matic assays using a wide range of in vitro conditions. Furthermore, once the proteins are prepared, proteome screening is signifi- cantly faster and cheaper. Using similar pro- cedures, it is clearly possible to prepare pro- tein arrays of 10 to 100,000 proteins for global proteome analysis in humans and other eukaryotes. References and Notes 1. S. Fields, Y. Kohara, D. J. Lockhart, Proc. Natl. Acad. Sci. U.S.A. 96, 8825 (1999) A. Goffeau et al., Science 274, 546 (1996). 2. P. Ross-Macdonald et al., Nature 402, 413 (1999) J. L. DeRisi, V. R. Iyer, P. O. Brown, Science 278, 680 (1997) E. A. Winzeler et al., Science 285, 901 (1999) P. Uetz et al., Nature 403, 623 (2000) T. Ito et al., Proc. Natl. Acad. Sci. U.S.A. 97, 1143 (2000). 3. H. Zhu, M. Snyder, Curr. Opin. Chem. Biol. 5, 40 (2001). 4. M. R. Martzen et al., Science 286, 1153 (1999). 5. H. Zhu et al., Nature Genet. 26, 283 (2000). 6. G. MacBeath, S. L. Schreiber, Science 289, 1760 (2000). 7. A. Caveman, J. Cell Sci. 113, 3543 (2000). 8. P. Arenkov et al., Anal. Biochem. 278, 123 (2000). 9. D. A. Mitchell, T. K. Marshall, R. J. Deschenes, Yeast 9, 715 (1993). The expression vector pEGH was created by inserting an RGS-HisX6 epitope tag between the GST gene and the polycloning site of pEG(KG). The yeast ORFs were cloned using the strategy described previously (5), except every step was done in a 96-well format. Plasmid DNAs confirmed by DNA sequencing were reintroduced into both yeast ( Y258) and E. coli (DH5a). The library contains 5800 unique ORFs. 10. For details of 96-well format protein purification protocol, a full list of results from all the experiments, and the design of the positive identification algo- rithms, please visit our public Web site (http:// bioinfo.mbb.yale.edu/proteinchip) and supplementa- ry material at Science Online (www.sciencemag.org/ cgi/content/full/1062191/DC1). 11. Biotinylated calmodulin (CalBiochem, USA) was add- ed to the proteome chip at 0.02 mg/ml in phosphate- buffered saline (PBS) with 0.1 mM calcium and incu- bated in a humidity chamber for 1 hour at room temperature. Calcium (0.1 mM) was present in buff- ers in all subsequent steps. The chip was washed three times with PBS at room temperature (RT, 25��C). Cy3-conjugated streptavidin (Jackson IR, USA) (1:5000 dilution) was added to the chip and incubat- ed for 30 min at RT. After extensive washing, the chip was spun dry and scanned using a microarray scan- ner the data was subsequently acquired with the GenePix array densitometry software (Axon, USA). 12. S. S. Hook, A. R. Means, Annu. Rev. Pharmacol. Toxi- col. 41, 471 (2001). 13. M. S. Cyert, R. Kunisawa, D. Kaim, J. Thorner, Proc. Natl. Acad. Sci. U.S.A. 88, 7376 (1991). 14. D. A. Stirling, K. A. Welch, M. J. Stark, EMBO J. 13, 4329 (1994). 15. F. Bohl, C. Kruse, A. Frank, D. Ferring, R. P. Jansen, EMBO J. 19, 5514 (2000) E. Bertrand et al., Mol. Cell 2, 437 (1998). 16. D. C. Winter, E. Y. Choe, R. Li, Proc. Natl. Acad. Sci. U.S.A. 96, 7288 (1999). 17. C. Schaerer-Brodbeck, H. Riezman, Mol. Biol. Cell 11, 1113 (2000). 18. K. Homma, J. Saito, R. Ikebe, M. Ikebe, J. Biol. Chem. 275, 34766 (2000). 19. J. Menendez, J. Delgado, C. Gancedo, Yeast 14, 647 (1998). 20. G. Odorizzi, M. Babst, S. D. Emr, Trends Biochem. Sci. 25, 229 (2000) D. A. Fruman et al., Annu. Rev. Biochem. 67, 481 (1998) T. F. Martin, Annu. Rev. Cell Dev. Biol. 14, 231 (2000) S. Wera, J. C. T. Bergsma, FEMS Yeast Res. 1, 1406 (2001). 21. Liposomes were prepared using standard methods (30). Briefly, appropriate amounts of each lipid in chloroform were mixed and dried under nitrogen. The lipid mixture was resuspended in TBS buffer by vor- texing. The liposomes were created by sonication. To probe the proteome chips, 60 ml of the different liposomes were added onto different chips. The chips were incubated in a humidity chamber for 1 hour at RT. After washing with TBS buffer for three times, Cy3-conjugated streptavidin (1:5000 dilution) was added to the chip and incubated for 30 min at RT. 22. Positives were identified using a combination of the GenePix software which computes a local in- tensity background for each spot and a series of algorithms we developed. Details can be found at http://bioinfo.mbb.yale.edu/proteinchip and at www.sciencemag.org/cgi/content/full/1062191/ DC1. 23. M. C. Costanzo et al., Nucleic Acids Res. 29, 75 (2001). 24. M. Gerstein, Proteins 33, 518 (1998). 25. K. Ansari et al., J. Biol. Chem. 274, 30052 (1999). 26. Y. Barral, M. Parra, S. Bidlingmaier, M. Snyder, Genes Dev. 13, 176 (1999). 27. Y. Li, T. Kane, C. Tipper, P. Spatrick, D. D. Jenness, Mol. Cell. Biol. 19, 3588 (1999). 28. I. Arnold et al., J. Biol. Chem. 274, 36 (1999). 29. S. Chu et al., Science 282, 699 (1998). 30. A. Casamayor et al., Curr. Biol. 9, 186 (1999) R. Guerra, M. L. Bianconi, Biosci. Rep. 20, 41 (2000). 31. M. Pardo et al., Yeast 15, 459 (1999). 32. Single-letter abbreviations for the amino acid resi- dues are as follows: A, Ala C, Cys D, Asp E, Glu F, Phe G, Gly H, His I, Ile K, Lys L, Leu M, Met N, Asn P, Pro Q, Gln R, Arg S, Ser T, Thr V, Val W, Trp and Y, Tyr. X indicates any residue. 33. We thank K. Nelson and S. Dellaporta for providing invaluable help. We also thank A. Kumar, G. Michaud, and C. Costigan for providing comments on the manuscript. This research is supported by grants from NIH. H.Z., A.C., and R.J. were supported by postdoc- toral fellowships from the Damon Runyon���Walter Winchell Foundation, the Spanish Ministerio de Cien- cia y Tecnologia, and by an IBM Graduate Research Fellowship, respectively. 2 May 2001 accepted 13 July 2001 Published online 26 July 2001 10.1126/science.1062191 Include this information when citing this paper. An fMRI Investigation of Emotional Engagement in Moral Judgment Joshua D. Greene,1,2* R. Brian Sommerville,1 Leigh E. Nystrom,1,3 John M. Darley,3 Jonathan D. Cohen1,3,4 The long-standing rationalist tradition in moral psychology emphasizes the role of reason in moral judgment. A more recent trend places increased emphasis on emotion. Although both reason and emotion are likely to play important roles in moral judgment, relatively little is known about their neural correlates, the nature of their interaction, and the factors that modulate their respective behavioral influences in the context of moral judgment. In two functional magnetic resonance imaging (fMRI) studies using moral dilemmas as probes, we apply the methods of cognitive neuroscience to the study of moral judgment. We argue that moral dilemmas vary systematically in the extent to which they engage emotional processing and that these variations in emotional engage- ment influence moral judgment. These results may shed light on some puzzling patterns in moral judgment observed by contemporary philosophers. The present study was inspired by a family of ethical dilemmas familiar to contemporary moral philosophers (1). One such dilemma is the trolley dilemma: A runaway trolley is headed for five people who will be killed if it proceeds on its present course. The only way to save them is to hit a switch that will turn the trolley onto an alternate set of tracks where it will kill one person instead of five. Ought you to turn the trolley in order to save five people at the expense of one? Most people say yes. Now consider a similar prob- lem, the footbridge dilemma. As before, a trolley threatens to kill five people. You are standing next to a large stranger on a foot- bridge that spans the tracks, in between the oncoming trolley and the five people. In this scenario, the only way to save the five people is to push this stranger off the bridge, onto the tracks below. He will die if you do this, but his body will stop the trolley from reaching the others. Ought you to save the five others by pushing this stranger to his death? Most people say no. Taken together, these two dilemmas cre- ate a puzzle for moral philosophers: What makes it morally acceptable to sacrifice one life to save five in the trolley dilemma but not in the footbridge dilemma? Many answers have been proposed. For example, one might suggest, in a Kantian vein, that the difference between these two cases lies in the fact that in the footbridge dilemma one literally uses a fellow human being as a means to some independent end, whereas in the trolley di- lemma the unfortunate person just happens to 1Center for the Study of Brain, Mind, and Behavior, 2Department of Philosophy, 1879 Hall, 3Department of Psychology, Green Hall, Princeton University, Princeton, NJ 08544, USA. 4Department of Psychiatry, University of Pittsburgh, Pittsburgh, PA 15260, USA. *To whom correspondence should be addressed. E- mail: jdgreene@princeton.edu R E P O R T S www.sciencemag.org SCIENCE VOL 293 14 SEPTEMBER 2001 2105

be in the way. This answer, however, runs into trouble with a variant of the trolley di- lemma in which the track leading to the one person loops around to connect with the track leading to the five people (1). Here we will suppose that without a body on the alternate track, the trolley would, if turned that way, make its way to the other track and kill the five people as well. In this variant, as in the footbridge dilemma, you would use some- one���s body to stop the trolley from killing the five. Most agree, nevertheless, that it is still appropriate to turn the trolley in this case in spite of the fact that here, too, we have a case of ���using���. These are just one proposed so- lution and one counterexample, but together they illustrate the sort of dialectical difficul- ties that all proposed solutions to this prob- lem have encountered. If a solution to this problem exists, it is not obvious. That is, there is no set of consistent, readily accessible moral principles that captures people���s intui- tions concerning what behavior is or is not appropriate in these and similar cases. This leaves psychologists with a puzzle of their own: How is it that nearly everyone manages to conclude that it is acceptable to sacrifice one life for five in the trolley dilemma but not in the footbridge dilemma, in spite of the fact that a satisfying justification for distinguish- ing between these two cases is remarkably difficult to find (2)? We maintain that, from a psychological point of view, the crucial difference between the trolley dilemma and the footbridge dilem- ma lies in the latter���s tendency to engage people���s emotions in a way that the former does not. The thought of pushing someone to his death is, we propose, more emotionally salient than the thought of hitting a switch that will cause a trolley to produce similar consequences, and it is this emotional re- sponse that accounts for people���s tendency to treat these cases differently. This hypothesis concerning these two cases suggests a more general hypothesis concerning moral judg- ment: Some moral dilemmas (those relevant- ly similar to the footbridge dilemma) engage emotional processing to a greater extent than others (those relevantly similar to the trolley dilemma), and these differences in emotional engagement affect people���s judgments. The present investigation is an attempt to test this more general hypothesis. Drawing upon re- cent work concerning the neural correlates of emotion (3���5), we predicted that brain areas associated with emotion would be more ac- tive during contemplation of dilemmas such as the footbridge dilemma as compared to during contemplation of dilemmas such as the trolley dilemma. In addition, we predicted a pattern of behavioral interference similar to that observed in cognitive tasks in which automatic processes can influence responses, such as the Stroop task (in which the identity of a color word can interfere with partici- pants��� ability to name the color in which it is displayed e.g., the ability to say ���green��� in response to the word ���red��� written in green ink) (6, 7). In light of our proposal that people tend to have a salient, automatic emo- tional response to the footbridge dilemma that leads them to judge the action it proposes to be inappropriate, we would expect those (relatively rare) individuals who nevertheless judge this action to be appropriate to do so against a countervailing emotional response and to exhibit longer reaction times as a result of this emotional interference. More general- ly, we predicted longer reaction times for trials in which the participant���s response is incongruent with the emotional response (e.g., saying ���appropriate��� to a dilemma such as the footbridge dilemma). We predicted the absence of such effects for dilemmas such as the trolley dilemma which, according to our theory, are less likely to elicit a strong emo- tional response. In each of two studies, Experiments 1 and 2, we used a battery of 60 practical dilemmas (8). These dilemmas were divided into ���mor- al��� and ���non-moral��� categories on the basis of the responses of pilot participants (8). (Typical examples of non-moral dilemmas posed questions about whether to travel by bus or by train given certain time constraints and about which of two coupons to use at a store.) Two independent coders evaluated each moral dilemma using three criteria de- signed to capture the difference between the intuitively ���up close and personal��� (and pu- tatively more emotional) sort of violation ex- hibited by the footbridge dilemma and the more intuitively impersonal (and putatively less emotional) violation exhibited by the trolley dilemma (8, 9). Moral dilemmas meet- ing these criteria were assigned to the ���moral- personal��� condition, the others to the ���moral- impersonal��� condition. Typical moral-per- sonal dilemmas included a version of the footbridge dilemma, a case of stealing one person���s organs in order to distribute them to five others, and a case of throwing people off a sinking lifeboat. Typical moral-impersonal dilemmas included a version of the trolley dilemma, a case of keeping money found in a lost wallet, and a case of voting for a policy expected to cause more deaths than its alter- natives. Participants responded to each di- lemma by indicating whether they judged the action it proposes to be ���appropriate��� or ���in- appropriate.��� In each experiment, nine participants (10) responded to each of 60 dilemmas (11) while undergoing brain scanning using fMRI (12). Figures 1 and 2 describe brain areas identified in Experiment 1 by a thresholded omnibus anal- ysis of variance (ANOVA) performed on the functional images (13). In each case, the Fig. 1. Effect of condition on activity in brain areas identified in Experiment 1. R, right L, left B, bilateral. Results for the middle frontal gyrus were not rep- licated in Experiment 2. The moral-personal condi- tion was significantly dif- ferent from the other two conditions in all other ar- eas in both Experiments 1 and 2. In Experiment 1 the medial frontal and posteri- or cingulate gyri showed significant differences be- tween the moral-imper- sonal and non-moral con- ditions. In Experiment 2 only the posterior cingu- late gyrus was significantly different in this comparison. Brodmann���s Areas and Talairach (28) coordi- nates (x, y, z) for each area are as follows (left to right in graph): 9/10 (1, 52, 17) 31 (���4, ���54, 35) 46 (45, 36, 24) 7/40 (���48, ���65, 26) 7/40 (50, ���57, 20). Fig. 2. Brain areas ex- hibiting differences in activity between con- ditions shown in three axial slices of a stan- dard brain (28). Slice location is indicated by Talairach (28) z co- ordinate. Data are for the main effect of condition in Experi- ment 1. Colored areas reflect the thresholded F scores. Images are reversed left to right to follow radiologic convention. R E P O R T S 14 SEPTEMBER 2001 VOL 293 SCIENCE www.sciencemag.org 2106