Psychiatry is still being largely practiced using a 19th century approach, relying on what patients are telling us and on our clinical impression. Most other fields of medicine are relying on more modern, 21st century approaches, using laboratory tests and other objective and quantitative assessments to help diagnose and treat people. The relative lack of accessibility of the brain to biopsies has created the need for indirect and/or peripheral ways of assessing brain function.
Biomarkers, as their name implies, are biological measures that serve as quantitative markers of the function of an organ or system. In the case of the brain , they can be molecular, electrophysiological, or imaging. For the brain, surrogate molecular markers can be found in peripheral tissues and fluids. The later include CSF, saliva, blood. In particular blood, containing secretion products of various tissues, and cells of the immune system, has become a useful accessible source of biomarkers. In cancer, the term of the art for this is “liquid biopsy”.
Why would markers in the blood correlate with brain function and with behavior? Next to the nervous system, the immune system is the most reactive, active and complex system in the body. It has some developmental commonalities with the nervous system. More importantly, it has bi-directional interactions during all of life. Common internal milieu (hormones, etc.) and external environmental factors (medications, etc.) lead to some common gene expression patterns in brain and white blood cells. These can be identified and used as biomarkers, using a careful four-step approach: discovery, prioritization, validation, and testing. These type of biomarkers ( epigenetic, RNA, protein, metabolite) vary over time, unlike DNA, i.e. they have a state component. They also have some integration of past events, and predictive ability for future events, i.e. have a trait component as well.
Step 1 in the process is discovery. Usually we focus on state, and choose quantitative feelings or thoughts or actions that we can use for our biomarker discovery, rather than using trait and broad diagnostic categories. A diagnosis is essentially describing a trait, i.e. feelings , thoughts and actions over time, i.e. state over time. Biomarkers discovered for state/assessment can have some utility for trait/diagnosis. It is easier to discover state biomarkers first, and then look at their trait capabilities, rather than the other way around. The reason is pragmatic- state biomarkers can be discovered using a powerful longitudinal within-subject design, that eliminates genetic variability and hence the cohorts sizes needed for detecting true signal. Moreover, the phenotype itself is more reliable within-subject, especially if it changes from one assessment to another, as the person can tell a difference between higher and lower intensity of feelings and thoughts. It may seem paradoxical that for discovering objective biomarkers for subjective feelings and thoughts, you have to rely on self-report of the subject. However, the subsequent prioritization of replicating findings using all sorts of human and non-human data, and especially the validation and testing steps in independent cohorts, will prove if you have discovered true signal or not.
The prioritization step 2, recommended after step 1 discovery, is an opportunity to not just rely on the Fischerian probability of your own results in the discovery cohort, but also inject some Bayesian probability by using other people’s prior results in the field. We do that through an approach we developed over the last two decades, called Convergent Functional Genomics. This approach integrates genetic and gene expression evidence, from brain and peripheral tissues, from human and non-human studies. We have built manually curated databases of the literature to date. We focus on primary data studies rather than secondary meta-analyses, and taking the findings that were deemed significant by the authors using their design and criteria. We also focus on unbiased discovery-type studies, such as whole genome and whole transcriptome studies, rather than biased hypothesis driven studies, to avoid “popularity” effects, i.e. genes that are studied by many people because they are fashionable, easy to study, or believed to be important. Moreover, to avoid this effect, when we score the convergent evidence, we cap points such as that multiple citations of the same type of evidence for the same gene do not count extra. We weigh more heavily the gene expression evidence than the genetic evidence; the brain evidence vs. the peripheral evidence; and the human evidence vs. comparable non-human evidence. The prioritization step with CFG increases the odds that the biomarkers are reproducible and relevant to the disorder. In essence it uses all the data in the field, thus avoiding overfitting based on the particular discovery cohort used, no matter how large.
The validation step 3 is usually done in an independent cohort of clinically severe subjects (or, in the case of suicide, a cohort of suicide completers). The goal is to see which of your candidate biomarkers are changed even more so in the validation cohort than in the discovery cohort. A stepwise change in expression from the low symptoms group in discovery to the high symptoms group in discovery to the validation group, accompanied by a significant ANOVA, filters biomarkers that additionally reproduce and are related in a quantitative way to the severity of the phenotype.
After the first three steps, a list of biomarkers is generated for additional testing in step 4, in additional independent cohorts of patients. The list of biomarkers can be generated using a cutoff based just on the validation step, or a more comprehensive convergent functional evidence (CFE) cutoff based on the first three steps. This testing is for additional reproducibility, for predictive ability, and for clinical utility. The goal is to see which of the biomarkers are good predictors of state (symptom severity), and which are good predictors of trait (future clinical worsening).
In the end, you have a list of biomarkers that are demonstrated to all around track and predict the phenotype of interest, and are ranked for that ability using a convergent functional evidence (CFE) score for all four steps. You can focus on single best markers for different populations: markers that work in all, that work best in a gender, or diagnostic group, or subtype of the disease, or combinations of these, i.e. male bipolars with anxious subtype of suicidality. Besides single best markers, you can make panels of markers for each of these groups, to be broader spectrum and account for biological variability in the population. Our liquid biopsy tests are currently whole-genome, so we can look at the results from larger panels, not just individual best biomarkers, or panels of top biomarkers. There is, however, a decrease in signal with larger panels, as the best biomarkers are diluted by the less good ones.
Once the biomarker results are in for a patient, their expression values get normalized using proprietary bioinformatic approaches. Those values then get compared to a large database of previously tested patients with the disease/phenotype of interest (for example, symptom intensity like suicidal ideation, future risk like future hospitalizations for suicidality, or suicide completion based on coroner’s cases), and the person gets a qualitative risk category assigned (high, intermediate, or low), as well as a quantitative score/rank percentile. This information, provided to their doctor, can inform their care, and repeat testing can also assess how well they respond to treatments.
In addition, each biomarker is tied to multiple existing psychiatric medications that can influence its levels of expression in a direction opposite to the one in disease, i.e. normalize its expression. This connection is done bioinformatically, using our databases where experimental data from human and animal model studies, from us and other groups, has looked at the gene expression effects of various drugs. Each biomarker that is altered in a patient gets a score if it is affected favorably by a specific psychiatric drug or not using our databases, and the whole panel of biomarkers gets a score for that drug. Thus, existing psychiatric drugs can be ranked for potential ability to treat/normalize the biomarkers of that particular patient, and their effect monitored with biomarkers.
Finally, a panel of biomarkers can also be used to march patients with non-psychiatric medications or nutraceuticals, using large drug gene expression databases. That may provide new avenues for treatment, especially of co-morbid medical and psychiatric conditions, as well as leads for new drug development in psychiatry with repurposed drugs.
We have demonstrated the above series of steps and approaches, in published studies to date for six indications: suicidality, pain, PTSD, memory/Alzheimer, longevity, and most recently mood disorders (depression/bipolar, in press). Other indications, such as anxiety, and schizophrenia (hallucinations/delusions) are forthcoming in the near future.