Case-control studies point to trans-diagnostic mechanisms in OCD: error-related negativity (ERN) and anxiety

Can OCD be better understood and treated from a dimensional perspective? Evidence for this comes from many sources (Hyman, Reference Hyman2010); of these, perhaps the most compelling (and frustrating) is the lack of specificity of various neurocognitive deficits observed in OCD (Endrass & Ullsperger, Reference Endrass and Ullsperger2014; Fineberg et al. Reference Fineberg, Chamberlain, Goudriaan, Stein, Vanderschuren, Gillan, Shekar, Gorwood, Voon, Morein-Zamir, Denys, Sahakian, Moeller, Robbins and Potenza2014). For example, ‘response inhibition’, the ability to cancel a prepotent motor response assayed using the stop-signal reaction-time task (Logan, Reference Logan, Dagenbach and Carr1994), is reliably impaired in not just OCD, but also attention deficit hyperactivity disorder (ADHD) and schizophrenia (Lipszyc & Schachar, Reference Lipszyc and Schachar2010). Pessimistically, one might conclude from this lack of specificity that our current neurocognitive markers are not useful for understanding the aetiology of OCD. Perhaps these markers reflect generalized executive impairment, a consequence of having any psychiatric disorder and the burdens thereof. Perhaps these markers are not diagnostic of any specific mental health issue, but simply place individuals into an at-risk state for many. What we will consider in the following sections is an admittedly well-trodden alternative: that it is our psychiatric taxonomy, not our neurocognitive models, that presents a problem for research and that the specificity of any neurocognitive model of a psychiatric diagnosis is fundamentally limited by the extent to which that diagnosis is biologically valid.

How can we determine if a neurocognitive model is specific to a trans-diagnostic phenotype, if not one of the existing DSM disorder categories? What does this mean for interpreting the results of prior work in these DSM diagnosed patient populations? A reasonable starting point is to first identify a model of biological relevance (e.g. neurocognitive or physiological marker) that exhibits partial specificity for psychiatric disorders that is perhaps not specific to one disorder, but certainly not common to all. One can then qualitatively identify phenotypic commonalities across the disorders that exhibit mechanistic commonalities and thus formulate hypotheses about trans-diagnostic phenotypes (Robbins et al. Reference Robbins, Gillan, Smith, de Wit and Ersche2012; Fineberg et al. Reference Fineberg, Chamberlain, Goudriaan, Stein, Vanderschuren, Gillan, Shekar, Gorwood, Voon, Morein-Zamir, Denys, Sahakian, Moeller, Robbins and Potenza2014). Publication bias is an obstacle here as studies showing no difference between patient cohorts and controls, which are needed to make a case for specificity, are less likely to be published or even submitted for publication (Rosenthal, Reference Rosenthal1979). One solution (albeit imperfect) to this problem is studying multiple disorders at once, thereby allowing for direct statistical comparison of disorders that have established deficits in a given domain and disorders that are hypothesized to be no different from controls. Failing this, one can gain insights by leveraging cases where a given neurocognitive mechanism is significantly enhanced in one cluster of disorders, but deficient in another (thereby avoiding publication bias), as is the case for one very promising neurocognitive deficit in OCD that we will discuss in some detail, ERN (Endrass & Ullsperger, Reference Endrass and Ullsperger2014).

The ERN is a negative deflection of the event-related potential (ERP) when subjects make an error (Falkenstein et al. Reference Falkenstein, Hohnsbein, Hoormann and Blanke1991), listed as the RDoC sub-construct ‘Performance Monitoring’. It is typically measured using electroencephalogram, but functional imaging has been coupled with this technique to confirm that the ERN has its source in the anterior cingulate cortex (ACC) (Debener et al. Reference Debener, Ullsperger, Siegel, Fiehler, von Cramon and Engel2005). A host of studies have found enhanced ERN in OCD (Gehring et al. Reference Gehring, Himle and Nisenson2000; Johannes et al. Reference Johannes, Wieringa, Nager, Rada, Dengler, Emrich, Münte and Dietrich2001b ; Ruchsow et al. Reference Ruchsow, Gron, Reuter, Spitzer, Hermle and Kiefer2005, Reference Ruchsow, Reuter, Hermle, Ebert, Kiefer and Falkenstein2007; Endrass et al. Reference Endrass, Klawohn, Schuster and Kathmann2008, Reference Endrass, Schuermann, Kaufmann, Spielberg, Kniesche and Kathmann2010, Reference Endrass, Riesel, Kathmann and Buhlmann2014; Hajcak et al. Reference Hajcak, Franklin, Foa and Simons2008; Stern et al. Reference Stern, Liu, Gehring, Lister, Yin, Zhang, Fitzgerald, Himle, Abelson and Taylor2010; Riesel et al. Reference Riesel, Endrass, Kaufmann and Kathmann2011, Reference Riesel, Endrass, Auerbach and Kathmann2015; Xiao et al. Reference Xiao, Wang, Zhang, Li, Tang, Wang, Fan and Fromson2011; Hanna et al. Reference Hanna, Carrasco, Harbin, Nienhuis, LaRosa, Chen, Fitzgerald and Gehring2012; Carrasco et al. Reference Carrasco, Harbin, Nienhuis, Fitzgerald, Gehring and Hanna2013a , Reference Carrasco, Hong, Nienhuis, Harbin, Fitzgerald, Gehring and Hanna b ; Grützmann et al. Reference Grützmann, Endrass, Kaufmann, Allen, Eichele and Kathmann2014; Klawohn et al. Reference Klawohn, Riesel, Grützmann, Kathmann and Endrass2014; Liu et al. Reference Liu, Hanna, Carrasco, Gehring and Fitzgerald2014) (but see Nieuwenhuis et al. Reference Nieuwenhuis, Nielen, Mol, Hajcak and Veltman2005; Agam et al. Reference Agam, Greenberg, Isom, Falkenstein, Jenike, Wilhelm and Manoach2014; Mathews et al. Reference Mathews, Perez, Roach, Fekri, Vigil, Kupferman and Mathalon2015; Weinberg et al. Reference Weinberg, Kotov and Proudfit2015), GAD (Ladouceur et al. Reference Ladouceur, Dahl, Birmaher, Axelson and Ryan2006; Weinberg et al. Reference Weinberg, Klein and Hajcak2012, Reference Weinberg, Kotov and Proudfit2015, Reference Weinberg, Olvet and Hajcak2010) (but see Xiao et al. Reference Xiao, Wang, Zhang, Li, Tang, Wang, Fan and Fromson2011), social anxiety disorder (Endrass et al. Reference Endrass, Riesel, Kathmann and Buhlmann2014) and depression (Chiu & Deldin, Reference Chiu and Deldin2007; Holmes & Pizzagalli, Reference Holmes and Pizzagalli2008, Reference Holmes and Pizzagalli2010; Georgiadi et al. Reference Georgiadi, Liotti, Nixon and Liddle2011; Aarts et al. Reference Aarts, Vanderhasselt, Otte, Baeken and Pourtois2013; Tang et al. Reference Tang, Zhang, Simmonite, Li, Zhang, Guo, Li, Fang, Xu and Wang2013; Mueller et al. Reference Mueller, Pechtel, Cohen, Douglas and Pizzagalli2015) (although less consistently than in generalized anxiety or OCD, see Ruchsow et al. Reference Ruchsow, Herrnberger, Wiesend, Grön, Spitzer and Kiefer2004, Reference Ruchsow, Herrnberger, Beschoner, Grön, Spitzer and Kiefer2006; Olvet et al. Reference Olvet, Klein and Hajcak2010; Georgiadi et al. Reference Georgiadi, Liotti, Nixon and Liddle2011; Weinberg et al. Reference Weinberg, Klein and Hajcak2012) (Table 1). Accounts regarding the precise function of ERN are varied (Holroyd & Coles, Reference Holroyd and Coles2002; Yeung et al. Reference Yeung, Botvinick and Cohen2004), but studies have shown that the ERN primes defensive responses, for example by potentiating startle responses (Hajcak & Foti, Reference Hajcak and Foti2008) and encouraging avoidance behaviour (Frank et al. Reference Frank, Woroch and Curran2005). Together with a qualitative examination of the symptom commonalities across these disorders, this suggests that anxiety might be a reasonable candidate for consideration as a dimensional phenotype related to these deficits. This thesis is supported by work in non-clinical samples, wherein the magnitude of the ERN is associated with self-reported worry (Hajcak et al. Reference Hajcak, McDonald and Simons2003; Zambrano-Vazquez & Allen, Reference Zambrano-Vazquez and Allen2014), high levels of trait anxiety (Paulus et al. Reference Paulus, Feinstein, Simmons and Stein2004) and OC-symptoms (Hajcak & Simons, Reference Hajcak and Simons2002; Zambrano-Vazquez & Allen, Reference Zambrano-Vazquez and Allen2014). The effect appears to be trait (rather than state) dependent; the ERN remains enhanced following successful treatment in both adult and paediatric OCD (Hajcak et al. Reference Hajcak, Franklin, Foa and Simons2008; Riesel et al. Reference Riesel, Endrass, Auerbach and Kathmann2015), does not differ as a result of chronic medication (Stern et al. Reference Stern, Liu, Gehring, Lister, Yin, Zhang, Fitzgerald, Himle, Abelson and Taylor2010) and an enhanced ERN has been observed in unaffected first-degree relatives of both adult and paediatric OCD sufferers (Riesel et al. Reference Riesel, Endrass, Kaufmann and Kathmann2011; Carrasco et al. Reference Carrasco, Harbin, Nienhuis, Fitzgerald, Gehring and Hanna2013a ), suggesting that this may be an important endophenotype that predisposes individuals to a range of disorders that involve anxiety.

Table 1. Summary of results for studies examining error-related negativity (ERN) across psychiatric disorders

n.s., Non-significant; PTSD, post-traumatic stress disorder; RT, reaction time; RL, reinforcement learning.

^a Unmedicated were free of selective serotonin re-uptake inhibitors or antipsychotics for at least 2 weeks, and free of stimulants for 17 h. In the case of addiction, unmedicated applies when subjects meet the above criteria and are abstinent.

^b Studies were included if they reported results of ERN analysis (either ERN or ERN corrected for correct-related negativity), included a case-control comparison, and reported generic ERN effects using a task that could be compared across studies.

^c Paediatric/adolescent sample.

^d Compared medicated and unmedicated groups, and no medication effect was observed.

^e Compared to other diagnoses in trans-diagnostic design.

^f Mixed anxiety (mostly generalized anxiety disorder).

^g Remitted patients.

^h Mixture adult and adolescent.

ⁱ Attention deficit hyperactivity disorder combined-type only.

Critically, the postulate that an enhanced ERN is a useful marker of an anxious phenotype rests on the specificity of this effect. We can begin to, albeit imperfectly, approach this issue using existing work in diagnosed patients – specifically if enhanced ERN is observed in disorders not in part characterized by pathological anxiety, then this suggests there is more evidence for the pessimistic view suggested earlier. Fortunately, evidence for such specificity is abundant for this particular neurocognitive marker; a large body of work has showed that a significantly decreased ERN magnitude is characteristic of a putatively distinct cluster (or clusters) of disorders, most consistently schizophrenia (Kopp & Rist, Reference Kopp and Rist1999; Alain et al. Reference Alain, McNeely, He, Christensen and West2002; Bates et al. Reference Bates, Kiehl, Laurens and Liddle2002, Reference Bates, Liddle, Kiehl and Ngan2004; Morris et al. Reference Morris, Yee and Nuechterlein2006, Reference Morris, Heerey, Gold and Holroyd2008, Reference Morris, Holroyd, Mann-Wrobel and Gold2011; Mathalon et al. Reference Mathalon, Jorgensen, Roach and Ford2009; Foti et al. Reference Foti, Kotov, Bromet and Hajcak2012; Horan et al. Reference Horan, Foti, Hajcak, Wynn and Green2012; Perez et al. Reference Perez, Ford, Roach, Woods, McGlashan, Srihari, Loewy, Vinogradov and Mathalon2012; Simmonite et al. Reference Simmonite, Bates, Groom, Jackson, Hollis and Liddle2012; Houthoofd et al. Reference Houthoofd, Morrens, Sabbe, Schrijvers, Vandendriessche, Hulstijn and de Bruijn2013; Kansal et al. Reference Kansal, Patriciu and Kiang2014; Minzenberg et al. Reference Minzenberg, Gomes, Yoon, Swaab and Carter2014; de la Asuncion et al. Reference de la Asuncion, Docx, Morrens, Sabbe and de Bruijn2015; Reinhart et al. Reference Reinhart, Zhu, Park and Woodman2015), but also bipolar disorder (Minzenberg et al. Reference Minzenberg, Gomes, Yoon, Swaab and Carter2014; Morsel et al. Reference Morsel, Morrens, Temmerman, Sabbe and de Bruijn2014) (but see Kopf et al. Reference Kopf, Volkert, Heidler, Dresler, Kittel-Schneider, Gessner, Herrmann, Ehlis and Reif2015), autism spectrum disorders (Vlamings et al. Reference Vlamings, Jonkman, Hoeksma, van Engeland and Kemner2008; Sokhadze et al. Reference Sokhadze, Baruth, El-Baz, Horrell, Sokhadze, Carroll, Tasman, Sears and Casanova2010, Reference Sokhadze, Baruth, Sears, Sokhadze, El-Baz, Williams, Klapheke and Casanova2012; South et al. Reference South, Larson, Krauskopf and Clawson2010) (but see Groen et al. Reference Groen, Wijers, Mulder, Waggeveld, Minderaa and Althaus2008; Henderson et al. Reference Henderson, Ono, McMahon, Schwartz, Usher and Mundy2015) and various forms of addiction (Forman et al. Reference Forman, Dougherty, Casey, Siegle, Braver, Barch, Stenger, Wick-Hull, Pisarov and Lorensen2004; Franken et al. Reference Franken, van Strien, Franzek and van de Wetering2007; Sokhadze et al. Reference Sokhadze, Stewart, Hollifield and Tasman2008; Luijten et al. Reference Luijten, van Meel and Franken2011; Zhou et al. Reference Zhou, Li and Zhu2013) (but see Schellekens et al. Reference Schellekens, de Bruijn, van Lankveld, Hulstijn, Buitelaar, de Jong and Verkes2010; Chen et al. Reference Chen, Jiang, Guo, Du, Wang and Zhao2013). Moreover, the amplitude of the ERN was found to predict treatment adherence for substance abuse (Steele et al. Reference Steele, Fink, Maurer, Arbabshirani, Wilber, Jaffe, Sidz, Pearlson, Calhoun, Clark and Kiehl2014) and decreases in the ERN are associated with self-reported impulsivity (Potts et al. Reference Potts, George, Martin and Barratt2006). Results in ADHD are mixed with some studies showing decreased ERN (Liotti et al. Reference Liotti, Pliszka, Perez, Kothmann and Woldorff2005; van Meel et al. Reference van Meel, Heslenfeld, Oosterlaan and Sergeant2007; Groen et al. Reference Groen, Wijers, Mulder, Waggeveld, Minderaa and Althaus2008; Herrmann et al. Reference Herrmann, Mader, Schreppel, Jacob, Heine, Boreatti-Hümmer, Ehlis, Scheuerpflug, Pauli and Fallgatter2010; Groom et al. Reference Groom, Liddle, Scerif, Liddle, Batty, Liotti and Hollis2013), but most reporting no differences (Wiersema et al. Reference Wiersema, van der Meere and Roeyers2005, Reference Wiersema, van der Meere and Roeyers2009; Groen et al. Reference Groen, Wijers, Mulder, Waggeveld, Minderaa and Althaus2008; Wild-Wall et al. Reference Wild-Wall, Oades, Schmidt-Wessels, Christiansen and Falkenstein2009; Zhang et al. Reference Zhang, Wang, Cai and Yan2009; Groom et al. Reference Groom, Cahill, Bates, Jackson, Calton, Liddle and Hollis2010; Sokhadze et al. Reference Sokhadze, Baruth, Sears, Sokhadze, El-Baz, Williams, Klapheke and Casanova2012) and only one study reporting an increase (Burgio-Murphy et al. Reference Burgio-Murphy, Klorman, Shaywitz, Fletcher, Marchione, Holahan, Stuebing, Thatcher and Shaywitz2007) (Table 1). Together, these data provide convergent evidence that an enhancement of the ERN is not ubiquitous in psychiatry, but rather appears to be directionally specific to disorders characterized predominantly by anxiety. This therefore may constitute a trans-diagnostic marker relevant for many disorders involving anxiety, including OCD.

Validating a trans-diagnostic mechanism: goal-directed learning and compulsivity

The suggestion that the consistent finding of enhanced ERN in OCD in fact reflects a trans-diagnostic phenotype representing anxiety is perhaps a compelling one given the pattern of results described above. However, without direct comparison across disorders and/or discrete symptom dimensions, conclusions regarding its phenomenological specificity cannot be drawn (indeed ‘anxiety’ is intended to serve as a placeholder, until such work is complete). We will now highlight an approach that can take this next step, drawing on data from an independent line of research in OCD. This body of research centres on the theory that an imbalance between ‘goal-directed control’ and ‘habit-learning’ drives symptoms in OCD (see Gillan & Robbins, Reference Gillan and Robbins2014 for detailed account) (see Supplementary material), such that compulsive behaviours in OCD are not goal-directed responses to anxiety/perceived threat, but are in fact stimulus-evoked, goal-insensitive habits (Dickinson, Reference Dickinson1985). This model suggests that compulsions are not a search for safety; they are a result of the need to realize a link that has developed between a stimulus or context and a given set of responses. In this view, the subjective experiences that accompany habits, such as ‘not just right experience’ (Coles et al. Reference Coles, Heimberg, Frost and Steketee2005; Ecker & Gönner, Reference Ecker and Gönner2008), or more complex obsessive thoughts are the result of the compulsive urge, not the cause. For example, there is preliminary evidence to suggest that obsessive thoughts can even arise as a result of compulsive habit formation in OCD, and not the other way around (Gillan et al. Reference Gillan, Morein-Zamir, Urcelay, Sule, Voon, Apergis-Schoute, Fineberg, Sahakian and Robbins2014b ; Gillan & Sahakian, Reference Gillan and Sahakian2015), constituting a sharp divergence from most cognitive models of OCD that focus on obsessive thoughts (Salkovskis, Reference Salkovskis1985). The original habit hypothesis of OCD was based on a convergence of neurobiological data illustrating that the pathophysiology of OCD overlaps extensively with that supporting the balance between goal-directed behaviour and habits (Graybiel & Rauch, Reference Graybiel and Rauch2000; Dolan & Dayan, Reference Dolan and Dayan2013). Empirical studies have found broad support for this idea; OCD patients show a reliable tendency to form habits excessively in both appetitive and aversive learning contexts (Gillan et al. Reference Gillan, Papmeyer, Morein-Zamir, Sahakian, Fineberg, Robbins and de Wit2011, Reference Gillan, Morein-Zamir, Urcelay, Sule, Voon, Apergis-Schoute, Fineberg, Sahakian and Robbins2014b ). Based on convergent evidence from follow-up investigations utilizing neuroimaging and computational modelling (Gillan et al. Reference Gillan, Morein-Zamir, Kaser, Fineberg, Sule, Sahakian, Cardinal and Robbins2014a , Reference Gillan, Apergis-Schoute, Morein-Zamir, Urcelay, Sule, Fineberg, Sahakian and Robbins2015a ; Voon et al. Reference Voon, Derbyshire, Rück, Irvine, Worbe, Enander, Schreiber, Gillan, Fineberg, Sahakian, Robbins, Harrison, Wood, Daw, Dayan, Grant and Bullmore2014), the working model theorizes that deficits in goal-directed control, mediated by the caudate and medial orbitofrontal cortex, are responsible for the excessive habits observed in OCD in the laboratory (Gillan & Robbins, Reference Gillan and Robbins2014). Clinically, these habits might manifest as compulsive behaviours and/or higher-order habitual needs or goals (Cushman & Morris, Reference Cushman and Morris2015).

Addressing the trans-diagnostic potential of this neurocognitive mechanism, goal-directed control has been suggested to play a role in a range of disorders that are clinically characterized by repetitive behaviours that persist despite negative events, or are experienced as ‘out of control’ (Gillan et al. Reference Gillan, Robbins, Sahakian, van den Heuvel and van Wingen2016a ). These disorders include addiction (Everitt & Robbins, Reference Everitt and Robbins2005), eating disorders (Godier & Park, Reference Godier and Park2014) and Tourette's syndrome (Groenewegen et al. Reference Groenewegen, van den Heuvel, Cath, Voorn and Veltman2003), and although this area of research is still in its infancy, the first empirical investigations have found evidence for goal-directed deficits in these disorders (Sjoerds et al. Reference Sjoerds, de Wit, van den Brink, Robbins, Beekman, Penninx and Veltman2013; Voon et al. Reference Voon, Derbyshire, Rück, Irvine, Worbe, Enander, Schreiber, Gillan, Fineberg, Sahakian, Robbins, Harrison, Wood, Daw, Dayan, Grant and Bullmore2014; Delorme et al. Reference Delorme, Salvador, Palminteri, de Wit, Roze, Hartmann and Worbe2015; Ersche et al. Reference Ersche, Gillan, Jones, Williams, Luitjen, de Wit, Ward, Sahakian, Bullmore and Robbins2016) (Table 2). As such, the candidacy of goal-directed deficits as a trans-diagnostic marker has been acknowledged in the RDoC matrix, which flags several relevant theoretical domains, i.e. ‘Habit’, ‘Goal Selection, Updating, Representation and Maintenance’ and ‘Response Selection, Inhibition or Suppression’. However, studies have recently been published showing similar deficits in goal-directed control in schizophrenia (Morris et al. Reference Morris, Quail, Griffiths, Green and Balleine2015), autism spectrum disorder (Alvares et al. Reference Alvares, Balleine, Whittle and Guastella2016), and most problematically, social anxiety disorder (Alvares et al. Reference Alvares, Balleine and Guastella2014, Reference Alvares, Balleine, Whittle and Guastella2016), which is not considered clinically to be a ‘compulsive’ disorder (Table 2).

Table 2. Summary of results for studies examining goal-directed learning

n.s., Non-significant.

Goal-directed learning was measured using devaluation, model-based (MB) learning or action-outcome (A-O) learning test. The latter two measures are proxies for devaluation sensitivity (Gillan et al. Reference Gillan, Papmeyer, Morein-Zamir, Sahakian, Fineberg, Robbins and de Wit2011, Reference Gillan, Otto, Phelps and Daw2015b ).

^a Unmedicated were free of selective serotonin re-uptake inhibitors or antipsychotics for at least 2 weeks, and free of stimulants for 17 h. In the case of addiction, unmedicated applies when subjects meet the above criteria and are abstinent.

^b Compared multiple diagnostic groups.

^c Compared medicated and unmedicated groups, and no medication effect was observed.

^d Paediatric sample.

Using a case-control design, it is difficult to disentangle a genuine lack of mechanistic specificity, as these data might indicate, from ‘DSM measurement error’, the confounding influence of co-morbid psychiatric disorders. For example, in the aforementioned studies in social anxiety disorder patients (Alvares et al. Reference Alvares, Balleine and Guastella2014, Reference Alvares, Balleine, Whittle and Guastella2016), because patients presented with multiple co-morbid disorders, one cannot know if the results are attributable to social anxiety symptoms or one or more of their co-morbid conditions. Indeed, like most disorders, OCD rates are higher in the social anxiety population (Grant et al. Reference Grant, Hasin, Blanco, Stinson, Chou, Goldstein, Dawson, Smith, Saha and Huang2005). To get around this issue, some researchers (ourselves included) have endeavoured to recruit only ‘pure cases’ of OCD, i.e. those that do not meet the criteria for any other psychiatric diagnoses (Table 2). Although somewhat effective, the problem with this approach is that it assumes that being one criterion short of diagnosis of depression, for example, is the same as having no depressive symptoms at all. In reality, OCD patients recruited in this way consistently have higher levels of sub-threshold symptoms of multiple other disorders, e.g. depression and anxiety (Gillan et al. Reference Gillan, Papmeyer, Morein-Zamir, Sahakian, Fineberg, Robbins and de Wit2011). While these confounding variables could be controlled for statistically, most patient studies simply do not have the necessary power to do so. Indeed, the difficulty in recruiting sufficient sample sizes for research studies with patients is compounded by the increasing number of (possibly confounding) disorders defined by the DSM in every new edition.

A novel solution to this problem was presented in a recent study that eschewed the traditional case-control design in favour of the greater sample sizes (and in turn statistical power) that can be achieved by leveraging normal variability in a large general population sample (Gillan et al. Reference Gillan, Kosinski, Whelan, Phelps and Daw2016b ). The study tested if the relationship between goal-directed control and psychopathology constitutes a trans-diagnostic trait by testing its generalizability and specificity. Just under 2000 subjects were recruited online and completed self-report questionnaires assessing multiple psychiatric symptoms and completed a web-based task that assessed goal-directed performance (i.e. ‘model-based learning’; Daw et al. Reference Daw, Gershman, Seymour, Dayan and Dolan2011). In line with the preceding case-control literature, normal variation in self-reported OCD symptomatology (Foa et al. Reference Foa, Huppert, Leiberg, Langner, Kichic, Hajcak and Salkovskis2002) was associated with decreases in goal-directed control in two separate samples collected in this study (also see a recent independent replication: Snorrason et al. Reference Snorrason, Lee, de Wit and Woods2016). Importantly, this effect was generalizable to symptoms of other aspects of compulsive psychopathology (as previously characterized in the literature: Everitt & Robbins, Reference Everitt and Robbins2005; Hogarth et al. Reference Hogarth, Chase and Baess2012; Godier & Park, Reference Godier and Park2014). Specifically, eating disorders (Garner et al. Reference Garner, Olmsted, Bohr and Garfinkel1982), impulsivity (Patton et al. Reference Patton, Stanford and Barratt1995) and alcohol addiction (Saunders et al. Reference Saunders, Aasland, Babor, de la Fuente and Grant1993) were also associated with deficits in goal-directed control, suggestive of a trans-diagnostic deficit linking these disorders. Crucially, these deficits were not associated with symptoms of trait anxiety (Spielberger et al. Reference Spielberger, Gorsuch, Lushene, Vagg and Jacobs1983), depression (Zung, Reference Zung1965), social anxiety (Liebowitz, Reference Liebowitz1987) or apathy (Marin et al. Reference Marin, Biedrzycki and Firinciogullari1991). Schizotypy (Mason et al. Reference Mason, Linney and Claridge2005) sat somewhere in the middle, showing a marginal association with goal-directed deficits (Fig. 2 a).

Fig. 2. Validating a trans-diagnostic dimension. (a) The strength of the association between self-report symptoms of various DSM disorders and deficits in goal-directed control. The pattern is strikingly non-specific. (b) The association between goal-directed deficits and three ‘trans-diagnostic symptom dimensions identified in a data-driven factor analysis. Factor 1 corresponds to ‘Anxious Depression’, Factor 2 is ‘Compulsive Behavior and Intrusive Thought’, and Factor 3 is ‘Social Withdrawal’. The association between Factor 2 (‘Compulsive Behavior and Intrusive Thought’) and deficits in goal-directed learning is greater than that of any of the nine DSM-inspired questionnaires and crucially, the relationship exhibits excellent specificity with respect to ‘non-compulsive’ aspects of psychopathology, i.e. Factor 1 (‘Anxious Depression’) and Factor 3 (‘Social Withdrawal’). Data reproduced with permission from Gillan et al. (Reference Gillan, Kosinski, Whelan, Phelps and Daw2016b ). *p < 0.05, **p < 0.01, ***p < 0.001.

Although these results are suggestive of generalizability and specificity, this kind of analysis does not take us much further than work in diagnosed patients, save to say that these effects appear to be continuous in the general population rather than being specific to diagnosed groups. A more important question is what the psychiatric phenotype that explains this pattern of results? Is there a trans-diagnostic symptom dimension shared by some disorders (e.g. OCD, addiction, eating disorders), but not others (e.g. social anxiety, depression)? To answer this question empirically, this study employed a two-stage validation methodology. First, a factor analysis was carried out on psychiatric self-report data to identify a smaller set of clinical phenotypes that more parsimoniously explain the self-report symptom data. This approach revealed that three factors could replace the nine original questionnaires and of particular interest was a factor linking disorders at the level of compulsive behaviour and thought, so-called ‘Compulsive Behaviour and Intrusive Thought’. This factor comprised select features of OCD, eating disorders, addiction and some aspects of impulsivity and schizotypy that pertain to a loss of control over repetitive behaviour and thought. Next, this phenotype was tested for biological validity by assessing if subjects’ scores on this factor predicted goal-directed performance in an independent analysis. ‘Compulsive Behaviour and Intrusive Thought’ was found to be more predictive of goal-directed deficits than any DSM-inspired questionnaire collected in this study (Fig. 2 b), illustrating the power of data-driven clinical phenotyping. For example, this trans-diagnostic symptom dimension predicted task performance better than total scores on questionnaires measuring severity of DSM disorders such as OCD, eating disorders and addiction. Crucially, this relationship was dissociable from other trans-diagnostic phenotypes identified in this study, ‘Anxious-Depression’ and ‘Social Withdrawal’ (which showed no relation to goal-directed deficits), which could be compared directly in this study, providing compelling evidence for the specificity of this deficit to ‘Compulsive Behaviour and Intrusive Thought’. This approach showcases how in an appropriately powered general population sample, a neurocognitive model can be tested for both generalizability and specificity, leading to a data-driven definition of a neurobiologically validated clinical phenotype.

Going deeper: ‘units of analysis’

In tandem with new approaches to empirically define psychiatric phenotypes, the task for basic research is to delineate the neurobiology that underpins them. RDoC proposes that this should be done, where appropriate, using units of analysis that link clinical phenomenology (e.g. self-reports) to behaviour/physiology (via paradigms) to the underlying circuit, cells, molecules and finally, genes. The aims of this approach are to facilitate evidence-based drug-development (e.g. based on pharmacological underpinnings), to guide new therapeutic approaches such as psychological therapies (e.g. based on neurocognitive changes) or brain stimulation (e.g. based on knowledge of brain circuits), and to identify and treat those at risk (e.g. using genetic tests) (Fig. 3).

Fig. 3. A trans-diagnostic approach to obsessive-compulsive disorder (OCD). The predominant symptoms of OCD may result from complex interaction between independent trans-diagnostic dimensions, of which anxiety and compulsivity are good (but not the only) candidates. In this simplified schematic, we show just two putative trans-diagnostic psychiatric dimensions, there are of course many others. The task for basic research, outlined in the Research Domain Criteria initiative is to understand these dimensions at units of analysis, linking self-report symptoms (such as anxiety and compulsivity) to paradigms, physiology, circuits, cells, molecules and genes. GAD, Generalized anxiety disorder.

Here, we will highlight just a couple of examples drawing on units of analysis that have the most direct implications for treatment. Might the ERN be useful for deciding on which treatment to prescribe to which individual? To begin to answer this, we must review what is known about the pharmacology of the ERN (i.e. as a molecular unit of analysis). Benzodiazepines, specifically alprazolam (Riba et al. Reference Riba, Rodríguez-Fornells, Münte and Barbanoj2005), lorazepam (de Bruijn et al. Reference de Bruijn, Hulstijn, Verkes, Ruigt and Sabbe2004) and oxazepam (Johannes et al. Reference Johannes, Wieringa, Nager, Dengler and Münte2001a ), reduce the amplitude of the ERN. This suggests a potential role for GABA (which is thought to directly inhibit ACC functioning) in the ERN and with that, a putative biological basis for the therapeutic effect of these drugs for GAD and therefore plausibly also for anxiety as a trans-diagnostic trait (Ravindran & Stein, Reference Ravindran and Stein2010). Unfortunately, this theory runs into issues when one considers that benzodiazepines are ineffective for treating OCD (e.g. clonazepam, Hollander et al. Reference Hollander, Kaplan and Stahl2003). Moreover, antidepressants, such as paroxetine and mirtazapine, which appear to have no effect on the ERN (de Bruijn et al. Reference de Bruijn, Hulstijn, Verkes, Ruigt and Sabbe2004, Reference de Bruijn, Sabbe, Hulstijn, Ruigt and Verkes2006), have good efficacy for treating both OCD and GAD. This apparent lack of pharmacological specificity is compounded by research on dopamine, which is probably been the most studied neurotransmitter with respect to the ERN. Studies in this area have shown that amphetamine administration increases the amplitude of the ERN (de Bruijn et al. Reference de Bruijn, Hulstijn, Verkes, Ruigt and Sabbe2004), while antipsychotics such as haloperidol and olanzapine reduce it (Zirnheld et al. Reference Zirnheld, Carroll, Kieffaber, O'Donnell, Shekhar and Hetrick2004; de Bruijn et al. Reference de Bruijn, Sabbe, Hulstijn, Ruigt and Verkes2006). Although antipsychotics can be useful for OCD in particularly treatment resistant cases, GAD does not respond to antipsychotics (Ravindran & Stein, Reference Ravindran and Stein2010).

Thus, the question of how an ERN measurement can help inform treatment response across disorders is far from clear. It may be the case that combined measures, such as ERN, structural scans and self-report, are needed to produce a useful therapeutic signal. For example, although the core symptoms of OCD do not typically reduce following benzodiazepines, this drug class might be useful for decreasing anxiety in certain OCD patients – and these particularly anxious patients might be driving the ERN findings from studies examining this patient population at the group level. Alternatively, the ERN might have no specific causal role for psychiatry whatsoever, it might arise from a range of discrete upsets to neural harmony – longitudinal designs are needed to answer these questions definitively.

Turning to circuits, cells and the link between self-report compulsivity and deficits in goal-directed control (Gillan et al. Reference Gillan, Kosinski, Whelan, Phelps and Daw2016b ), studies point to the involvement of the medial orbitofrontal cortex (mOFC) and caudate based on their involvement in goal-directed control across species (Dolan & Dayan, Reference Dolan and Dayan2013) and convergent data suggesting these are critically involved in OCD and other compulsive disorders. For example, hyperactivity in the caudate nucleus is linked to failures in goal-directed behaviour observed in OCD patients (Gillan et al. Reference Gillan, Apergis-Schoute, Morein-Zamir, Urcelay, Sule, Fineberg, Sahakian and Robbins2015a ). In terms of brain structure, deficits in goal-directed control are associated with reductions of grey-matter volume in the caudate and mOFC in healthy individuals (Voon et al. Reference Voon, Derbyshire, Rück, Irvine, Worbe, Enander, Schreiber, Gillan, Fineberg, Sahakian, Robbins, Harrison, Wood, Daw, Dayan, Grant and Bullmore2014). Moreover, in this same study, the observation that these regional volumes are reduced in binge-eating disorder patients was statistically explained by their deficits of goal-directed control relative to controls (Voon et al. Reference Voon, Derbyshire, Rück, Irvine, Worbe, Enander, Schreiber, Gillan, Fineberg, Sahakian, Robbins, Harrison, Wood, Daw, Dayan, Grant and Bullmore2014). Although imaging studies in humans are important for validating animal models of psychiatric populations, animal models are crucial as they can uniquely assess causality (Ahmari, Reference Ahmari2015). To this end, one exemplary study showed that chronic stimulation of mOFC neurons that project to the ventromedial striatum [i.e. mimicking the hyperactivity observed in human OCD patients (Gillan et al. Reference Gillan, Apergis-Schoute, Morein-Zamir, Urcelay, Sule, Fineberg, Sahakian and Robbins2015a )] induces compulsive grooming behaviour in mice (Ahmari et al. Reference Ahmari, Spellman, Douglass, Kheirbek, Simpson, Deisseroth, Gordon and Hen2013). Furthermore, the authors found that excessive grooming was ameliorated with chronic SSRI treatment, the first line pharmacotherapy for OCD (Fineberg et al. Reference Fineberg, Reghunandanan, Brown and Pampaloni2013b ), thereby showing convergence across multiple units of analysis, from behaviour to brain circuits to molecules. What remains to be seen is whether this kind of chronic stimulation also induces deficits in goal-directed control in tandem with compulsive grooming, and if deficits in goal-directed control are remediated by treatment with SSRIs. These are testable hypotheses that, if supported, could mean that computerized tasks that assess this neurocognitive mechanism might be used to inform treatment allocation in the future. In addition to guiding pharmacotherapy decisions, work at the circuit level can be used to select target sites for deep-brain stimulation, of which there are many candidates for OCD (Mallet et al. Reference Mallet, Polosan, Jaafari, Baup, Welter, Fontaine, du Montcel, Yelnik, Chéreau, Arbus, Raoul, Aouizerate, Damier, Chabardès, Czernecki, Ardouin, Krebs, Bardinet, Chaynes, Burbaud, Cornu, Derost, Bougerol, Bataille, Mattei, Dormont, Devaux, Vérin, Houeto, Pollak, Benabid, Agid, Krack, Millet, Pelissolo and Group2008; Nuttin et al. Reference Nuttin, Gabriëls, Cosyns, Meyerson, Andréewitch, Sunaert, Maes, Dupont, Gybels, Gielen and Demeulemeester2008; Denys et al. Reference Denys, Mantione, Figee, van den Munckhof, Koerselman, Westenberg, Bosch and Schuurman2010; Greenberg et al. Reference Greenberg, Gabriels, Malone, Rezai, Friehs, Okun, Shapira, Foote, Cosyns, Kubu, Malloy, Salloway, Giftakis, Rise, Machado, Baker, Stypulkowski, Goodman, Rasmussen and Nuttin2010) and indeed to guide the allocation of distinct forms of behavioural therapy (Saxena et al. Reference Saxena, Gorbis, O'Neill, Baker, Mandelkern, Maidment, Chang, Salamon, Brody, Schwartz and London2009). The question of whether deficits in goal-directed control diminish with successful treatment (and are akin to state phenomena) or remain elevated and stable (trait) awaits investigation. But even more interesting in our minds is the question of whether or not goal-directed performance at baseline predicts better response to one treatment over another. This may be how the success of the marker should be ultimately judged.

Precision therapeutics: insights and trans-diagnostic opportunities

Although defining biologically valid clinical phenotypes as accurately and discretely as possible is valuable, it may not be the most important thing that trans-diagnostic psychiatry research can contribute. Even if a phenotype such as ‘compulsivity’ possesses good neurobiological validity, specificity and homogeneity, this may not translate into homogeneity of treatment response. Rather, discrete neural pathophysiology may give rise to common network-level dysfunctions and behavioural manifestations, and these distal phenotypic markers might therefore offer minimal insight for clinicians in deciding on whether or not an SSRI will be effective in an individual. One way to address this is to work from the bottom-up – to take a step back from phenotypic classification and focus instead on establishing robust and direct links between neurocognitive markers and treatment response. This kind of work has already begun, typically within the confines of one diagnostic category or another and we will briefly mention some of the most promising data. We will take a particular focus on pharmacological interventions, but the general approach can similarly be applied to behavioural interventions. Then we will highlight how treatment prediction research involving sufferers of many DSM categories that are prescribed the same treatment is likely to be an important approach of the future.

Taking neural markers as an example, several studies have had early success at using functional imaging at the group level to predict treatment response within the OCD population (for review see Ball et al. Reference Ball, Stein and Paulus2014). One potential marker of SRI response in OCD is reduced baseline activity in the OFC and ACC (Swedo et al. Reference Swedo, Pietrini, Leonard, Schapiro, Rettew, Goldberger, Rapoport, Rapoport and Grady1992, Reference Swedo, Schapiro, Grady, Cheslow, Leonard, Kumar, Friedland, Rapoport and Rapoport1989; Brody et al. Reference Brody, Saxena, Schwartz, Stoessel, Maidment, Phelps and Baxter1998; Saxena et al. Reference Saxena, Brody, Maidment, Dunkin, Colgan, Alborzian, Phelps and Baxter1999). Unfortunately several studies that came later failed to replicate (Saxena et al. Reference Saxena, Brody, Ho, Zohrabi, Maidment and Baxter2003; Carey et al. Reference Carey, Warwick, Niehaus, van der Linden, van Heerden, Harvey, Seedat and Stein2004; Ho Pian et al. Reference Ho Pian, van Megen, Ramsey, Mandl, van Rijk, Wynne and Westenberg2005; van der Wee et al. Reference van der Wee, Ramsey, van Megen, Denys, Westenberg and Kahn2007), finding no significant effects or in some cases entirely different neural markers. A recent study tested whether a task-related ACC measure like ERN might fare better than these resting studies at predicting clinical response. Although the result should be taken cautiously as the sample size was particularly small (only 10 patients returned at post-treatment assessment), the authors were unable to find a relationship between baseline ERN and cognitive behaviour therapy response in a paediatric OCD sample (Hajcak et al. Reference Hajcak, Franklin, Foa and Simons2008).

One possibility is that the predictive value of ACC activity (if there is any) could be exclusive to a putative anxiety dimension and this might get washed out in studies relying on heterogeneous diagnostic categories and small samples. Recruiting large samples across diagnostic categories and recording changes in the symptoms of dissociable psychiatric dimensions like anxiety and compulsivity might help isolate reliable effects. Indeed, in the one study that adopted a trans-diagnostic approach, recruiting individuals with OCD, depression or both, the authors were able to identify dissociable neural markers of subsequent depressive v. obsessive-compulsive response to paroxetine (Saxena et al. Reference Saxena, Brody, Ho, Zohrabi, Maidment and Baxter2003). This means that not only is the same drug effective in treating distinct psychiatric problems, its efficacy at treating depression and OCD symptoms (sometimes in the same individual) is predicted by dissociable patterns of abnormal baseline neural activity. Given that individuals with a diagnosis of OCD are aetiologically and phenotypically heterogeneous, the task for future research is not to identify mean baseline activity patterns in responders v. non-responders on some singular measure of function, but to use large datasets to tease apart the mechanisms through which a given drug can reduce symptoms of largely independent psychiatric dimensions, and use these insights to identify the patients for whom a symptom reduction is likely to happen following treatment.

To conclude this section with some practical statements, morphometric predictors of SRI response in OCD already appear to be somewhat more robust than functional ones. The only two structural MRI studies to date have both shown that decreased grey matter in the lateral OFC is also linked to SRI response (Hoexter et al. Reference Hoexter, Dougherty, Shavitt, D'Alcante, Duran, Lopes, Diniz, Batistuzzo, Evans, Bressan, Busatto and Miguel2013, Reference Hoexter, Diniz, Lopes, Batistuzzo, Shavitt, Dougherty, Duran, Bressan, Busatto, Miguel and Sato2015). Hoexter and colleagues showed that OFC grey-matter volume had 77% sensitivity and 81% specificity at predicting SRI response in OCD. If these figures hold up to replication using the same analysis pipeline, the interpretation is that if this marker were used to determine whether or not an SRI is prescribed to a drug-naive individual, just two out of every 10 people assigned to SSRI would not respond, while another two of the 10 not assigned the treatment would have missed out on the benefits of the SSRI (Hoexter et al. Reference Hoexter, Diniz, Lopes, Batistuzzo, Shavitt, Dougherty, Duran, Bressan, Busatto, Miguel and Sato2015). This would make it a useful tool for clinicians, were MRI scans not expensive and impractical for most clinics to use diagnostically (at least for now). This is where we suggest cognitive neuroscience might be able to intervene by replacing brain scans with cognitive tasks that are good indicators of neurobiology, such those linked to normal and pathological variation in OFC grey-matter volume, of which goal-directed control is one notable example (Voon et al. Reference Voon, Derbyshire, Rück, Irvine, Worbe, Enander, Schreiber, Gillan, Fineberg, Sahakian, Robbins, Harrison, Wood, Daw, Dayan, Grant and Bullmore2014). Cognitive behavioural therapy for OCD similarly might benefit from the addition of components that focus more directly on habit formation and goal-directed control in those patients for whom it is more relevant. For example, one might focus psychoeducation: patients could be taught about identifying habit ‘trigger’ situations, or identifying the development of new habits that might later develop into more elaborate compulsions. Alternatively, habit-reversal training (Snorrason et al. Reference Snorrason, Berlin and Lee2015) may be used to complement standard exposure and response prevention techniques. These behavioural treatments might be further enhanced with targeted pharmacotherapy, for which basic data are accruing (Lovinger, Reference Lovinger2010).

Combining trans-diagnostic with existing approaches

We have argued strongly in favour of the trans-diagnostic model as a rational approach to advance basic science research in psychiatry, but must make clear that to date this approach has yet to facilitate gene discovery or the development of new, better treatments. It will take time for sufficient data to accrue and as such, dismissing symptom-based approaches entirely is premature. Existing clinical trial datasets, based on patients grouped according to conventional diagnostic criteria such as DSM, have demonstrated a clear specificity of effect for certain psychiatric treatments [e.g. serotonin v. noradrenaline reuptake inhibitors in OCD (Fineberg et al. Reference Fineberg, Chamberlain, Goudriaan, Stein, Vanderschuren, Gillan, Shekar, Gorwood, Voon, Morein-Zamir, Denys, Sahakian, Moeller, Robbins and Potenza2014)], and a magnitude of efficacy for some of our treatments commensurate with the effect sizes seen in other branches of medicine (Davis et al. Reference Davis, Giakas, Qu, Prasad and Leucht2011; Leucht et al. Reference Leucht, Helfer, Gartlehner and Davis2015). We have also seen that specific drug response can be linked to easily observable symptoms, such as the presence of tics (Bloch et al. Reference Bloch, Landeros-Weisenberger, Kelmendi, Coric, Bracken and Leckman2006). The hope is that by revisiting how we define disorder and/or dimensions using biological and cognitive data will further enhance the precision with which treatments can be delivered to individuals. In addition, new ways of assessing the outcomes of treatment trials, e.g. using statistical techniques such as group-based mixture models, that divide ‘diagnostic’ patient samples into clinically meaningful subgroups with greater predictive value, hold promise for the advancement of personalized psychiatric treatment (Thase et al. Reference Thase, Larsen and Kennedy2011). Thus, most progress is likely to be made with a balanced approach, taking into account and if possible combining the most valid and reliable diagnostic and trans-diagnostic methodologies.