Proc. Nat. Acad. Sci. USAVol. 69, No. 6, pp. 1569-1573, June 1972

Structures of the Visual Chromophores and Related Pigments:A Conformational Basis of Visual Excitation

(11-cis retinal/vitamin A/xanthenoid/crystal structure/carotenoid)

M. SUNDARALINGAM AND C. BEDDELL

Department of Biochemistry, College of Agricultural and Life Sciences, University of Wisconsin, Madison, Wis. 53706

Communicated by Henry Lardy, April 6, 1972

ABSTRACT A detailed conformational analysis of theknown crystal structures of vitamin A, carotenoid, andxanthenoid derivatives has been made. These studiessuggest that the visual chromophore li-cis retinal and itsisomers may each occur in several different conformationalstates, which would be expected to have different light-absorbing characteristics. The presence of such states forli-cis retinal and all-trans retinal may contribute to theobserved rhodopsin intermediates that have been charac-terized by different Xmax values, and to the breadth ofspectral response in scotopic and color vision.

In the process of scotopic vision, absorption of light by thevisual pigment rhodopsin gives rise to nerve transduction andthe physiological process of vision. Rhodopsin is believed to beintegrated into the disc membrane of retinal-rod outer seg-ments as a protein-lipid complex (1). It has been known formany years that the absorption of light by the rhodopsinvisual chromophore, retinal, results in the geometrical isom-erization of the 1 1-cis to the all-trans form (2). It has alsobeen shown that the chromophore is linked covalently to thelipoprotein (3, 4), and that it may possibly undergo a changein covalent bonding (5). The mechanism by which the nerveimpulse is initiated is still obscure; an early receptor potentialaccompanying retinal illumination has been observed (6); itmay be related to isomerization or to the transfer (5) of thechromophore, but it is not clear whether this potential is partof the mechanism of excitation. 8While both the chromophore and the 'lipoprotein are es-

sential components in the primary processes of vision, little isknown of their conformations in rhodopsiff. A knowledge ofthe detailed conformation of rhodopsin and of its constituentswill, however, be essential for a precise undersAtanding of thephotochemical events. In the present paper, we give a con-formational analysis of the visual chromophores and discussits relevance to the visual process. In addition, using informa-tion from crystallographic studies on the interactions of smallmolecules with proteins, we suggest models for the interactionof the chromophore with opsin in rhodopsin.

Structural analysis

Crystal structures of the visual chromophores 11-cis retinaland all-trans retinal (7), and of eight other carotenoid andxanthenoid derivatives are available (8-15). The chemicalstructures and atomic numbering of these compounds aregiven in Fig. 1. Detailed coordinates have been published forthe first six of these compounds, while for compounds 7 and 8,only preliminary reports have been made and the coordinatesare not yet available.

1569

The geometry around the double bonds in the carotenoidchains can be either cis or trans. Therefore, several geometricalisomers are possible (16). In addition, the conformation aboutthe ring-chain 6-7 linkage can be gauche ('s-cis') or anti('s-trans'), and each geometrical isomer can potentially exhibitdifferent conformations for the ring and 'single' bonds in theside chain.

Torsion Angles. The torsion angles in the known compoundswere calculated from published atomic coordinates; they arelisted in Tables 1 and 2.

Ring Conformation. The torsion angles about the ring bondsfor the compounds are listed in Table 1. All of the cyclohexenerings are puckered in the half-chair conformation. The atomsC(2) and C(3) are displaced (about 0.3 A) on opposite sides ofthe plane through the remaining four atoms. The largest torsionangle is about the C(2)-C(3) bond, which is opposite thedouble bond, while the smallest torsion angle is about thedouble bond. In the case of compounds 5 and 6, the ketogroup at position C(4) influences the conformation of thecyclohexene ring such that the torsion angles about 2-3 and3-4 are, respectively, reduced by about 100 and 200 relativeto those of the ,3-ionylidene ring; the former becomes aboutequal to the torsion angle about the 1-2 bond. Therefore, ingeneral, the cyclohexenone rings are less puckered than thecyclohexene rings. In vitamin A2 (retinal 2) there is an addi-tional double bond at position 34 in comparison with vitaminA1 (retinal 1). It is expected that the cyclohexadiene ring sys-tem will preferably adopt a conformation similar to that ofthe cyclohexenone ring, or more planar. The substituentC(16) lies close to the ring plane, whereas one of the gemdimethyl groups on C(1) displays a pseudo-axial orientation,while the other displays a pseudo-equatorial orientation.

Ring-chain conformation. The stereochemistry of the ringrelative to the chain is an important parameter in the retinalsand the carotenoid derivatives. The ring-chain conformationis described by the torsion angle X [5-6-7-8] [or 1-6-7-8], whichis the angle between the ring double-bond 5-6 and the chaindouble-bond 7-8 (Table 2). The usual usage of the terms transand cis denotes the configuration around double bonds.Where these terms are applied to geometries around singlebonds, they are written 's-trans' and 's-cis', respectively.However, the observed torsion angles cannot be adequatelydescribed in this way. Therefore, we shall use the conventionalterminology for single bonds, anti and gauche. Rotation aboutthe 'single' bond 6-7 is not free; instead, it is highly restricted

1570 Chemistry: Sundaralingam and Beddell

1 OH

10

FIG. 1. The chemical structures and numberingof the known crystal structures of vitamin A,caratenoid, and xanthenoid derivatives.1 trans-g3-ionylidene--y-crotonic-acid (8)2 Vitamin A acid (9)3 #-Carotene (10)4 15,15'-Dehydro-j3-carotene (11)5 Canthaxanthine (12)6 15,15'-dehydrocanthaxanthine (15)7 cis-i6-ionylidene--y-crotonic acid (13)8 retro-g-ionylidene acetyl-p-bromoanilide (14)9 11-cis retinal (7)10 All-trans retinal (7)

0

P.

FIG. 2. A diagrammatic illustration of possibleconformations for the cis-and trans-isomer of theretinal chromophore in rhodopsin. The chromophoreis shown bound to a lysine residue ("Model 9").A The Schiff base can exist as the 8yn or anti

isomer. It is conceivable that isomerization of theSchiff base may accompany isomerization at the11-12 bond.

B The presence of the 6-7 anti conformer of 11-Ci8retinal (shown) might be expected to enhancethe red shift.

C An induced conformational change in opsin,such as that shown figuratively for Ca of thelysine residue, might lead to initiation of thenervous impulse.

Proc. Nat. Acad. Sci. USA 69 (1972)

4

Conformational Basis of Visual Excitation 1571

to three possible ranges. In the anti conformation, the X angleis about 1800, while in the gauche conformations, the angle isin the range 39 to 520 (+gauche) or -39 to -52° (-gauche).Compound 1 is the only example that exhibits the anti con-formation, X = 1690. The cis planar conformation about the6-7 bond is forbidden because of steric interaction betweenthe substituents in the ring positions C(5) and C(1) and thehydrogen atoms in the side-chain at C(8) and C(7). There-fore, although the anti planar conformation is allowed, the+gauche conformation is preferred (17). Both 11-cis and all-trans retinals exhibit the gauche conformation (7).

Chain Conformation. Here one has to consider two pos-sibilities, an all-trans chain and one containing cis linkages.In the known trans compounds, the torsion angles about thedouble and 'single' bonds in the side chain lie, on an average,in the range 180 i 50, emphasizing the fact that small dis-tortions from planarity are often encountered, even in con-jugated double-bonded systems. In general, the chain dis-plays a bend normal to its plane (15). In compound 7 there isonly a small deviation from the cis planar arrangement,perhaps due to steric interaction between the hydrogen atomsattached to positions C(8) and C(11). Similarly, small distor-tions from anti planarity are observed in the retinals (7). Invitamin A acid (2), the torsion angle 13-14-15-0 involving thecarbonyl group is 160, while that involving the hydroxylgroup 13-14-15-OH is 1710. However, in 11-cis retinal and all-trans retinal, the torsion angle 13-14-15-0 for the carbonylgroup is about 1800. For the Schiff base complex betweenretinal and opsin, either of the corresponding conformations,or an intermediate one, might be present.

Discussion

Conformations of the Chromophore. In the absence ofmolecular crowding, one would expect that the conformations

TABLE 1. Torsion angles about the cyclohexene ring bonds

01-2 02-3 103-4 1045 W0- q61

1 trans-fl-ionylidene-7-crotonic acid ±44 T60 ±41 F11 T3 +12

2 Vitamin A acid ±39 T59 ±45 T14 +F5 +73 6-Carotene 421 T35 ±26 +4 +9 ± 14 15,15'-Dehydro-ft-

carotene 442 T60 444 + 12 +F6 T95 Canthaxanthine

Residue 1 450 T49 423 T0 44 + 28Residue 2 ±45 T40 415 T4 42 T27

6 15,15'-Dehydro-cantha-xanthine 449 +51 ±27 T2 41 +24

The torsion angle (.0w) about a bond is defined as the anglemade by the projection of the 1-2 bond with respect to the 3-4bond. When viewed down the 2--3 bond, positive angles (0 to+1800) are for a clockwise rotation of the far bond 3-4 relativeto the near bond 1-2, while negative angles (0 to -180°) arefor a counterclockwise rotation. Zero angle is defined for aneclipsed position of the 2-3 and 3-4 bonds. The upper andlower signs for the torsion angles indicate the conformations ofthe enantiomers. Since all of the compounds belong to a centro-symmetric space group, both enantiomers are present in thecrystal. Although the torsion angles are not available for com-pound 7 and the retinals, the retinals exhibit the preferred half-

TABLE 2. The ring-chain torsion angles

5-6-7-8 1-6-7-8 Conformation

1 trans-,0-ionylidene--y-crotonic acid +169 413 anti

2 Vitamin A acid T48 ±143 +gauche3 d-Carotene F39 4143 +gauche4 15,15'-Dehydro-,5-

carotene T49 ± 132 +Fgauche5 Canthaxanthine

Residue 1 452 T132 ±gaucheResidue 2 454 +127 ±gauche

6 15,15'-Dehydrocantha-xanthine 443 +141 ±gauche

7 cis-B-ionylidene-y-crotonic acid 80 gauche

8 retro-ft-ionylidene acetyl-p-bromoanilide 180 - anti

9 11-cis retinal ±400 - gauche10 All-trans retinal ±59 - gauche

Since all of these crystal structures possess a center of inversion,both enantiomers are present. The upper and lower signs in thetorsion angles designate the enantiomeric conformations. Thedetailed structuies of compounds 7, 8, 9, and 10 have not ap-peared; only approximate values of the torsion angles 5-6-7-8are available. It is noteworthy that compound 8 has a double bondlinking the ring to the chain and, therefore, it is not strictly an-alogous to the other compounds.

about single bonds in a conjugated system will be planar, tomaximize conjugation. It has long been recognized that in1 1-cis retinal there is molecular overcrowding, involving themethyl group at position 13 and the hydrogen atom at position10, but it was not clear whether the steric interactions arerelieved by rotation about the cis double bond, the adjacentsingle bonds, or both (16). In the recently determined x-raystructure of 11-cis retinal, steric interaction was relieved byrotation about the 12-13 bond; the conformation about the10-11 bond was close to antiplanar (7). Honig and Karplus(18) have also predicted a similar conformation from cal-culations of the torsional potentials of retinal for some con-formations about the 10-11, 11-12, and 12-13 bonds. It isinteresting to compare this situation with systems containingisolated double bonds. An analysis of the conformations aboutisolated cis and trans double bonds demonstrated that thesingle bonds adjacent to the double bond show considerable"flexibility" to rotation, and that equal rotations of oppositesign (1250 ± 200) about these bonds are preferred (19).However, in conjugated systems the relative rotations aboutsingle bonds appear to be those that favor conjugation.Rotation about the 10-11 bond will decrease the degree ofconjugation to a larger extent than rotation about the 12-13bond, and a smaller rotation about the 10-11 bond would beexpected for the ground state. Hence, conjugation, in part,differentiates the flexibilities of the single bonds. It should,however, be noted that the presence of a bulkier substituentat position 13 than at position 10 in retinal might also lead tounequal rotations about the single bonds.

In its complex with opsin, the visual chromophore may notexist in the identical conformation as in free solution or in acrystal. Due to the possibility of 'flexibility' about the singlebonds, a range of conformations are possible that differ only

chair conformation. slightly in energy. Indeed, Honig and Karplus (18) have

Proc. Nat. Acad. Sci. USA 69 (1972)

1572 Chemistry: Sundaralingam and Beddell

found the presence of two energy minima corresponding totwo conformations about the 12-13 bond, one about 600 andthe other about 1300. From energy calculations (18), it ap-pears that the energy difference between these two conforma-tions is small and that the entire range of conformations 600-1300 about the 12-13 bond can be exhibited by the retinals.The presence of different conformers in rhodopsin would beexpected to broaden the absorption spectrum, since individualconformers will possess different degrees of conjugation and,hence, different spectral characteristics. However, this mech-anism alone does not explain the red shift observed for rhodop-sin, relative to retinal (20).

A Conformational Model for Rhodopsin. In the absence ofdetailed structural information concerning the conformationof the chromophore and its environment in rhodopsin, onemight use the present knowledge of possible conformationsfor the chromophore, and of the types of interactions that havebeen observed between ligands and proteins in x-ray crys-tallographic studies, to construct a model for the active siteof rhodopsin. In view of the markedly different Xmax valuesfor rhodopsin and for the free chromophore in homopolarsolvents (20), one may suggest that while the chromophore islikely to interact in similar ways with ligands of comparable;hydrophobicity, the presence of specific polar groups in suit-able positions in opsin may be responsible for the red shift.As models for the interaction of the cyclohexene ring inrhodopsin, one might consider the interaction of .the hemegroup in myoglobin (21), hemoglobin (22), and ferricyto-chrome c (23), of the tyrosyl side chain of glycyl-ityrosine inits complex with carboxypeptidase A (24), and of the indolylside chain of formyl-itrypotophan in its complex with a-chymotrypsin (25). In all of these examples, the hydrophobicgroup binds in a pocket in the protein, with the formation ofmany van der Waals contacts; similarly, we imagine that thecyclohexene ring of retinal is enclosed in a pocket in the opsinwithin which it makes favorable contacts. To elaborate themodel further, we must consider whether the chromophoreundergoes a change of covalency. It has been suggested thatthe chromophore undergoes a change in covalent bonding in adark process after illumination (5). Retinal was linked tophosphatidyl ethanolamine in dark rhodopsin, and in theillumination product metarhodopsin I, but to have beometransferred to a lysine residue of the opsin in metarhodopsinII. However, it has since been suggested that the observedlinkage of the chromophore to phospholipid is induced by theanalysis and isolation procedures (26). We will, therefore,consider both the case where the chromophore is transferredfrom phospholipid (Model 1) and the case where it is cova-lently linked only to lysine (Model 2).

Model 1. If we assume that in rhodopsin, metarhodopsin I,and metarhodopsin II the cyclohexene ring remains bound atthe same site, then evidently conformational changes in thechromophore are likely during the transfer of the Schiff baselinkage from the primary amine of phosphatidyl ethanolamineto that of lysine. These conformational changes are additionalto those induced directly by the absorption of light quanta.In the model we describe here, we suggest that the isomeriza-tion induced by absorption of light will induce a strain in thechromophore, in view of its attachment at each end to proteinand phospholipid, and that the strain may be relieved by a

Since these components constitute an integral part of thedisc membrane of the rod outer segment, such changes mayinitiate ion passage through the membrane and, thus, initiatethe process of nerve transduction. We further suggest that themovement of the Schiff base brings it close to a lysyl sidechain, which displaces the phospholipid from the chromo-phore. The phospholipid is then free to move to its originalposition in the membrane, hence stopping the local inducedion flow and terminating the pulse of current. It is this currentthat may correspond to the early receptor potential. A featureof this theory is that a phospholipid molecule is stronglybound through the chromophore to the protein in unbleached,but not in bleached, rhodopsin. Significantly, more phos-pholipid can be extracted from rod segments after photo-bleaching (27). The visual cycle is completed by the reverseisomerization of the chromophore, and its transfer to thephospholipid.

Model 2. If we assume that the chromophore does notchange covalent linkage but that, as in Model 1, the cyclo-hexene ring remains bound at the same site in the lipoproteinduring the 11-cis to all-trans isomerization, we must considerwhether there is sufficient flexibility in the side chain of thelysyl residue and in the chromophore for the isomerization tooccur without inducing further large conformational changesin the opsin. In Fig. 2 we show schematically possible con-formations for the 11-cis and all-trans isomers of the chromo-phore linked to the lysyl residue. It is evident that there couldbe sufficient flexibility of the lysyl side chain and the chromo-phore side chain beyond the 11-12 double bond to permit theisomerization. The reverse isomerization may occur spon-taneously to relieve the strain in the rhodopsin, or may occurafter hydrolysi:f of the chromophore from the opsin.

In both of these models, the 11-cis -- all-trans isomeriza-tion of the chromophore is considered to involve conforma-tional changes primarily distal to the 10-11 bond in thechromophore, as shown in the crystal structure of the 11-cisisomer (7). These changes are considered responsible for themetarhodopsin I-metarhodopsin II transition. Metarhodop-sin I (Xm.. = 478 nm) shows a red shift relative to meta-rhodopsin II (X.ma = 380 nm) and all-trans retinal (Xmax =387 nm) (28); this shift may arise from the nature of the localenvironment at the distal end of the 11-cis chromophore. Ithas been suggested that the chromophore may exist in rhodop-sin as a protonated Schiff base (29); recently, it was shownthat the retinylideneiminium chromophore can show a redshift similar to that of rhodopsin (30). A large bathochromicshift was observed for the retinylideneiminium chromophoreat low temperature in the presence of trichloroacetic acid, butnot in the presence of hydrochloric acid, acetic acid, or mono-chloroacetic acid; it was suggested that a specific associationof trichloracetic acid with the retinylideneiminium chromo-phore might be responsible (30). Such an association is likelyto be dependent on the conformation of the retinylidene-iminium chromophore, and the interaction with trichloroaceticacid may be associated with the occurrence of a specificconformation of the chromophore at low temperature. In itscomplex with opsin, the red shift observed for rhodopsinand metarhodopsin I may be due to specific interactions be-tween the 11-cis-chromophore and the opsin that are lostwhen the chromophore isomerises to all-trans. It has beensuggested that the red shift may arise from delocalization of

Proc. Nat. Acad. Sci. USA 69- (1972)

conformational change of the phospholipid and/or protein.

Conformational Basis of Visual Excitation 1573

the positive charge of the protonated Schiff base to the side-chain carbon atoms of the chromophore (28); alternatively,it has been suggested that the Schiff base is uncharged, butthat carbonium ions are induced in the side chain by appro-

priately placed Lewis acids or protonating groups of rhodopsin(31). It is interesting to compare this situation with thatproposed for lysozyme from crystallographic studies (32).In lysozyme, it appears that the glycosyl oxygen atom of thesubstrate is protonated by glutamic acid residue 35, and theformation of a carboxonium ion at C-05 is promoted bydistortion of the sugar residue towards a half-chair conforma-tion and by the proximity of an anionic carboxylate group,

aspartate 52. By analogy with Fig. 2, the cis chromophore inrhodopsin or metarhodopsin I is shown protonated by an

acidic group, while the delocalized positive charge is stabilizedon carbon atoms at the end of the chromophore side-chainby an adjacent negative group (or dipole). Isomerizationto the all-trans isomer positions the distal end of chromophorein a different local environment, which does not promotecarbonium ion formation, and the red shift is lost. The spectraldifference between rhodopsin and metarhodopsin I may reflectsmall differences in conformation of the 11-cis chromophorethat position it in slightly different environments in the opsin.We therefore suggest that some of the illumination products

of rhodopsin with different Xmax values are characterizedprimarily by different chromophore conformations, ratherthan by different opsin conformations, as has been postulated(33). Changes in chromophore conformation could also ac-

count for the unmasking of certain functional groups in therhodopsin that have been observed to accompany the transi-tion between the illumination products (33).

It should be possible to perform structure determinationsand a conformational analysis similar to that described hereboth for the retinylideneiminium compounds and for the so-

called retinal2 (vitamin A2) chromophore (15). The additionalring double-bond in retinal2 would be expected to contributeto a greater red shift than observed for the retinals themselves,as would also the occurrence of the 6-7 's-trans' (anti) con-

former. Different visual pigments, arising from combinationsof the retinal chromophores with different opsins, may pos-

sibly be involved in color vision.

We thank the National Institutes of Health of the UnitedStates Public Health Service for support of this work throughGrants GM-17378 and GM-18455.

1. Vanderkooi, G. & Sundaralingam, M. (1970) Proc. Nat.Acad. Sci. USA 67, 233-238.

2. Hubbard, R. & Kropf, A. (1958) Proc. Nat. Acad. Sci. USA44, 130-139; Kropf, A. & Hubbard, R. (1958) Ann. N.Y.Acad. Sci. 74, 266-280.

3. Bownds, D. (1967) Nature 216, 1178-1181.

4. Akhtar, M., Blosse, P. T. & Dewhurst, P. B. (1967) Chem.Commun. 631-632.

5. Poincelot, R. P., Millar, P. G., Kimbel, R. L. & Abraham-son, E. W. (1969) Nature 221, 256-257.

6. Brown, K. T. & Murakami, M. (1964) Nature 201, 626-628.7. Gilardi, R., Karle, I. L., Karle, J. & Sperling, W. (1971)

Nature 232, 187-188.8. Eichhorn, E. L. & MacGillavry, C. H. (1959) Acta

Crystallogr. 12, 872-883.9. Stam, C. H. & MacGillavry, C. H. (1963) Acta Crystallogr.

16, 62-68.10. Sterling, C. (1964) Acta Crystallogr. 17, 1224-1228.11. Sly, W. G. (1964) Acta Crystallogr. 17, 511-528.12. Bart, J. C. J. & MacGillavry, C. H. (1968) Acta Crystallogr.

Sect. B 24, 1587-1606.13. Koch, B. & MacGillavry, C. H. (1963) International Union

of Crystallography Congress, Rome, Aug. 1963, Abstracts,p. A48.

14. Paul-Roy, S., Scherk, H. & MacGillavry, C. H. (1969)Chem. Commun. 1517-1518.

15. Bart, J. C. J. & MacGillavry, C. H. (1968) Acta Crystallogr.Sect. B 24, 1569-1587.

16. Hubbard, R. & Wald, G. (1968) in "Structural Chemistryand Molecular Biology," eds. Rich, A. & Davidson, N.(Freeman and Co., San Francisco), pp. 545-554; Abraham-son, E. W. & Weisenfeld, J. R., Hand&. Sensory Physiol.VII, part. 1, in press.

17. Honig, B., Hudson, B., Sykes, B. D. & Karplus, M. (1971)Proc. Nat. Acad. Sci. USA 68, 1289-1293.

18. Honig, B. & Karplus, M. (1971) Nature 229, 558-560.19. Sundaralingam, M. (1971) in "Membrane structure and its

biological applications," New York Academy of SciencesSymposium, New York, June 2-4, 1971.

20. Abrahamson, E. W. & Ostroy, S. E. (1967) Progr. Biophys.Mol. Biol. 17, 179-215, and references therein.

21. Kendrew, J. C., Dickerson, R. E., Strandberg, B. E., Hart,R. G., Davies, D. R., Phillips, D. C. & Shore, V. C. (1960)Nature 185, 422-427.

22. Cullis, A. F., Muirhead, H., Perutz, M. F., Rossmann, M.G. & North, A. C. T. (1961) Proc. Roy. Soc. Ser. A 265, 15-38; (1961) Proc. Roy. Soc. Ser. A 265, 161-187.

23. Dickerson, R. E., Kopka, M. L., Weinzierl, J. E., Varnum,J. C., Eisenberg, D. & Margoliash, E. (1967) J. Biol. Chem.242, 3015-3018.

24. Lipscomb, W. N., Reeke, G. N., Hartsuck, J. A., Quiocho,F. A. & Bethge, P. H. (1970) Phil. Trans. Roy. Soc. London,Ser. B 257, 177-214.

25. Steitz, T. A., Henderson, R. & Blow, D. M. (1969) J. Mol.Biol. 46, 337-348.

26. Hall, M. & Bachrach, A. D. E. (1970) Nature 225, 637-638;Anderson, R. E. (1970) Nature 227, 954-955.

27. Krinsky, N. I. (1958) Arch. Ophthalmol. 60, 688-694.28. Matthews, R. G., Hubbard, R., Brown, P. K. & Wald, G.

(1963) J. Gen. Physiol. 47, 215-240.29. Kropf, A. & Hubbard, R. (1958) Ann. N.Y. Acad. Sci. 74,

266-280.30. Hirtenstein, M. D. & Akhtar, M. (1971) Nature 233, 94-95.31. Blatz, P. E. (1965) J. Gen. Physiol. 48, 753-760.32. Blake, C. C. F., Johnson, L. N., Mair, G. A., North, A. C.

T., Phillips, D. C. & Sarma, V. R. (1967) Proc. Roy. Soc.Ser. B 167, 378-388.

33. Wald, G. (1968) Nature 219, 800-807.

Proc. Nat. Acad. Sci. USA 69 (1972)