Borsboom et al. argue that reductionism in psychopathology research has not yielded sufficient insights to understand the complexity of the systems that cause psychopathologies, nor has it provided effective treatments. This somewhat pessimistic verdict may apply when the philosophically driven and publicly nourished expectations are a full understanding and full cure of the known psychiatric disorders (Müller Reference Müller2018). The leading framework for the last decades of neurobiological research in psychopathology was reductionism. Leading hypotheses were generated that suggested a single psychiatric disorder can be reduced, that is, causally explained and treated, to a dysfunction in a single target or single functional system of the brain. It can be argued that this may have been supported by the technical limitations of new empirical techniques. For instance, the view that a dysfunction in the dopaminergic (DA) system of the brain is the sole causal mediator for drug addiction development was massively driven by the advent of in vivo microdialysis (Westerink Reference Westerink1995). This technique proved that all drugs with addiction potential acutely enhance DA activity in the brain's reward circuitry, while non-addictive drugs do not (Di Chiara & Imperato Reference Di Chiara and Imperato1988). Thereby, a single prominent and many-times-replicated finding was generalized to a functional theory (Koob Reference Koob1992; McBride et al. Reference McBride, Murphy and Ikemoto1999; Wise Reference Wise2002) which did not pay tribute to the emerging complexity of the system (Salamone Reference Salamone1996). The therapeutic predictions based on this model, however, failed in practice (McCreary et al. Reference McCreary, Müller and Filip2015; Spanagel & Kiefer Reference Spanagel and Kiefer2008), suggesting that the DA theory of addiction is at least incomplete (Nutt et al. Reference Nutt, Lingford-Hughes, Erritzoe and Stokes2015). One possible reason for that could have been the initial limitation of a key technique, in that only DA was measured and only dopaminergic innervated brain structures were considered. Empirical techniques advanced and showed that many more transmitter systems of the brain are dysregulated during and after drug consumption (Heilig & Koob Reference Heilig and Koob2007; Müller & Huston Reference Müller and Huston2006; Schneider et al. Reference Schneider, Levant, Reichel, Gulbins, Kornhuber and Müller2017; Williams & Adinoff Reference Williams and Adinoff2008). What is required in this field is not to deny the massive gain in knowledge based on single-target empirical approaches (Koob & Volkow Reference Koob and Volkow2016), but rather, a constructive synthesis of those findings (Spanagel Reference Spanagel2009).

Borsboom and colleagues suggest a systems approach in which syndromes of distinct diagnostic categories interact. Only a systems understanding that also comprises proximal and distal environmental factors would allow for overcoming what they consider an epistemological failure. While this radical approach has the merit of considering syndrome-syndrome interactions (e.g., Schuckit et al. Reference Schuckit, Tipp, Bergman, Reich, Hesselbrock and Smith1997), as well as environmental factors in psychopathology (Caspi & Moffitt Reference Caspi and Moffitt2006), it falls short of a major criterion for a scientific approach: its testability. Here it is argued that a major driving force for a bio-reductionistic approach in psychopathology is actually its testability. Hypotheses can be falsified on empirical grounds, when single variables – biological and environmental – can be addressed. This is difficult, and probably impossible, with syndromes that naturally involve a plethora of single causal elements which cannot be systematically manipulated at once. Empirical network approaches simply purport testability by bioinformatic large-data analysis. Although they are not worthless, they are just correlative by nature. By that way, they may generate hypotheses; however, these require explicit testing (Easton et al. Reference Easton, Lucchesi, Lourdusamy, Lenz, Solati, Golub, Lewczuk, Fernandes, Desrivieres, Dawirs, Moll, Kornhuber, Frank, Hoffmann, Soyka, Kiefer, Schumann, Giese and Müller2013; Mielenz et al. Reference Mielenz, Reichel, Jia, Quinlan, Stöckl, Mettang, Zilske, Kirmizi-Alsan, Schönberger, Praetner, Huber, Amato, Schwarz, Purohit, Brachs, Spranger, Hess, Büttner, Ekici, Perez-Branguli, Winner, Rauschenberger, Banaschewski, Bokde, Büchel, Conrod, Desrivieres, Flor, Frouin, Gallinat, Garavan, Gowland, Heinz, Martinot, Lemaitre, Nees, Paus, Smolka, Schambony, Bäuerle, Eulenburg, Alzheimer, Lourdusamy, Schumann and Müller2018; Schumann et al. Reference Schumann, Liu, O'Reilly, Gao, Song, Xu, Ruggeri, Amin, Jia, Preis, Segura, Akira, Barbieri, Baumeister, Cauchi, Clarke, Enroth, Fischer, Hallfors, Harris, Hieber, Hofer, Hottenga, Johansson, Joshi, Kaartinen, Laitinen, Lemaitre, Loukola, Luan, Lyytikainen, Mangino, Manichaikul, Mbarek, Milaneschi, Moayyeri, Mukamal, Nelson, Nettleton, Partinen, Rawal, Robino, Rose, Sala, Satoh, Schmidt, Schraut, Scott, Smith, Starr, Teumer, Trompet, Uitterlinden, Venturini, Vergnaud, Verweij, Vitart, Vuckovic, Wedenoja, Yengo, Yu, Zhang, Zhao, Boomsma, Chambers, Chasman, Daniela, de G., Deary, Eriksson, Esko, Eulenburg, Franco, Froguel, Gieger, Grabe, Gudnason, Gyllensten, Harris, Hartikainen, Heath, Hocking, Hofman, Huth, Jarvelin, Jukema, Kaprio, Kooner, Kutalik, Lahti, Langenberg, Lehtimaki, Liu, Madden, Martin, Morrison, Penninx, Pirastu, Psaty, Raitakari, Ridker, Rose, Rotter, Samani, Schmidt, Spector, Stott, Strachan, Tzoulaki, van der Harst, van Duijn, Marques-Vidal, Vollenweider, Wareham, Whitfield, Wilson, Wolffenbuttel, Bakalkin, Evangelou, Liu, Rice, Desrivieres, Kliewer, Mangelsdorf, Müller, Levy and Elliott2016). Borsboom et al. need to be asked for this crucial but elegantly avoided point in their theory: How do they want to test causal factors at the systems (network) level without reducing it to measurable units? So far, the examples they provide are just disguised reductionistic approaches.

In order to rescue the innovative momentum of the authors’ proposal, a combination of both approaches may be considered: a reductionistic empirical frame which allows empirical testability, interwoven with a constructivistic systems approach which brings single elements together to a higher-order network of syndromes: a constructive reductionism. Instances for that can already be found, for example, in drug addiction research. A systems approach proposed the interaction of disorders, like depression or schizophrenia causing the abuse and addiction to various psychoactive drugs (Müller & Schumann Reference Müller and Schumann2011). The reductionistic testing of mechanisms has recently yielded first testing of elements of a depression-alcoholism (Gulbins et al. Reference Gulbins, Palmada, Reichel, Luth, Bohmer, Amato, Müller, Tischbirek, Groemer, Tabatabai, Becker, Tripal, Staedtler, Ackermann, van Brederode, Alzheimer, Weller, Lang, Kleuser, Grassme and Kornhuber2013; Müller et al. Reference Müller, Kalinichenko, Tiesel, Witt, Stöckl, Sprenger, Fuchser, Beckmann, Praetner, Huber, Amato, Mühle, Büttner, Ekici, Smaga, Pomierny-Chamiolo, Pomierny, Filip, Eulenburg, Gulbins, Lourdusamy, Reichel and Kornhuber2017) and schizophrenia-nicotine addiction syndrome interaction (Koukouli et al. Reference Koukouli, Rooy, Tziotis, Sailor, O'Neill, Levenga, Witte, Nilges, Changeux, Hoeffer, Stitzel, Gutkin, DiGregorio and Maskos2017). A constructivistic approach is now picking that up to a systems view that includes various syndrome-syndrome interactions (Müller & Kornhuber Reference Müller and Kornhuber2017; Schneider et al. Reference Schneider, Levant, Reichel, Gulbins, Kornhuber and Müller2017). Given the increasingly recognized complexity of the biological base of psychiatric disorder syndromes (Schumann et al. Reference Schumann, Binder, Holte, de Kloet, Oedegaard, Robbins, Walker-Tilley, Bitter, Brown, Buitelaar, Ciccocioppo, Cools, Escera, Fleischhacker, Flor, Frith, Heinz, Johnsen, Kirschbaum, Klingberg, Lesch, Lewis, Maier, Mann, Martinot, Meyer-Lindenberg, Müller, Müller, Nutt, Persico, Perugi, Pessiglione, Preuss, Roiser, Rossini, Rybakowski, Sandi, Stephan, Undurraga, Vieta, van der Wee, Wykes, Haro and Wittchen2014), we should not expect a full-blown constructive synthesis happening in a single step, that is, by just interlinking single reductionistic findings.

Reductionism is, in fact, a multi-level reductionism, where functional networks are step-wise reduced to functional elements and sub-elements. Mirroring this, a multi-level constructivism should also be acknowledged. At the lowest level, reductionistic findings with the highest possible technical resolution are brought together to a nano-systems view (e.g., Kreek et al. Reference Kreek, Nielsen, Butelman and Laforge2005; Ungless et al. Reference Ungless, Argilli and Bonci2010). At the next constructive level, nano-systems views are brought together to constitute micro-systems views (e.g., Hyman et al. Reference Hyman, Malenka and Nestler2006; Robinson & Kolb Reference Robinson and Kolb2004). This can be continued up to the level of the whole brain (Müller & Homberg Reference Müller and Homberg2015; Spanagel Reference Spanagel2009) and syndrome behaviours (Schumann et al. Reference Schumann, Binder, Holte, de Kloet, Oedegaard, Robbins, Walker-Tilley, Bitter, Brown, Buitelaar, Ciccocioppo, Cools, Escera, Fleischhacker, Flor, Frith, Heinz, Johnsen, Kirschbaum, Klingberg, Lesch, Lewis, Maier, Mann, Martinot, Meyer-Lindenberg, Müller, Müller, Nutt, Persico, Perugi, Pessiglione, Preuss, Roiser, Rossini, Rybakowski, Sandi, Stephan, Undurraga, Vieta, van der Wee, Wykes, Haro and Wittchen2014). This is not the end, however. An extended endophenotype of psychiatric disorders should comprise the whole body periphery as a proximal environment, and the social as well as physical environments of the organisms as distal environments (Badiani Reference Badiani2013; Zinberg Reference Zinberg1984). The constructivistic synthesis may then end at a level where also the time component (i.e., personal, social and historical developments) is incorporated in this systematic construction. However, there will always be an error term in the reductionistic as well as constructivistic part of the approach. Simple chance events in biological systems, for example, de novo mutations (Michaelson et al. Reference Michaelson, Shi, Gujral, Zheng, Malhotra, Jin, Jian, Liu, Greer, Bhandari, Wu, Corominas, Peoples, Koren, Gore, Kang, Lin, Estabillo, Gadomski, Singh, Zhang, Akshoomoff, Corsello, McCarroll, Iakoucheva, Li, Wang and Sebat2012), may act as single-case determinants. This cannot be overcome and controlled by either reductionism or constructivism, nor by their combination, and we may have to live with a certain degree of uncertainty.