Burning down the house: reinventing drug discovery in psychiatry for the development of targeted therapies

Abstract

Despite advances in neuroscience, limited progress has been made in developing new and better medications for psychiatric disorders. Available treatments in psychiatry rely on a few classes of drugs that have a broad spectrum of activity across disorders with limited understanding of mechanism of action. While the added value of more targeted therapies is apparent, a dearth of pathophysiologic mechanisms exists to support targeted treatments, and where mechanisms have been identified and drugs developed, results have been disappointing. Based on serendipity and early successes that led to the current drug armamentarium, a haunting legacy endures that new drugs should align with outdated and overinclusive diagnostic categories, consistent with the idea that “one size fits all”. This legacy has fostered clinical trial designs focused on heterogenous populations of patients with a single diagnosis and non-specific outcome variables. Disturbingly, this approach likely contributed to missed opportunities for drugs targeting the hypothalamic-pituitary-adrenal axis and now inflammation. Indeed, cause-and-effect data support the role of inflammatory processes in neurotransmitter alterations that disrupt specific neurocircuits and related behaviors. This pathway to pathology occurs across disorders and warrants clinical trial designs that enrich for patients with increased inflammation and use primary outcome variables associated with specific effects of inflammation on brain and behavior. Nevertheless, such trial designs have not been routinely employed, and results of anti-inflammatory treatments have been underwhelming. Thus, to accelerate development of targeted therapeutics including in the area of inflammation, regulatory agencies and the pharmaceutical industry must embrace treatments and trials focused on pathophysiologic pathways that impact specific symptom domains in subsets of patients, agnostic to diagnosis. Moreover, closer collaboration among basic and clinical investigators is needed to apply neuroscience knowledge to reveal disease mechanisms that drive psychiatric symptoms. Together, these efforts will support targeted treatments, ultimately leading to new and better therapeutics in psychiatry.

INTRODUCTION

Despite tremendous progress in basic neuroscience and the understanding of fundamental cellular, molecular, and circuit-based mechanisms of neurobiological processes and pathology, the field of psychiatry continues to struggle to develop new and better medications. Indeed, the pharmacological armamentarium for psychiatric disorders has not meaningfully changed for over 50 years [1, 2]. In most cases, psychiatric diseases are treated with some combination of drugs that fall into a limited number of categories, with limited specificity for the disorder being treated, and an unknown mechanism of action in humans beyond what is known about their basic pharmacological properties (Table 1). More concerning is that despite efforts to embrace a “precision medicine” approach to drug treatment and development, including breaking down disorders into symptom domains as reflected by Research Domain Criteria (RDoC) or grouping patients through agnostic clustering algorithms using a host of biomarkers and advanced computational modeling [3, 4], the field is bereft of a fundamental prerequisite for progress—an understanding of the pathophysiologic mechanisms that drive symptoms in patients with psychiatric disorders [1, 2, 5, 6]. Committing to a better grasp of such mechanisms will allow identification of specific subgroups of patients who are likely to respond to a specific set of therapeutics that can be tested and confirmed through target engagement and related behavioral outcomes. This strategy will provide the foundation for targeted treatments in psychiatry and ultimately new and better pharmacotherapies.

Table 1.

Nonspecificity of medication interventions in psychiatry.

Class of medication	Representative disorders treated	Pharmacology	Disease specific mechanism of action (demonstrated in humans)
Antipsychotics	Schizophrenia, Mood Disorders, Anxiety Disorders, PTSD, OCD, Psychotic Disorders, Autism, Personality Disorders [78–80]	Multiple MOA (e.g., D2 antagonism, 5HT receptor antagonism, ACh receptor antagonism) [81]	D2 antagonism for positive symptoms in schizophrenia; [82] otherwise unknown
Antidepressants (e.g., TCAs, SSRIs, SNRIs, MAOIs)	Mood Disorders, Anxiety Disorders, PTSD, OCD, Psychotic Disorders, Autism, ADHD, Personality Disorders [79, 83, 84]	Multiple MOA (e.g., 5HT, NE, DA reuptake blockade and 5HT receptor agonism and antagonism), increased neurogenesis [6, 85]	Unknown [6]
Mood Stabilizers	Mood Disorders, Schizophrenia, Anxiety Disorders, PTSD, OCD, Autism, Personality Disorders [79, 80, 86, 87]	Multiple MOA (e.g., modulation of GABA-ergic and glutamatergic signaling, modulation of voltage-gated ion channels), increase of BDNF, inhibition of GSK3 [86, 88]	Unknown
Stimulants	ADHD, Mood Disorders, Anxiety Disorders, PTSD, Autism [89–91]	Increased release and decreased reuptake of NE and DA [92]	Unknown
Ketamine	Mood Disorders, Anxiety Disorders, PTSD, OCD [93]	Multiple MOA (e.g., NMDA receptor antagonism, opioid receptor agonism), increase of BDNF [94, 95]	Unknown
Psychedelics (e.g., MDMA, psylocibin)	Mood Disorders, Anxiety Disorders, PTSD, OCD, Autism [96]	Multiple MOA (e.g., 5HT2A receptor agonism) [96]	Unknown
THC	Mood Disorders, Anxiety Disorders, PTSD, ADHD, Autism [97, 98]	Multiple MOA (e.g., CB1 receptor agonism) [99]	Unknown

5-HT 5-hydroxytryptophan, Ach acetylcholine, ADHD Attention-Deficit/Hyperactivity Disorder, CB cannabinoid, D2 dopamine receptor 2, DA dopamine, GABA gamma-aminobutyric acid, GSK glycogen synthase kinase, MAOI monoamine oxidase inhibitor, MDMA 3,4-methylenedioxy-methamphetamine, MOA mechanism of action, NE norepinephrine, PTSD post-traumatic stress disorder, OCD obsessive compulsive disorder, TCA tricyclic antidepressant, THC tetrahydrocannabinol, SNRI serotonin norepinephrine reuptake inhibitor, SSRI selective serotonin reuptake inhibitor.

Although the development of targeted treatments in psychiatry has obvious appeal, there are many countervailing forces. Given the pressure to develop novel therapeutics to address the plethora of treatment non-responsive patients or those with persisting symptoms, therapeutic interventions such as ketamine and psychedelic drugs whose mechanisms of action are even more elusive than those of conventional psychotropic medications have taken center stage (albeit with a passing nod to eventually reverse engineering the mechanisms involved). Moreover, regulatory agencies remain tied to outdated and overinclusive diagnostic entities whose heterogeneity thwarts efforts to develop therapeutic approaches that target specific pathologies and related behaviors, especially when they occur across diagnostic categories [7]. In addition, profit motives in industry may drive the development of drugs with modest, broad spectrum activity within and across disorders, sacrificing greater efficacy within patient subgroups in order to maximize market share. With these forces in mind, the question then becomes how can we do a better job of reinventing drug discovery in psychiatry to address these challenges.

Successes in the field of oncology may provide a blueprint for a way forward. Like psychiatry, oncology has been dominated for many years by non-specific, cytotoxic treatment regimens, notably chemotherapies. Similar to psychotropic medications, early chemotherapies were discovered by serendipity, based in part on the surprising therapeutic effects of nitrogen mustard (developed as mustard gas in World War II) and related alkylating compounds on disease activity in patients with lymphoma [8]. Moreover, as is the case with current psychiatric treatments, chemotherapy is effective but has a burdensome side effect profile not entirely unlike that seen with some psychotropic medications including notably antipsychotic drugs and their association with involuntary movement disorders and marked metabolic disturbances [9]. In fact, given the relative non-specificity and broad spectrum efficacy of both treatment regimens, one could consider current pharmacologic therapies in psychiatry as “chemotherapy for the brain”. It should be noted that chemotherapeutic regimens were developed as treatments for cancers defined in large part by their tissue-specific location (e.g., cancers of the lung, breast, prostate, blood, etc.) [10, 11]. Nevertheless, with the advent of a greater understanding of the cellular, molecular and more recently immunologic mechanisms that support cancer cell growth and proliferation, a revolution in oncology has occurred [10–12]. A whole new class of therapeutics that target these mechanisms has been developed. These targeted treatments including immunotherapies have exhibited remarkable efficacy, with responses and even cures beyond wildest expectations [10–13]. Importantly, targeted therapies are agnostic to cancer type and tissue location [10–12]. Indeed, the old boundaries that siloed various cancers have broken down, and clinicians can identify tumors relevant for a given treatment through molecular profiling, regardless of cancer type or location. This revolution has also extended to the strategy for clinical trials with the emergence of so called “basket trials”, in which patients with multiple sorts of cancers are grouped by their common pathophysiologic mechanisms/molecular profiling and not by their cancer type or location [10, 12]. Imagine a similar trial in psychiatry where patients with a common pathophysiologic mechanism (e.g., inflammation) leading to specific symptoms were included in a trial, irrespective of diagnosis. Such a trial would be a “non-starter” for psychiatric drug development because regulatory agencies do not accept applications for new psychiatric therapeutics unless they treat a disorder directly aligned with current diagnostic nomenclature [1, 3, 7]. The gatekeepers of drug development have reified diagnostic entities as opposed to pathophysiologic brain states and associated symptoms. Given that psychiatric disorders as currently conceived and defined are highly heterogeneous with multiple mechanisms for symptoms within and across disorders, the likelihood of developing targeted therapies in psychiatry under current conditions is low. The time for change is now.

In this Perspective, we will address how psychiatry has arrived at its current state along with missed opportunities from the past and lessons to be learned. Using inflammation as a candidate mechanistic pathway to pathology, we will also outline how missteps in translation are jeopardizing this field, repeating the mistakes of the past. Finally, we will provide recommendations for the development of targeted therapies in psychiatry including action items for researchers, funding agencies, industry, and regulatory bodies as well as clinical trial design.

HOW DID WE GET HERE?

The seduction of serendipity; the reification of diagnostic nomenclature and the notion that “one size fits all”

Early identification of pharmacologic therapies for psychiatric diseases was the result of serendipitous observations regarding the efficacy of drugs like thioridazine and impramine, which were found to be effective in psychotic and mood disorders, respectively [1, 2, 14]. Given the broad spectrum of efficacy of these compounds in patients with diagnoses of schizophrenia and depression, these early observations instantiated the existence of these disorders as ontologically-valid, monolithic entities and supported the notion that “one size fits all” in psychiatric drug development and response [1, 2, 15]. Moreover, the seduction of serendipity led to unrealistic expectations that better treatments for these disorders would be well within reach of the emerging field of “biological psychiatry” [2]. Unfortunately, as noted above, there has been little progress beyond these drugs, in part related to a disconnect between the significant advances in basic neuroscience and the limited understanding of the neurobiology of psychiatric diseases as they manifest in clinical settings [15, 16]. Indeed, second generation therapeutics have primarily addressed the side effect burden of first-generation medications without materially improving efficacy or specificity of mechanism [1, 2, 5]. In addition, the field continues to be burdened by the reification of psychiatric nosology and the notion of “one size fits all” that is reflected in the approach of regulatory agencies including the United States (US) Federal Drug Administration [7]. These agencies continue to pose hurdles that bespeak this legacy, eschewing the well-known heterogeneity of psychiatric disorders and the specificity of mechanisms leading to specific symptoms (e.g., inflammation-induced anhedonia – see below).

Lost in translation: a cautionary tale and lessons to be learned

Translational missteps that occurred in attempts to clinically translate the large body of research on the role of the hypothalamic-pituitary-adrenal (HPA) axis in depression and anxiety disorders illustrate fundamental problems that have contributed to the lack of progress in developing drugs aligned with pathophysiologic mechanisms that cut across disorders and contribute to specific symptomatology. Conservative estimates based on a search of grants funded by the National Institute of Mental Health in NIH Reporter using the search terms “HPA axis” or “cortisol” indicate that over $6.9 billion dollars were allocated to investigators for research on the HPA axis and cortisol from 1985 to the present [17]. Of note, these funds do not include dollars distributed by international funding agencies or for earlier studies that identified the HPA axis regulatory peptide corticotropin releasing factor (CRF) as a major mediator of behavioral pathology in animal models of stress [18, 19]. CRF was also found to be elevated in the cerebrospinal fluid of patients with depression and post-traumatic stress disorder (PTSD) [20, 21], and quickly became a therapeutic target for drug development. Despite the massive investment by funding agencies as well as the pharmaceutical industry in the development of drugs to block CRF (and other HPA axis hormones including cortisol), pharmacotherapeutics targeting the HPA axis failed to exhibit any meaningful efficacy in clinical trials for major depression, anxiety disorders, or PTSD [22–24]. Although the field might interpret these failures as being related to “wrong drug, wrong disorder” or a consequence of limitations in relevant animal models or even more worrisome, willful blindness, it is apparent that fundamental mistakes were made regarding clinical trial design that derived from the previously noted beliefs in the monolithic nature of psychiatric disorders and the idea that “one size fits all”. For example, in a review of the failed trials using CRF antagonists, critics suggested that prescreening patients for some form of HPA axis alteration prior to enrollment might have led to more positive outcomes [22]. Only about one third or less of outpatients with major depression have evidence of HPA axis alterations as reflected for example by an abnormal dexamethasone suppression test [25, 26]. Therefore, only a fraction of the patients included in the failed trials likely had the pathology of interest. If the clinical trial populations for these studies had been enriched for HPA axis pathology (e.g., using a biomarker related to abnormal HPA axis activity), the results may have been very different. The same criticism can be levied at studies using anti-inflammatory drugs to treat depression (see below) as well as trials examining a host of other drugs under development for the treatment of psychiatric disorders, including those targeting glutamate and its receptors. In virtually all cases, no prescreening for the pathology of interest was done (thus enrolling heterogeneous patient populations), no determination of target engagement was conducted (thus no verification that the drug hit the target), outcome variables were nonspecific measures of symptom severity (thus no evaluation of the specificity of behavioral response), and as a consequence, results have been wholly underwhelming. Given the heterogeneity of depression (and other psychiatric disorders), it is incumbent upon investigators, funding agencies and regulatory bodies to embrace the notion of specific pathologies that occur in subsets of patients with a given disorder (and across disorders) that are associated with pathology-specific biomarkers and symptoms. Without this realignment of drug development with the realities of pathology-specific changes in the brain and behavior, the field will be challenged to advance, and opportunities (e.g., CRF antagonists, anti-inflammatories, and glutamate receptor modulators) will be lost. These are the risks of maintaining the status quo.

THE PROMISE OF BIG DATA?

Increasing attention has been paid to large datasets (“big data”) including muti-omics and neuroimaging often in conjunction with advanced analytic strategies using machine learning to identify relevant genes, proteins, molecular pathways and neurocircuits (alone or in combination) associated with disease [27–31]. This with the goal of developing better prediction models of disease subgrouping and treatment response and hopefully better and more effective, targeted therapeutics. These approaches have tremendous power to detect relatively small effect sizes and using machine learning algorithms can explore and leverage the heterogeneity within patient samples to predict relevant patient subgroups (e.g., biotypes) and outcomes [30, 31]. However, the validity of these analyses is limited by the phenotypic characterization of the populations under study, with much better correlations for example among gene-wide association studies (GWAS) in depression where patients are stratified based on selected clinical features [32]. Furthermore, disorders are most often approached within the context of a given diagnosis, without full appreciation of subgroups across disorders that may be as or more relevant to pathophysiologic mechanisms, thus repeating mistakes of the past (e.g., the reification of diagnostic categories within current psychiatric nomenclature) [28]. As a case in point, recent studies in mood disorders have demonstrated that genetic liability for major depression is also associated with an augmented risk for anxiety disorders, supporting the transdiagnostic nature of genetic risk [33]. Moreover, once a gene or other biomarkers (neuroimaging or otherwise) are associated with a disorder or a biotype within a disorder, these associations do not immediately translate into an understanding of functional relevance as it relates to a pathophysiologic mechanism for a given disease process [34]. Indeed, hypotheses need to be formulated and tested in multiple model systems and ultimately in humans before a viable therapeutic can be developed and tested. Thus, finding for example that a series of genes are associated with a disorder is just the beginning of a journey and not an end in itself. In fact, despite the advantage of large datasets to identify genes relevant to psychiatric disorders, very few genes from these studies have been translated into mechanistic pathways relevant to the development of new or better treatments for psychiatric patients [29].

One gene that is in the process of being translated in this fashion is FK506 binding protein 5 (FKBP5). FKBP5 is involved in the regulation of glucocorticoid receptor function and ultimately the response to stress including the inflammatory response [35–37]. It is increasingly apparent that FKBP5 is relevant to pathologic responses to stress that may in turn drive symptoms across stress-related psychiatric disorders [36–38]. Of note, however, the story of FKBP5 reveals additional cracks in the reliance on big data to uncover relevant pathophysiologic mechanisms in psychiatric disease. In early case-control studies and a meta-analysis of these studies, the gene for FKBP5 was identified as being related to depression as well as PTSD, notably in the setting of prior exposure to childhood trauma [37, 39–41]. Subsequent studies have revealed that these findings are a result of epigenetic changes that occur in the FKBP5 gene as a result of early adverse experiences [37]. Thus, FKBP5 is a gene whose relevance to pathology is manifest primarily in the context of a gene X environment interaction that leads to epigenetic modifications [37], ultimately making its detection, even in large datasets, challenging, especially if deep phenotyping (e.g., environmental exposures such as childhood trauma) is not available and gene X environment interactions are not explored in the analyses. Consistent with this “blind spot” in GWAS as well as the practice of limiting analyses to isolated psychiatric disorders (e.g., depression), the FKBP5 gene has not been revealed as a gene relevant to psychiatric disease notably depression in large GWAS studies [24, 30], despite being one of the few genes where the pathophysiologic mechanisms are being worked out, and a therapeutic is likely not far behind [37, 42]. As development of FKBP5 antagonists proceed, with risk genotypes, epigenetic modifications and environmental experiences as well as possibly HPA axis dysfunction as qualifying biomarkers for targeting transdiagnostic subgroups of relevance, a pressing question is how will regulatory agencies respond? As it stands now, therapeutic development of FKBP5 will face a number of the hurdles noted above and would greatly benefit from an approach (and designation) more akin to that of targeted therapies in oncology, which are solely based on the pathophysiologic pathway targeted, agnostic to diagnosis.

Of note, leveraging large datasets and applying machine learning strategies has additional limitations [43]. There is no question that the application of machine learning to large datasets can reliably identify disease subgroups and predict response to a variety of treatment strategies [30, 31]. Such approaches are also increasingly being used to identify circuit-based changes associated with symptom clusters and disorders, which can provide the foundation for circuit-based therapeutic (neuromodulation) strategies including repetitive transcranial stimulation, deep brain stimulation, and magnetic resonance imaging-targeted ultrasound [27, 44–46]. However, machine-learning models lack the determination of causality that is inherent and essential to mechanistic modeling approaches [43]. Thus, agnostic predictors of disease subgroups and treatment response based on machine learning alone, while useful, will not advance the mechanistic understanding of pathophysiologic processes and therefore the development of targeted therapeutics, circuit-based or otherwise. Indeed, circuit-based changes inherently derive from related underlying alterations in brain states at the molecular and cellular level that ultimately drive the pathophysiologic mechanisms involved.

ANTI-INFLAMMATORY DRUG DEVELOPMENT: A CRISIS IN THE MAKING

Despite an impressive literature delineating specific mechanisms by which inflammation affects the brain and behavior as well as clear-cut strategies for identifying patients likely to respond to targeted treatments, clinical trials using anti-inflammatory drugs for relevant psychiatric disorders have failed to leverage available knowledge, repeating the same mistakes of “one size fits all” and the reification of diagnostic categories and nomenclature [47–50]. As a result, this field of research is floundering, and as was the case with the HPA axis may fail to deliver on its early promise.

A large body of data has been amassed demonstrating a cause-and-effect relationship across multiple psychiatric disorders between inflammation and behavioral changes, especially motivational deficits (e.g., anhedonia) and anxiety (Fig. 1) [48]. Subjects administered a variety of inflammatory stimuli including inflammatory cytokines [e.g., interferon-alpha], endotoxin and typhoid vaccination reliably exhibit changes in brain circuitry involving reward processing and threat sensitivity in association with symptoms of anhedonia and anxiety [48, 51–54]. In addition, blockade of inflammatory cytokines in patients with increased inflammation improves symptoms of anhedonia and anxiety, with evidence that such treatments target the specific brain circuits involved [52, 55–57]. These data complement a large body of research demonstrating reliable associations of neurovegetative symptoms with inflammatory markers that occur across diagnostic categories [58]. GWAS and case-control studies have also indicated that relevant genes associated with psychiatric disorders reliably represent pathways associated with inflammation and the immune response [59, 60]. Finally, a multitude of studies have reproducibly found increased inflammatory markers within subgroups of patients with a variety of psychiatric disorders including mood and anxiety disorders, PTSD and schizophrenia [61–63].

Fig. 1. Representative targets for treatment, verification of target engagement and primary outcomes in inflammation-induced behavioral change.

Stimulation of immune cells by microbial or danger-associated molecular patterns activate nuclear factor kappa B and inflammasome signaling pathways that drive the inflammatory response and are associated with metabolic shifts from energy efficient oxidative phosphorylation (OXPHOS) to energy expedient glycolysis. Metabolic reprogramming is also accompanied by increased fatty acid (FA) and amino acid (AA) synthesis that further supports rapid growth and proliferation of relevant immune cells. The peripheral inflammatory response including the release of inflammatory cytokines such as tumor necrosis factor (TNF), interleukin (IL)-1 and IL-6 as well as the acute phase reactant C-reactive protein (CRP) along with immune cell trafficking to the brain can in turn transmit inflammatory signals to the brain, ultimately influencing neurotransmitter metabolism, brain circuitry and behavior. Neurotransmitter systems implicated in the effects of inflammation on the brain include dopamine and glutamate. Primary neurocircuits that are engaged include those involving the basal ganglia and its connectivity to the ventromedial prefrontal cortex (vmPFC) and subgenual anterior cingulate cortex (sgACC) as well as the supplementary motor area (SMA), leading to behavioral changes including deficits in motivation and psychomotor activity. Brain regions including the insula, dorsal ACC (dACC), hippocampus and amygdala are also affected with resultant increased threat sensitivity, increased sensitivity to loss and ultimately hypervigilance and anxiety.

Clinical trials using anti-inflammatory drugs have seized upon these findings, and a host of studies using drugs with putative anti-inflammatory properties have been conducted [64]. Unfortunately, virtually every trial to date has enrolled patients from a single population of subjects within a given psychiatric disorder (e.g., depression, schizophrenia, bipolar disorder). In addition, only a few studies have focused on patients as a function of their inflammatory status, and these have yielded more promising results [56, 57, 65, 66]. Moreover, in all studies, the primary endpoints have been measures of total scores derived from standardized scales of symptom severity based on diagnostic categorization (e.g., the Hamilton Depression Rating Scale for studies on depression), as opposed to focusing on symptoms directly related to the impact of inflammation on the brain. Unfortunately, to date, the results have been underwhelming at best and non-significant at worse [64, 67, 68]. Indeed, a recent negative study on celecoxib and minocycline including one of the largest number of patients studied to date concluded “This large trial casts doubt on the potential therapeutic benefits of adjunctive anti-inflammatory drugs for the acute management of bipolar depression.” [67] Given the previously noted missteps with therapeutics targeting the HPA axis, these trials and related interpretation of their findings are “déjà vu all over again”.

WHAT IS THE SOLUTION?

Recommendations for the development of targeted therapeutics in psychiatry

For clinical trial design.

A multitude of opportunities exist for reinventing drug discovery in psychiatry with an eye toward developing targeted treatments. Using the impact of inflammation on the brain and behavior as a case in point, we have previously provided guidelines for clinical trial designs for anti-inflammatory drugs that are equally applicable to general recommendations for the development and testing of targeted treatments (Table 2) [47–49]. Briefly, within the context of an identified pathophysiologic mechanism, clinical trials should identify and enrich for specific subgroups of patients who exhibit evidence of the relevant pathology, agnostic to diagnosis. Although the choice of biomarkers to achieve this enrichment can be complex, experimental medicine studies will be required to tie relevant biomarkers to specific neurocircuits and behavior as well as their response to appropriate treatment interventions. Demonstration that the therapeutic being applied has targeted the pathology of interest as reflected by these biomarkers is also required, and primary outcome variables should be aligned with the specific neurocircuits engaged by the relevant pathology and associated symptom domains. Examples from the impact of inflammation on the brain exemplify potential therapeutic targets as well as pathologies that can be used for verification of target engagement and relevant circuit-based and behavioral outcome variables (Fig. 1).

Table 2.

Clinical trial guidelines for targeted therapies.

• Enrich the sample population using neurobiological and/or symptom variables that are associated with the pathophysiologic mechanism being targeted

• Consider a match/mismatch design whereby all patients receive active treatment but are blinded as to whether they have the biomarker and/or symptom profile consistent with the pathophysiology of interest and the hypothesized response

• Verify target engagement by the use of biomarkers in the brain and periphery including peripheral blood, cerebrospinal fluid and/or neuroimaging biomarkers

• Use objective biomarkers of response including neuroimaging and objective measures of behavior that are aligned with treatment effects on pathology-specific outcomes

• Align behavioral outcome variables with the pathophysiologic pathway of interest

• Treat traditional rating scales of treatment response as exploratory and interpret with caution

For researchers

Basic science researchers:

The chasm that currently exists between basic scientists and clinical researchers, if anything, has widened over the past decades [69]. Moreover, there is an ongoing and worsening physician-scientist shortage that is challenging the translation of the vast basic neuroscience knowledge to the clinic [70, 71]. Drawing from similar past challenges in oncology, the Nobel Laureate and immunologist Ralph Steinman (who discovered the dendritic cell) echoed sentiments that are painfully pertinent to psychiatric patients today [72]. “Patients,” he said at the time “have been too patient with basic research”. “Most of our best work is in lab animals, not people,” said Steinman, “But this has not resulted in cures or even significantly helped most patients” [72]. A greater effort must be made to train basic researchers in translational work, and colloquiums should be established that bring basic and clinical researchers together to realign translational and clinical research with findings from the bench. Of note is that in a recent recommendation of core competencies for graduate students in neuroscience, sadly there is no mention of training in translational research [73]. Finally, a greater push for smoothing the path for physician-scientists to develop careers as independent investigators in psychiatric research must occur [70, 71, 74, 75].

Clinical researchers (and clinicians):

There should be a greater appreciation that current therapeutics as well as novel drugs (e.g., ketamine, psychedelics, and THC) are highly non-specific (“chemotherapies for the brain”), and for ketamine and psychedelics, reverse engineering of the mechanisms involved is a low-yield strategy. Moreover, although circuit-based therapies are highly effective, more attention should be paid regarding the cellular and molecular mechanisms that are responsible for circuit dysfunction. In addition, although DSM categories of diagnoses have a role in clinical communication and billing purposes, there must be openness to identifying and labeling patients by their specific pathologies (e.g., “inflammation-related behavioral disorder with depression and anxiety”) and psychiatric drugs by their mechanisms of action as established in clinical populations (e.g., FKBP5 antagonists).

For funding agencies (NIMH).

Although the importance of rapid deployment of low-cost treatments that can improve availability and access to and reduce disparities in psychiatric care is undeniable, funding agencies need to make special efforts to balance their portfolios, limiting investment in non-specific therapies with no clear targeted mechanism of action. Funding truly transdiagnostic studies that address mechanisms and therapeutics involving clearcut pathophysiologic pathways aligned with relevant neurocircuits and behavioral domains (e.g., as defined in RDoC or otherwise) must be a priority [3, 76]. Increased emphasis should be placed on the translational relevance of basic science and clinical studies, including review criteria specifically addressing the likelihood of advancing targeted therapeutics. Clinical studies should also be encouraged to take the “basket trial” approach of oncology and include patients with convergent mechanistic pathologies across diagnostic categories with outcomes agnostic to diagnosis along with deep phenotyping beyond simple measures of depression or anxiety. Furthermore, studies using big data to identify -omics and neuroimaging profiles associated with pathology, should focus on heterogeneity within and across disorders, possibly through identifying subgroups (using machine learning) prior to analyses. These efforts should actively eschew current diagnostic nomenclature and emphasize environmental context and associated epigenetic biomarkers of risk. Limitations of machine learning from a mechanistic standpoint should also be appreciated.

For industry and regulatory agencies.

Clinical trials should follow the guidelines recommended in Table 2. Trials should not only embrace targeted subgroups of patients across diagnoses and verify target engagement at multiple levels, but also allow outcomes to be specific to the symptoms derived from the mechanism involved. This realignment of clinical trials will come at the expense of current DSM diagnostic categories and standardized symptoms scales (such as the Hamilton Depression Rating Scale) to measure outcomes. Thus, the reification of psychiatric diagnoses and the notion that “one size fits all” must be discarded and replaced with scales that capture specific symptom targets of the treatment in conjunction with possibly more inclusive measures of outcome (e.g., quality of life), which can capture overall treatment response. Moreover, in failed trials, efforts should be made to identify patient responders to reveal additional data for the refinement of hypotheses and patient subgrouping and ultimately the development of more specific therapeutics. Such an approach has found success in oncology where for example the response to everolimus, a drug targeting the mTORC1 (mammalian target of rapamycin) complex was tied to tumors harboring TSC1 somatic mutations [77]. Finally, the pharmaceutical industry should recognize that working with more specific mechanisms of pathologies that operate across disorders may lose market share in one disorder but pick up market share in others.

SUMMARY

Legacies of outdated and now reified diagnostic categories and the idea that “one size fits all” have plagued the development of targeted therapies in psychiatry and have relegated the field to broad spectrum therapeutics with unknown mechanisms of action, representing the equivalent of “chemotherapy for the brain”. The pressing need to address issues of treatment non-response and persistent symptoms has led to a revival of drugs from past eras including ketamine and psychedelics that add even more fuel to the fire of a field that is bereft of well-delineated pathophysiologic mechanisms that drive symptoms within and across psychiatric disorders. Further, despite the availability of big data and advanced machine learning modeling strategies, we have yet to make mechanistic advances using these approaches. Moreover, although circuit-based therapeutics are effective, they too are largely devoid of an understanding of the cellular and molecular mechanisms of circuit dysfunction. Of greater concern is that in cases where an understanding of the mechanism is available (e.g., the impact of inflammation or HPA axis pathology on the brain and behavior), clinical trials continue to use traditional methodologies that suffer from the legacies of the past that are blind to the heterogeneity of psychiatric disorders and the non-specificity of standard outcome variables. The way forward as modeled by the successes in oncology is to embrace mechanistic approaches that identify subgroups of patients with a given pathology within and across disorders targeting relevant pathways along the chain of pathology with verification and use of outcome variables directly linked to the circuits and symptoms involved. Researchers, regulatory agencies, and the pharmaceutical industry must all participate in the process and reject the legacy of non-specific, broad spectrum therapies and embrace the future of targeted treatments in psychiatry. If not now, when?