Running head: REVISITING THE SEROTONIN-AGGRESSION RELATION 1 Revisiting the Serotonin-Aggression Relation in Humans: A Meta-analysis Aaron A. Duke University Pierre Mendès-France, Grenoble and University of Kentucky Laurent Bègue University Pierre Mendès-France, Grenoble Rob Bell Queen’s University Belfast Tory Eisenlohr-Moul University of Kentucky Author Note Aaron A. Duke, University Pierre Mendès-France, Grenoble and Department of Psychology, University of Kentucky; Laurent Bègue, University Pierre Mendès-France, Grenoble; Rob Bell, School of Psychology, Queen’s University Belfast; Tory Eisenlohr-Moul, Department of Psychology, University of Kentucky. This research was supported in part by a doctoral grant from the French Ministry of Research and grant T32DA007304 from the National Institute on Drug Abuse. We would like to thank the several researchers who responded enthusiastically to our requests for unpublished data. Correspondence concerning this article should be addressed to Aaron A. Duke, Department of Psychology, University of Kentucky, Lexington, KY 40506. Email: [email protected] REVISITING THE SEROTONIN-AGGRESSION RELATION 2 Abstract The inverse relation between serotonin and human aggression is often portrayed as “reliable,” “strong,” and “well-established” despite decades of conflicting reports and widely recognized methodological limitations. In this systematic review and meta-analysis we evaluate the evidence for and against the serotonin deficiency hypothesis of human aggression across four methods of assessing serotonin: (a) cerebrospinal fluid levels of 5-hydroxyindoleacetic acid (CSF 5-HIAA), (b) acute tryptophan depletion, (c) pharmacological challenge, and (d) endocrine challenge. Results across 175 independent samples and over 6,500 total participants were heterogeneous, but, in aggregate, revealed a small, inverse correlation between central serotonin functioning and aggression, anger, and hostility, r = -.12. Pharmacological challenge studies had the largest mean weighted effect size, r = -.21, and CSF 5-HIAA studies had the smallest, r = -.06, p = .21. Potential methodological and demographic moderators largely failed to account for variability in study outcomes. Notable exceptions included year of publication (effect sizes tended to diminish with time) and self- versus other-reported aggression (other-reported aggression was positively correlated to serotonin functioning). We discuss four possible explanations for the pattern of findings: unreliable measures, ambient correlational noise, an unidentified higher-order interaction, and a selective serotonergic effect. Finally, we provide four recommendations for bring much needed clarity to this important area of research: acknowledge contradictory findings and avoid selective reporting practices, focus on improving the reliability and validity of serotonin and aggression measures, test for interactions involving personality and/or environmental moderators, and revise the serotonin deficiency hypothesis to account for serotonin’s functional complexity. Keywords: serotonin, 5-HT, aggression, anger, hostility REVISITING THE SEROTONIN-AGGRESSION RELATION 3 Revisiting the Serotonin-Aggression Relation in Humans: A Meta-analysis “The greatest enemy of knowledge is not ignorance, it is the illusion of knowledge.” — Stephen Hawking There has been considerable enthusiasm at pinpointing human aggression’s neurochemical origins ever since Brown and colleagues’ original finding that cerebrospinal fluid (CSF) levels of the principle serotonin metabolite, 5-hydroxyindoleacetic acid (5-HIAA), “accounted for 80% of the variance in aggression scores” among a group of military men diagnosed with personality disorders (Brown, Goodwin, Ballenger, Goyer, & Major, 1979, p. 133). These results were interpreted to suggest that a deficiency in serotonin was largely responsible for these men’s aggressive behavior. This serotonin deficiency hypothesis of human aggression has been tested hundreds of times over the past several decades and remains the most common hypothesis of serotonin’s role in pathological aggression (e.g., Montoya, Terburg, Bos, & van Honk, 2012; Passamonti et al., 2012; Raine, 2008; Stadler et al., 2007; Steiger et al., 2004; Yanowitch & Coccaro, 2011). One widely cited author called the inverse relationship between serotonin activity and aggression “perhaps the most reliable finding in the history of psychiatry” (Fishbein, 2001, p. 15). Glancing through several recently published psychology and psychiatry textbooks reveals a concordant lack of controversy—no mention of contradictory findings (e.g., Bushman & Bartholow, 2010; Carlson, 2010; Higley & Barr, 2007; Hollander & Berlin, 2008). One sine qua non of science is a healthy skepticism, and with a little bit of effort one can find numerous examples of studies finding no relation, or a positive relation between serotonin and human aggression (e.g., Booij, Tremblay, et al., 2010; Castellanos et al., 1994; REVISITING THE SEROTONIN-AGGRESSION RELATION 4 Gardner, Lucas, & Cowdry, 1990; Germine, Goddard, Woods, Charney, & Heninger, 1992; Handelsman et al., 1998; Koszycki, Zacharko, Le Melledo, Young, & Bradwejn, 1996; Lopez et al., 1996; Tuinier, Verhoeven, & van Praag, 1996; Wood, Rilling, Sanfey, Bhagwagar, & Rogers, 2006). Understanding of the relation between serotonin and human aggression will be advanced by revisiting this “most reliable finding” in a more inclusive and objective light. We begin this review by critically evaluating the evidence for the serotonin deficiency hypothesis across four methods of assessing serotonin: CSF 5-HIAA, acute tryptophan depletion, pharmacological challenge, and endocrine challenge. We then introduce a comprehensive meta-analysis designed to test whether divergent findings in the literature can be explained by taking into account a variety of methodological and sample- level moderators. CSF 5-HIAA There is too much overlap between the mean 5-HIAA levels of “normal” and “affected” populations for the term “low 5-HIAA levels” to have a scientifically legitimate meaning … Thus, low 5-HIAA levels are inappropriate markers for increased “risk” for any specific psychiatric condition or behavior in the general population or in comparison among individuals. (Balaban, Alper, & Kasamon, 1996, p. 29). Several reviews have been published examining the relation between 5-HIAA and aggressive, disruptive, or antisocial behaviors (e.g., Åsberg, 1994; Balaban et al., 1996; Moore, Scarpa, & Raine, 2002; Tuinier, Verhoeven, & van Praag, 1995, 1996). The results from these reviews have been equivocal. The quotation at the beginning of this section was taken from an early meta-analysis (Balaban et al., 1996), which found that 5-HIAA levels did not differ REVISITING THE SEROTONIN-AGGRESSION RELATION 5 significantly between aggressive and unaggressive clinical groups and that both aggressive and unaggressive clinical groups had a small, but significant reduction in 5-HIAA levels when compared to nonclinical controls. Another early review (Tuinier et al., 1996) concluded that only 8 out of 23 studies published at the time on 5-HIAA and aggression provided support for an inverse relation. Furthermore, these 8 studies were almost entirely restricted to personality- disordered young adult males such as those in Brown and colleague’s original study (Brown et al., 1979). A 2002 meta-analysis of 16 studies looking at 5-HIAA and adult antisocial behavior (Moore et al., 2002) revealed a “medium” (Cohen, 1988, p. 157) mean weighted effect size of d = -0.45. Results revealed no significant effects of gender, alcoholism, suicidality, or whether violence was against persons or property. Age, however, was found to significantly moderate the relation between 5-HIAA and aggression such that studies with young adult, antisocial individuals exhibited a much larger effect than studies with older antisocial individuals. There are a number of limitations in the two meta-analyses just mentioned that require further qualification. Both Balaban et al. (1996) and Moore et al. (2002) failed to account for study quality, excluded studies with limited statistical information (e.g., studies reporting only that findings were “non-significant”), and did not test, or attempt to account for, publication bias. Thus, there is reason to believe that these meta-analyses may have included a sample of studies biased towards finding statistically significant results. The failure to clarify the ambiguous relation between 5-HIAA and aggression may be in part due to difficulties in establishing reference levels of CSF 5-HIAA. CSF 5-HIAA has been shown to be moderated by age (e.g., Hedner, Lundell, Breese, Mueller, & Hedner, 1986; Seifert, Foxx, & Butler, 1980; Takeuchi et al., 2000), gender (e.g., Blennow et al., 1993; Hagenfeldt, Bjerkenstedt, Edman, Sedvall, & Wiesel, 1984), height and weight (e.g., Blennow et al., 1993; REVISITING THE SEROTONIN-AGGRESSION RELATION 6 Hartikainen et al., 1991; Strömbom et al., 1996), physical activity (Eklundh, Gunnarsson, & Nordin, 2001; Nordin, Lindstöm, & Wieselgren, 1996), season (e.g., Brewerton, Berrettini, Nurnberger, & Linnoila, 1988; Hartikainen et al., 1991), atmospheric pressure (Eklundh, Fernström, & Nordin, 1994; Nordin, Swedin, & Zachau, 1992), and intra-spinal pressure (Eklundh & Nordin, 2001) along with several psychiatric and neurological conditions (for a review of CSF 5-HIAA moderators see Dhondt, 2004). One of the most serious confounding influences associated with measuring CSF 5-HIAA may be an inverse relation between CSF5- HIAA levels and stress associated with the lumbar puncture procedure itself (Hill et al., 1999). Despite CSF 5-HIAA’s important historical role as the first biomarker used to support the serotonin deficiency hypothesis in humans, there remain misgivings about the reliability and validity of using CSF 5-HIAA as an index of central serotonergic functioning (e.g., Balaban et al., 1996; Hyland, 2008; van der Vegt, Lieuwes, Cremers, de Boer, & Koolhaas, 2003). Acute Tryptophan Depletion (ATD) Although ATD … [is] important in the investigation of the monoamine systems, monoamine depletion does not directly decrease mood. (Ruhé, Mason, & Schene, 2007). Acute tryptophan depletion represented the first broad attempt at experimentally manipulating central serotonin levels in humans (Young, Smith, Pihl, & Ervin, 1985). Acute tryptophan depletion relies on the assumption that modifying plasma levels of L-tryptophan, the primary precursor of serotonin, will lead to corresponding modifications in levels of brain serotonin (Hood, Bell, & Nutt, 2005). Because the blood brain barrier prevents peripheral serotonin from entering directly into the central nervous system, serotonin in the brain must be synthesized locally from its precursor tryptophan. Humans receive all of their tryptophan REVISITING THE SEROTONIN-AGGRESSION RELATION 7 from dietary sources (Moore et al., 2000), and therefore, dietary tryptophan intake can be experimentally manipulated to influence central serotonin availability. However, free plasma levels of tryptophan are relatively resilient to short periods of tryptophan-free dieting with a typical reduction in only about 15–20% of central serotonin availability (Delgado, Charney, Price, Landis, & Heninger, 1989). Thus, while often used as an experimental control, tryptophan-free diets are less effective at reducing serotonin availability than acute tryptophan depletion. Acute tryptophan depletion relies on two additional physiological processes in order to reduce central serotonin availability (for an overview of the rationale and methodology of acute tryptophan depletion, see Hood et al., 2005). First, tryptophan must compete with five other large neutral amino acids (LNAA) for active-transport across the blood brain barrier; therefore, the ratio of free plasma tryptophan to other LNAAs determines the rate by which tryptophan can cross the blood brain barrier and be synthesized into serotonin. The tryptophan/LNAA ratio can be augmented by increasing available tryptophan relative to other LNAAs (e.g., through direct administration of tryptophan supplements), and can be attenuated (i.e., in acute tryptophan depletion) by increasing the availability of other LNAAs relative to tryptophan (e.g., through administration of non- tryptophan supplements including other LNAAs). The second process has to do with synthesis of proteins in the liver. When there is a dietary influx of amino acids, the liver will begin to synthesize various new proteins, many of which require tryptophan. If an amino-acid mixture (containing both LNAAs and non-LNAAs) is administered without any tryptophan included, the liver will use existing plasma tryptophan for protein-synthesis, thereby further reducing tryptophan’s ability to be synthesized into serotonin. While a variety of different amino acid mixtures have been utilized in acute tryptophan REVISITING THE SEROTONIN-AGGRESSION RELATION 8 depletion studies, they can be roughly divided into three categories: tryptophan-free (T-) amino acid mixtures, balanced (B) amino acid mixtures, and tryptophan-enhanced (T+) mixtures. Young, as the first to use acute tryptophan depletion in humans, standardized the 100 gram T- amino mixture with 15 amino acids in the proportion that are found in human breast milk minus the tryptophan (Young et al., 1985). Young’s balanced mixture consisted of adding 2.3 grams of tryptophan to the 100 gram mixture while the T+ mixture included 10.3 grams of tryptophan (i.e., a higher than normal ratio of tryptophan compared to the other amino acids). In these mixtures, the amino acids aspartic acid and glutamic acid are typically excluded due to concerns about toxicity (Hood et al., 2005). Further, these amino acid mixtures are renowned for having a very “unpalatable” taste to them—an issue that has been addressed differently by different researchers. Often something is added to the mixture to improve its taste (e.g., chocolate syrup). Also, researchers began administering amino acids containing sulfur separately from the rest of the mixture in the form of capsules, as these amino acids (methionine, cystine, and arginine) are considered the most unpalatable. Similar amino-acid depletion techniques have been adopted for other monoamines and a number of reviews have been conducted examining the effects of monoamine depletion on mood (Ruhé et al., 2007; van der Does, 2001), anxiety (Anderson & Mortimore, 1999), psychiatric disorders (Bell, Hood, & Nutt, 2005; Booij, Van der Does, & Riedel, 2003; Moore et al., 2002; Reilly, McTavish, & Young, 1997; Sobczak, Honig, Duinen, & Riedel, 2002), neural activation (Fusar-Poli et al., 2006), memory (Sambeth et al., 2007), executive functioning (Mendelsohn, Riedel, & Sambeth, 2009), and cognitive flexibility (Evers, van der Veen, Fekkes, & Jolles, 2007). Booij, Van der Does, and Riedel analyzed the literature on acute tryptophan depletion and aggression and found several contradictory findings. These REVISITING THE SEROTONIN-AGGRESSION RELATION 9 authors concluded that there was evidence for acute tryptophan depletion increasing aggression in individuals predisposed to act aggressively, but not in individuals with low trait levels of aggression. A recent systematic review of acute tryptophan depletion’s effects on executive functioning concluded that the majority of findings indicate that acute tryptophan depletion does not impair key executive functioning processes including response inhibition, decision making, planning, sustained attention, and set shifting (Mendelsohn et al., 2009). It should be noted that impaired executive functioning has been argued to be important in predicting many forms of impulsive aggression (e.g., Hancock, Tapscott, & Hoaken, 2010). In the only relevant meta-analysis conducted on monoamine depletion studies, acute tryptophan depletion was found to have no effect on mood in healthy individuals, but to moderately decrease mood in individuals with a history of depression (Ruhé et al., 2007). Finally, findings from a recent review of an alternate, but much less common approach to serotonin manipulation—tryptophan augmentation—have suggested similarly equivocal results with respect to mood and similarly negative results with respect to executive functioning (Silber & Schmitt, 2010). Pharmacological/Endocrine Challenge Caution seems to be indicated in drawing inferences about the functional status of central serotonergic neurotransmission from neuroendocrine responses elicited by challenge studies using serotonergic agents. (Murphy, Aulakh, Mazzola- Pomietto, & Briggs, 1996, p. 213). Coinciding with the increase in availability of drugs that more selectively target various serotoninergic receptors, acute pharmacological challenge tests began to be applied to the study of human aggression in the late 1980s. In these studies, a serotonergic agent is administered and acute changes in aggression and related constructs are measured over the REVISITING THE SEROTONIN-AGGRESSION RELATION 10 course of several hours. Similar to acute tryptophan depletion, pharmacological challenge studies allow for experimental manipulation of central serotonin levels and typically involve comparing mean responses across different drug dosage conditions. However, group level data derived from pharmacological challenge studies does not allow for measurement of individual differences in responsivity to manipulation of central serotonin levels. The endocrine challenge paradigm, on the other hand, provides a putative index of individual responsivity to serotonergic agents. Evidence indicating increased hypothalamus-pituitary-adrenal activity following serotonergic pharmacological challenges led to the use of pituitary hormones (e.g., prolactin, adrenocorticotropic hormone, growth hormone) and adrenal hormones (e.g., cortisol) as indices of individual serotoninergic functioning (Cowen, Anderson, & Gartside, 1990). Thus, endocrine challenge studies are pharmacological challenge studies in which hormonal endpoints are used as an index of central serotonergic functioning. Some concerns have been raised concerning the implications of differential hormonal responses following serotonergic challenges. For example, the quotation above was taken from Murphy et al. (1996) in a review of “neuroendocrine responses to serotonergic agonists as indices of the functional status of central serotonin neurotransmission in humans,” in which a high level of discordance was found between different endocrine endpoints (e.g., prolactin and cortisol), temperature, and behavioral measures. Despite the popularity of endocrine challenge studies, information concerning their validity remains somewhat sparse. A number of different pharmacological agents have been used in challenge studies, though the most popular choice has been d,l-fenflurmaine (d,l-FEN), the racemic mixture of