Proc. Natl. Acad. Sci. USA Vol. 91, pp. 380-384, January 1994 Neurobiology Topography of a binding site for small amnestic deduced from structure-activity studies: Relation to amnestic effect of amyloid f8 protein JAMES F. FLOOD*, EUGENE ROBERTStt, MARK A. SHERMAN§, BRUCE E. KAPLAN§, AND JOHN E. MORLEY* Geriatric Research Education and Clinical Center (GRECC), Veterans Administration Medical Center, St. Louis, MO 63106, and St. Louis University School of Medicine, Division of Geriatric Medicine, St. Louis, MO 63104; and tDepartment of Neurobiochemistry and §Biology Division, Beckman Research Institute of the City of Hope, Duarte, CA 91010 Contributed by Eugene Roberts, September 7, 1993

ABSTRACT Four peptides homologous to amyloid (B pro- With the exception of commercially purchased peptides tein containng the Val-Phe-Phe (VFF) sequence administered VF and FV, all peptides used in these studies were synthe- intracerebroventricularly after training caused amnesia for sized and analyzed to establish purity by standard methods. footshock active avoidance training in mice. Results with VFF Prior to neutralization and appropriate dilution with saline, and other peptides containing VFF or portions thereof were peptides were dissolved in (i) saline, (ii) dimethyl sulfoxide, used to generate a topographic map for a hypothetical binding (iii) glacial acetic acid, or (iv) NaOH. Upon testing for surface for amnestic peptides, termed Z. Effects on retention of retention of FAAT, groups receiving posttraining icv admin- footshock active avoidance training were rationalized in terms istration of2 1d ofappropriate dilutions ofthe above vehicles of flt to Z, making possible design of potential memory- showed no significant differences among them (ANOVA, F modulating peptidic and nonpeptidic substances. Three pep- < 1). The mean trials to criterion of the most frequently tides that neither improved nor impahred retention blocked the emploved vehicle (vehicle ii above; final concentration of amnestic effects of -(12-28), a homologous to amyloid 13 protein, opening the way to development of substances that dimethyl sulfoxide, 0.001%) were compared to that of mice can antagonize the neurotoxic effects of amyloid (3 protein on receiving test peptides. neural structures and thus attenuate symptoms and progres- With the exception of dose-response curves, the experi- sion of Alzheimer disease. ments reported below tested whether or not peptides im- paired retention at 6.14 nmol per mouse. Although f-(1-28) Intracerebroventricular (icv) administration of a synthetic (1) and (-(12-28) were significantly amnestic at 1.5 nmol per peptide homologous to amyloid 13protein (AP), [Glnl1]13-(1- mouse, the 6.14-nmol dose was chosen so that substances 28) (DAEFRHDSGYQVHHQKLVFFAEDVGSNK), im- with weaker effects might be detected. mediately after training caused amnesia for footshock active avoidance training (FAAT) in mice in a dose-dependent RESULTS fashion. Also amnestic were peptides l3-(12-28), -(18-28), and /3-(12-20) (1). The above amnestic peptides have the Topography of a Binding Site Deduced from a Structure- sequence VFF in common. ,B-(1-11), which does not contain Activity Study of Amnestic Effects. Examination of Corey- VFF, is not amnestic (see below). FAAT experiments were Pauling-Koltun (CPK) models of the peptides tested (Table performed with 28 peptides, 22 of which contained the VFF 1) and superimposition of appropriate configurations of the sequence or portions thereof. A topographic model for the amnestically active ones led to devisal of a provisional binding site ofamnestic peptides, termed Z, was derived from topographic model ofthe site (Z) to which all ofthe amnestic the structure-activity study of the latter. substances might bind. A more refined hypothetical three- In separate experiments, 3 of 15 nonamnestic peptides dimensional binding surface for the peptides (described be- were found to counteract the amnestic effects of ,B-(12-28). low) then was generated by surrounding with small molecules that mimic the properties of amino acids the peptide Gln- MATERIALS AND METHODS Phe-Phe-y-aminobutyric acid that fit optimally to all of the proposed contact points on the original map. A solvent- The care and handling of test mice, surgical procedures for accessible surface was generated using the algorithm of preparation of mice for icv administration, the testing appa- Connolly (2). Peptides then were docked onto the surface and ratus, training and testing procedures, and methods of sta- energy-minimized to optimize contacts [Dreiding II forcefield tistical treatment of data have been described (1). Immedi- (3)]. Images were created with the BIOGRAF program (Mo- ately after training, groups of mice (10-15 mice per group) lecular Simulations, Waltham, MA). The hypothetical recep- were given icv injections (2 ,ul) of vehicle alone or vehicle tor map thus generated (Z) is presented at the outset. Without containing 6.14 nmol of a peptide. Tests for retention of FAAT were performed 1 week later. Results were expressed a detailed molecular definition of the structure of Z, quanti- as the mean and SEM. Significance of overall effects of tative assessments of binding energies of ligands are not treatment was determined by ANOVA run on trials to reach possible. Qualitative and semiquantitative arguments devel- a footshock avoidance criterion in the retention test (trials to oped with the data at hand were employed to interpret the criterion). Tukey's t test was used to make statistical com- results. Numbering clockwise from the lower left in Fig. 11, parisons among group means. Treatment was judged to be the postulated sites on Z are characterized as: 1, H-bonding; amnestic when the mean of vehicle and a treatment group gave a P value of <0.01. Abbreviations: AA3, amyloid ,B protein; FAAT, footshock active avoidance training; icv, intracerebroventricular; VIP, vasoactive intestinal peptide. The publication costs of this article were defrayed in part by page charge *To whom reprint requests should be addressed at: Department of payment. This article must therefore be hereby marked "advertisement" Neurobiochemistry, Beckman Research Institute of the City of in accordance with 18 U.S.C. §1734 solely to indicate this fact. Hope, Duarte, CA 91010. 380 Downloaded by guest on September 23, 2021 Neurobiology: Flood et al. Proc. Natl. Acad. Sci. USA 91 (1994) 381 Table 1. Effect of 6.14 nmol of icv-administered peptides on amino group, for a total of five groups per molecule. Water retention of T-maze FAAT (15 mice per group) clusters may be organized at those sites, the sizes and shapes Peptide Trials to of which are determined by combinatorial influences of Peptide tested criterion, no. molecular and environmental factors. Unbound Z site is presumed to have water clusters associated with sites 1, 3, 1* VFF 9.67 0.40t and 5 (Fig. 11). VFF (Fig. 14) is posited to orient on Z in such 2* VFFAE 8.73 0.38t a manner that the attractive energy between VFF and Z 3* KLVFF 7.73 0.42 becomes great enough to squeeze out intervening water, in 4* KLVFFAE 6.87 0.27 this way minimizing the energy of the system. The molar 5* FFV 9.07 0.41t amnestic efficacy ofVFF was not significantly different from 6t FFVG 8.47 0.33t+ that ofthe APhomologue [Gln11]f-(1-28) (Fig. 2A). Ifbinding 7t DFFV 7.27 0.38 of its VFF segment to Z is key to the amnestic effect of the 8t DFFVG 6.87 0.33 latter, its flanking regions must contribute to strength of 9 KLFFVAE 9.58 0.39t attachment since neither the a-amino group of Val nor the lo§ VF 9.13 0.94t carboxyl group of F-2 is available to contribute to binding ii§ VFT 9.73 0.31t efficacy. Similar considerations apply to other such compar- 12§ AVFT 9.73 ± 0.21t isons, as for AVFT and VIP in Fig. 2B. 13§ AVF 6.80 ± 0.25 If residues with H-bonding side chains that do not bind to 14 VVF 9.27 ± 0.37t Z are added at either end of a peptide that, by itself, binds to 15 VFV 9.07 ± 0.43t Z, water clusters on the nonbinding moieties may become 16 FVF 6.93 ± 0.29 sufficiently large and stable so that resulting physical and 171 FV 9.40 ± 0.46t energetic barriers deter effective attachment ofthe peptide to 1811 SFVG 7.47 ± 0.45 Z. A curvilinearly decremental trend in amnestic potency was 19** QFVG 6.80 ± 0.39 noted with increasing numbers of side-chain H-bonding 20ft FF 7.00 ± 0.42 groups in such peptides related to VFF (Fig. 3A). 21#* QFFG 9.00 ± 0.46t FFV, also amnestic (Table 1 and Fig. 3B), attaches to 22 SFFG 7.33 ± 0.40 aromatic sites 2 and 4 on Z, but the C-terminal carboxyl group 23§H AFIG 6.67 ± 0.26 of Val in FFV is directed to cationic site 3 (Fig. 17). Addition 24§§ AIFT 7.00 ± 0.29 of H-bonding groups to FFV also reduced amnestic potency 25§§ GFMT 6.80 ± 0.34 (Fig. 3B). It is presumed that a water cluster can be mobilized 26§§ NLIT 7.07 ± 0.36 on the amide nitrogen of a peptide bond when Gly is the 27 vv 6.80 ± 0.31 C-terminal residue because, uniquely among the amino acids, 28* DAEFRHDSGYQ 6.93 ± 0.28 the C-terminal Gly offers little side-chain interference to Data for trials to criterion are expressed as the mean t SEM. P water cluster formation. For the latter reason, the number of values were obtained for selected comparisons after obtaining a H-bonding groups assigned to a peptide was increased by one significant F value by ANOVA; F(16,204) = 8.22, P < 0.001. for peptides with a C-terminal Gly (Fig. 3B). *Residues in A,B numbered from the N terminus: peptide 1, 18-20; KLVFFAE (Fig. 1 8 and 9), residues 16-22 ofA13, was not peptide 2, 18-22; peptide 3, 16-20; peptide 4, 16-22; peptide 28, amnestic (Table 1 and Fig. 3A), while the inversion of VFF 1-11. 1 tp < 0.01 for comparison with vehicle alone. to give KLFFVAE resulted in an amnestic peptide (Table *Residues in : peptide 5, 5-7; peptide 6, 5-8; peptide 7, and Figs. 2C and 3). KLFFVAE is presumed to make a 4-7; peptide 8, 4-8. five-point attachment to Z in the following manner (Fig. 110): §Residues in VIP: peptide 10, 5 and 6; peptide 11, 5-7; peptide 12, site 1, a-amino group of K; site 2, F-1; site 3, a-carboxyl 4-7; peptide 13, 3-5. group of E; site 4, F-2; site 5, 8-amino group of K. ¶Peptide 17 residues are also found in the following: residues 8 and Ifthe F-1 and F-2 residues of nonamnestic KLVFFAE are 9, kassinin; residues 6 and 7, ; residues 6 and 7, arranged so as to fit sites 2 and 4, respectively, the a-carboxyl neurokinin B; residues 17 and 18, y; residues 32 and group of Glu could interact coulombically with cationic site 33, . 3 (Fig. 1 8 and 9); but the unrealistic energetic demand of IlPeptide 18 residues are also found in the following: residues 5-8 in LVF neurokinin A; residues 16-19, ; residues 31-34, naked aqueous exposure of the hydrophobic segment neuropeptide K. that arises in this configuration (Fig. 1 8 and 9) would **Peptide 19 residues are also residues 7-10 in kassinin. preclude the existence of such a conformation of the peptide ttPeptide 20 residues are also residues 7 and 8 in and in this instance and, therefore, its binding to Z. residues 5 and 6 in neurokinin B. VFFAE, which is amnestic (Table 1 and Fig. 3A), is #*Peptide 21 residues are also residues 6-9 in substance P. presumed to attach as follows (figure not shown): site 5, §§Peptide 23 residues are also residues 6-9 in ; peptide 24 a-amino group of Val; site 4, F-1; site 2, F-2; and site 3, residues are also residues 4-7 in growth hormone releasing factor; -carboxyl group ofGlu. The lesser potency of VFFAE than peptide 25 residues are also residues 3-6 in ,B-endorphin; peptide that of VFF may be attributable to the greater energy 26 residues are also residues 29-32 in neuropeptide Y. required for desolvation of VFFAE (Fig. 3A). Peptide ,B-(1- and Three 11), without a VFF sequence, is not amnestic (Table 1) and 2, aromatic; 3, cationic; 4, aromatic; 5, anionic. presumably does not bind to Z. views of the proposed surface of Z are shown (Fig. 1 1-3). Both Phe residues in VFF are not required to produce Attachment of amnestic VFF (Table 1) to Z (Fig. 14-6) is amnesia, since VF (Fig. 111) and FV (Fig. 112) also were presumed to take place by interaction of the a-amino group amnestic (Table 1). VF and FV both fit well to portions ofZ, of the N-terminal Val with anionic site 5 and by association their Phe groups binding to aromatic site 4, the free amino ofthe (Phe residues,F-1 and F-2, residues with aromatic sites group of Val in VF binding to anionic site 5, and the free 4 and 2, respectively. The Phe residues in a particular peptide carboxyl group of Val in FV binding to cationic site 3. are designated F-1 and F-2 in the order of their occurrence However, neither FF (Fig. 113) nor VV (data not shown) was from the N terminus. amnestic (Table 1). There are five potential sites on the side chains ofVFF that The best fit of FF to Z (Fig. 113) would require there to be may H-bond with water, one on each ofthe two oxygen atoms closely grouped hydrophilic N-terminal amino, C-terminal of the C-terminal carboxyl group and three on the N-terminal carboxyl, and peptidic carbonyl groups (Fig. 114). The en- Downloaded by guest on September 23, 2021 382 Neurobiology: Flood et al. Proc. Natl. Acad. Sci. USA 91 (1994)

l~~~~~~~~~~~~~~~~~~~~~~ 6 1|_ ~~~~~~~~~~~--~~~~~~~: =~~~~~~~

FIG. 1. Fit of conformers of peptides chosen for docking with a computer-generated model of a putative topographic site, Z, for the binding of amnestic peptides. Atoms are color coded as follows: carbon, grey; oxygen, red; nitrogen, blue; hydrogen, white. Downloaded by guest on September 23, 2021 Neurobiology: Flood et al. Proc. Natl. Acad. Sci. USA 91 (1994) 383 was amnestic with a molar efficacy similar to that ofVIP (Fig. 2B). VFT (residue 5-7) was also amnestic (Figs. 115 and 2B); but AVF (residue 4-6; Fig. 117) was not (Table 1 and Fig. 2B). VFT fits well to a portion of Z, with Thr interacting coulombically with site 3 through its free carboxyl group and Val and Phe associating with the hydrophobic aromatic sites 2 and 4, respectively. Although the Ala residue ofAVFT (Fig. 116) protrudes from the surface of Z into the aqueous medium, the disposition of the hydrophobic portions of Ala and Val is such as to tend to prevent formation oflarge water a( Vehicle alone clusters around the closely lying hydrophilic groups. AVF .0 would bind to Z less effectively than does AVFT because (D coulombic interaction with site 3 is not possible.

0 Tentative Support for the Model. Every amnestic peptide

co C D tested originally contained a Val residue, and the smaller co amnestic peptides were fragments oflarger amnestic peptides KLFFVAEAT (A, and VIP). To enhance credibility ofthe model, we sought 9 a Val-free amnestic peptide that is a fragment of a larger memory-enhancing peptide and fits Z. We selected the tet- rapeptide QFFG (Fig. 118), residues 6-9 of the 11-residue 8. substance P (RPKPQQfFGLM-NH2). Substance P has been reported to enhance memory retention (e.g., see refs. 5-7) and to counteract trophic and toxic effects of AP3 on hippo- 7 KLVFFAEI campal neurons in culture (see ref. 8). QFFG fits well to Z (Fig. 118) and is amnestic (Table 1). The -t-amide group of QFFG H-bonds to site 1 and .its a-amino group interacts electrostatically with site 5. Residues F-1 and F-2 associate 2 4 6 2 4 6 with aromatic sites 4 and 2, respectively, and the carboxyl Dose group of the C-terminal Gly falls close enough to site 3 to FIG. 2. Dose-response curves comparing amnestic effects of engage in coulombic interaction with it. The structural re- VFF with those of the AP homologue [Gln11],-(1-28) (A) and quirements for amnestic effects of QFFG appear to be quite amnestic effects ofAVFT with those of VIP (B). The data for curves fastidious since the similar peptides QFVG, SFFG, and for [Gln11],-(1-28) and VIP were taken from refs. 1 and 4, respec- SFVG were not amnestic (Table 1). With -aminobutyric acid tively. The dose-response curve for KLFFVAE is shown in C. substituting for Gly, the C-terminal carboxyl group fits per- Curves for VFF, AVFT, and KLFFVAE are compared in D. Dose fectly onto site 3. QFF-raminobutyric acid was used to is reported as nmol per mouse. Trials to criterion are reported as no. devise the final topographic map (Fig. 1 1-3). QFFG Has Substance P-Like Activity. QFFG was tested for ergy required to remove water clusters from the latter prob- substance P-like activity on cultured contractile smooth ably would be too great to permit effective binding of FF to muscle cells from the avian amnion (9, 10). Thirty of 105 Z. If VV were oriented on Z in a manner similar to VF, its (28%) cells responded with contractile responses to locally affinity forZwould be less than that ofVF because the energy applied 1 ,uM QFFG, and 80 of 105 (76%) cells responded to of binding of the C-terminal Val to aromatic site 4 would be considerably less than that of Phe. Table 2. Effects on retention of T-maze FAAT of icv Vasoactive intestinal peptide (VIP) is amnestic in the coadministered nonamnestic peptides on amnestic effects FAAT paradigm (4). AVFT (Fig. 116), residues 4-7 of VIP, of P-(12-28) Peptide Trials to criterion, no. VFF to Vehicle alone 6.85 ± 0.20 ** (A) P-(12-28) alone 9.62 ± 0.30*t KLFFVAE, + AVF 9.31 ± 0.36 + DFFV 9.31 ± 0.38

9 ** (B) \ + KLVFF 9.23 ± 0.34 + NLIT 9.15 ± 0.41 0 FF.V + KLVFFAE 9.08 ± 0.30

KLVFF + FVF 8.92 ± 0.26 + AFIG 8.92 ± 0.38 o 8 FF + VV 8.85 ± 0.42 + SFVG 8.85 ± 0.41

1t \DFFV + GFMT 8.85 ± 0.47 FVy \~~KLVFA Vehicle alone DFFVGKLVFFAE + FF 8.69 ± 0.46 7 + SFFG 8.08 ± 0.40 + AIFT 6.92 ± 0.32t ± ** P < O1 by conparison with vehicle alone + DFFVG 6.80 0.38t + QFVG 6.69 ± 0.22* 5 6 7 8 9 10 Data are the mean ± SEM. The higher the mean, the less the No. of side-chain H-bonding groups efficacy of a peptide in blocking the amnestic effect of P-(12-28). *P < 0.01 for comparison with vehicle alone. FIG. 3. Decrease ofamnestic potency ofpeptides as a function of tp values were obtained for selected comparisons after obtaining a H-bonding capacity. The special role of the C-terminal Gly is significant F value by ANOVA; F(6,63) = 11.89; P < 0.001. discussed in the text. Trials to criterion are reported as no. *P < 0.01 for comparison with ,-(12-28) alone. Downloaded by guest on September 23, 2021 384 Neurobiology: Flood et al. Proc. Natl. Acad. Sci. USA 91 (1994) 11 0.25; DFFVG, 9.14 ± 0.32; AIFT, 9.43 ± 0.30; and QFVG, 9.64 ± 0.28. 10 * o 1 * - P < O1 by DISCUSSION .~ comparison If A13 binds to the site that Z was designed to represent, then o) with bar 1 our observations suggest the possibility that for attachment 0 AJ3 and some ofits larger amnestic fragments [e.g., P-(12-28)] may require only a small portion of their sequence that includes the VFF sequence or a portion thereof. A,B and the larger amnestic fragments derived from it possibly may combine with Z in the form of aggregates rather than as l monomers. Such peptides are known to self-associate, form- 658 6 1 2 3 4 5 6 7 ing 3-pleated sheet structures of still undetermined nature. It is of particular interest to the present study that residues FIG. 4. Antagonism by DFFVG of amnestic effect of 3-(12-28) 17-20 (LVFF) appear to play crucial roles in the formation of when DFFVG was administered before or after 1&(12-28). The 3-sheet structure and amyloidogenicity of AP (12). AP3 also groups receiving DFFVG plus (-(12-28) gave significantly fewer binds with high avidity to apolipoprotein E, with which it trials to criterion (reported as no.) than those receiving -(12-28) plus coexists in senile plaques (13). This complex might bind to Z. saline (P < 0.001) and were not significantly different from the Investigation of the potential therapeutic effects of QFVG, controls (saline). Bars: 1, saline/saline; 2, l3-(12-28)/saline; 3, saline/ DFFVG, and AIFT in Alzheimer disease is warranted be- /-(12-28); 4, saline/DFFVG; 5, DFFVG/saline; 6, /-(12-28)/ cause upon icv administration they clearly blocked the am- DFFVG; 7, DFFVG/P-(12-28). nestic effects of /-(12-28), a peptide homologous to AP3, 0.1 ,uM substance P. Since cross-desensitization was found without toxicity or elicitation of grossly observable physio- logical effects. It does not seem likely that the above antiam- between QFFG and substance P but not between substance nestic peptides act directly on the same site (Z) as that P and carbachol or , other agents that cause proposed for the amnestic peptides because they showed contraction, it is likely that QFFG activates neurokinin/ neither amnestic nor memory-enhancing effects when given substance P receptors on these cells, albeit less effectively alone. Allosteric effects locking Z into a nonreceptive mode than substance P, itself. Whether or not this is the case with for amnestic peptides are a possibility. neurons ofthe mammalian central nervous system remains to be determined. After completion of this work, a report We thank Dr. Lisa M. Dahm for kindly performing the physio- appeared that AP3 activates the above receptors, expressed in logical experiments with QFFG cited in the text. The work of J.F.F. Xenopus oocytes, in synergy with glutamate (11). and J.E.M. was supported by the Medical Research Service of the Department of Veterans Affairs and that of E.R., M.A.S., and Three Peptides That Are Neither Amnestic Nor Memory B.E.K. was supported by institutional funds from the Beckman Enhancing Block the Amnestic Effects of (-(12-28), a Peptide Research Institute of the City of Hope. Homologous to A.B. Fifteen of the peptides tested were found to have no significant amnestic effect (Table 1). Three of 1. Flood, J. F., Morley, J. E. & Roberts, E. (1991) Proc. Natl. them, QFVG, DFFVG, and AIFT, blocked the amnestic Acad. Sci. USA 88, 3363-3366. on 2. Connolly, M. L. (1983) Science 221, 709-713. effect of ,B-(12-28) retention of FAAT when coadminis- 3. Mayo, S. L., Olafson, B. D. & Goddard, W. A., III (1990) J. tered with isomolar amounts (6.14 nmol) of /-(12-18) (Table Phys. Chem. 94, 8897-8909. 2). 4. Flood, J. F., Garland, J. S. & Morley, J. E. (1990) Peptides 11, Subsequently, 2 ,l of DFFVG and 2 ,ul of 3-(12-28) were 933-938. given icv separately after training either before or after saline 5. Wetzel, W. & Matthies, H. (1982) Acta Biol. Med. Ger. 41, s or first DFFVG and then were 647-652. (60 apart), P-(12-28) given, 6. Schlesinger, K., Pelleymounter, M. A., van de Kamp, J., or first l3-(12-18) and then DFFVG were given (Fig. 4). Bader, D. L., Stewart, J. M. & Chase, T. N. (1986) Behav. Whether saline was given before or after 3-(12-28) did not Neural Biol. 45, 230-239. affect the result, indicating that increase in volume adminis- 7. Hasenohrl, R. U., Gerhardt, P. & Huston, J. P. (1990) Peptides tered icv from 2 ,.1 to 4 Al did not matter. The order of 11, 163-167. administration of ,B-(12-28) and DFFVG did not affect the 8. Kosik, K. & Coleman, P., eds. (1992) Neurobiol. Aging 13, 535-625. ability ofthe latter to block the amnestic effect ofthe former. 9. Bowers, C. W. & Dahm, L. M. (1993) Am. J. Physiol. 264, These results suggest, but do not prove, that direct interac- C229-C236. tion ofthe counter-amnestic peptides with ,l3(12-28) is not the 10. Bowers, C. W. & Dahm, L. M. (1992) Proc. Natl. Acad. Sci. reason for their protective action. .Separate experiments with USA 89, 8130-8134. the amnesia blockers in groups of 14 weakly trained animals 11. Kimura, H. & Schubert, D. (1993) Proc. Natl. Acad. Sci. USA showed them not to have any memory-enhancing effects on 90, 7508-7512. 12. Hilbich, C., Kisters-Woike, B., Reed, J., Masters, C. L. & retention of T-maze FAAT, indicating that amnestic effects Beyreuther, K. (1992) J. Mol. Biol. 228, 460-473. of ,B-(12-28) were not being overcome by independent mem- 13. Strittmatter, W. J., Saunders, A. M., Schmechel, D., Pericak- ory-enhancing effects. Values for the number of trials to Vance, M., Enghild, J., Salvesen, G. S. & Roses, A. D. (1993) criterion (mean ± SEM) were as follows: vehicle, 9.07 + Proc. Natl. Acad. Sci. USA 90, 1977-1981. Downloaded by guest on September 23, 2021