Hallucinating mice, psychotherapy in the brain, and computational psychiatry: the future of psychiatry is now

Stein et al1 provide a philosophically grounded overview of the conceptual foundations of psychiatry, from its basis in psychopathology to its applications in clinical practice. They emphasize: a) the importance of facts and values in health care; b) the necessity of taking a pluralist approach to science, regarding ontology, explanation and values; and c) the usefulness of "embodied cognition" as an integrative concept in theory and in clinical practice. I have worked as a clinical psychiatrist for over 40 years, with a research focus on systems neuroscience for over the last 30 years, and applaud their thoughts on the foundations of the field. In what follows, I just want to add a few considerations from my personal and scientific perspective. I endorse Stein et al's points a) and c) wholeheartedly. Whoever maintains that medicine is science, and science is (by definition) only about facts, must be reminded that the terms "pain", "suffering", "illness" and "disease" imply negative valuation, i.e., cannot be understood by anybody who does not grasp what statements like "I do not like x", or "I am so happy because of y" mean. Together with the spatial and temporal finiteness of our minds as embodied minds, joy and pain give us agency and purpose, and let us – as the most social primates that evolution brought about – strive for the betterment of our lives. In case of success, psychiatrists are rarely needed; failure, in contrast, calls upon us. As regards the point b), I want to propose a few considerations. Instead of a pluralist approach, I would like to speak of a multimodal approach with respect to science (ontology and explanation), and of an open and tolerant approach when it comes to cultural differences. Let me explain. There is no British, Chinese, French, San or Papuan science; there is only science. At its empirical connections with reality, science has always been changing, and battles between different schools of thought have been a big part of its history. However, science is remarkably stable at its inner theoretical core. Set theory; Newtonian, Einsteinian and quantum physics; the periodic system, evolution, genetics, etc. have incrementally grown and produced a huge body of knowledge. It is hard to imagine which "ugly fact" would have the epistemic force/weight to contradict any of the approaches just mentioned. Sure, there have always been (and there will always be) statements which are questioned by scientific progress. But, the closer we get from science's empirical connections with reality to its inner core, the smaller the probability of doubt. In what follows, I want to exemplify a multimodal approach to psychopathology and psychiatry by briefly touching upon three recent domains of research: a) mouse models of various forms of psychopathology that allow to establish causality between brain states and states of mind; b) the concepts of neuromodulation (by small molecules, such as dopamine, acetylcholine, norepinephrine, serotonin and many others) and neuroplasticity (from the synaptic to the cortical level); and c) computational psychiatry as part of artificial intelligence research. For the larger part of my career, I could not have cared less about mouse models in medicine, especially in psychiatry. They worked for testing new drugs (even though most of our arsenal was discovered by serendipity) and provided some quite limited insights into their mechanisms of action. However, for understanding the core symptoms of psychosis (hallucinations, delusions and formal thought disorder), as well as for providing insights into the interactions between genes and complex human environment, mouse models appeared useless. This has changed completely within the past five years. I will provide some examples. Eye movement desensitization and reprocessing (EMDR) is a form of psychotherapy used to treat post-traumatic stress disorder (PTSD) with alternating bilateral sensory stimulation (ABS). In order to study its mechanism of action, a mouse model was implemented that allowed the induction of conditioned fear, its extinction, and monitoring the influence of ABS on extinction. In conjunction with ABS, extinction was more pronounced and lasted longer. Once this was established, a neural pathway (from the superior colliculus via the mediodorsal thalamus) which suppressed the activity of fear-encoding cells in the basolateral amygdala, mediating persistent attenuation of fear, could be worked out. "Together, these results reveal the neural circuit that underlies an effective strategy for sustainably attenuating traumatic memories"2. In order to study auditory hallucinations in a mouse model3, mice had to be trained to hallucinate, i.e., firstly, to perceive a salient stimulus (a tone) within random noise (false perception) and, secondly, take this perception more or less for real (reality check). Both responses could be reliably induced. By optogenetic and chemogenetic methods, a neural pathway was discovered, with elevated dopamine levels in the ventral tegmental area and the striatum, where distinct subregions encode different kinds of expectations. "These findings support the idea that hallucinations arise… due to elevated dopamine producing a bias in favor of prior expectations against current sensory evidence"3. The term "dissociation" denotes a broad spectrum of states involving perception, thought, consciousness, and the experience of time, space, reality and self. In a mouse model of the effects of drugs that reliably produce a dissociative state, a very specific rhythmic oscillation in layer 5 of the retrosplenial cortex was discovered to be necessary and sufficient for this state. Moreover, in a patient with epilepsy, a similar oscillation was found at a corresponding brain area during the dissociative aura4. This finding catapults an ill-defined, little understood and frequently used concept from the 19th into the 21st century. Within a developmental window in mouse brain development (days 2-9), adversity (murine analogs to human poverty, immigration, neglect and abuse) establishes a circuit which leads to behavioral dysfunction in adulthood (decreased reward sensitivity), very much akin to human PTSD5. It is important to realize that optogenetic models allow, for the first time, to establish causality – and not mere correlation – in the realm of mind and brain. Therefore, this method should be highly fruitful in the near future of psychopathology. The concepts of neuromodulation and neuroplasticity allow neuroscience-based parsimonious explanations of a wide range of mental phenomena, such as acute states of mind (induced by increased, decreased or dysregulated neuromodulatory agents) as well as long-term ("chronic") changes of mind caused by synaptic plasticity. In the mouse model, neuromodulatory changes (increased dopamine in the ventral tegmental area) drive auditory hallucinations, which can be readily treated with dopamine antagonists (in mice and acutely ill patients). Once these experiences become entrenched by synaptic plasticity, they are less dependent upon dopamine hyperactivity and therefore less treatable by antipsychotic agents8, 6. Finally, neural networks started to be used in computational models of mental phenomena in the late 1980s and 1990s by a few psychiatrists8, 6, 7. While we expected that computational psychiatry would take off eventually, none of us would have imagined that this new form of artificial intelligence would be applied to anything from deciphering cuneiform script to protein folding, drug design, weather forecasting, as well as designing next-generation hardware and software9. Artificial intelligence is not only going to change psychiatry, but to change the world, just as the invention of writing, the printing press, or the Internet.