Depression and the Nervous System

Depression is one word covering many conditions

A person cannot get out of bed and cannot say why. Food has no taste. The phone rings and answering it feels like lifting something heavy. Everything that used to be worth doing is still there, and none of it pulls.

That description fits tens of millions of people, and it is where the trouble starts. Two psychiatrists were asked to write the clinical seminar on depression for the Lancet, the piece a doctor reads to find out where the field stands. They open by conceding the problem.

Detection, diagnosis and management are all difficult, they write, because depression presents in many different ways, runs an unpredictable course, and responds to treatment variably. The reference work on the condition begins by admitting that the category does not behave like one disease.

1,030 symptom profiles in one trial

Then somebody counted. Eiko Fried is a psychologist who studies how psychological measurements behave. Randolph Nesse is a psychiatrist who asks what a symptom is for. Together they took the largest depression trial ever run and asked a question nobody had bothered with. If 3,703 people all carry the same diagnosis, how many different symptom pictures are they actually showing? They counted 1,030 distinct profiles.

Of those, 501 were reported by exactly one person. The single most common pattern covered 1.8% of the sample. Controlling for how severe each person was did not shrink the spread. The diagnosis names the endpoint everybody can see. It says nothing about the road anybody took to get there.

STAR*D and the disputed remission rate

The clinical consequence was measured by that same trial. STAR*D put 3,671 outpatients through up to four successive treatment steps, moving anyone who did not remit on to the next one. Remission fell at every step: 36.8%, then 30.6%, then 13.7%, then 13.0%. Two thirds got there in the end, and the people who needed more steps relapsed more often afterwards.

That headline figure is itself disputed. A second team reanalyzed the same patient-level data with fidelity to the original protocol and reported cumulative remission of 35.0% rather than 67%. The gap came from a non-blinded clinic rating standing in for the scale the protocol specified, and from counting patients already in remission when a step began. That disagreement is itself a finding, about how the field reports what it finds.

Symptom clusters answer treatment differently

One group then showed that the averages were hiding structure. The psychologist Adam Chekroud and colleagues at Yale split the symptom checklists into empirical clusters, replicated the split in a second trial, and asked whether the clusters answered treatment differently. They did.

Eight of nine treatments worked better on core emotional symptoms than on sleep, or on the atypical cluster of heavy sleep and increased appetite. The gap between two drugs on a single cluster was sometimes wider than the gap between drug and placebo overall.

Heterogeneity on that scale is what the rest of the evidence follows from. Depression is also dangerous. A meta-review screened 407 systematic reviews and pooled the 20 that reported mortality, covering more than 1.7 million patients and over a quarter of a million deaths. Every psychiatric disorder carried raised mortality, many of them at levels the authors compared with heavy smoking. Depression sat among the four conditions with the highest suicide risk.

What has to be ruled out before depression is treated as depression

Several findable conditions produce the pattern that gets diagnosed as depression, and a blood test catches some of them. Looking for them is a physician's first duty, and it comes before every other question raised here.

The thyroid gland is the clearest example. It sets the pace of metabolism across the whole body. When it runs slightly slow without yet crossing into overt disease, the state is called subclinical hypothyroidism, and a blood test finds it. One group pooled data on 12,315 people and found that those with it carried more than twice the risk of depression. In older people the risk was about 1.7 times.

Then they asked the question that matters. Does correcting the thyroid correct the mood? Across the trials available, levothyroxine did not significantly improve depression scores.

Finding the driver and reversing the driver are different achievements. Medicine keeps discovering that the two come apart.

Not every illness reduces to poor regulation. Structural, genetic, infectious, toxic and malignant causes are real, and they have to be treated as what they are. The responsible formulation is narrower and more useful. Every illness has a tonal expression, meaning a change in how the whole system is organized, and many illnesses are started, maintained or amplified by failures of that organization. Depression is a condition where the second clause carries a great deal of weight, and the first clause still applies in full.

The presentations that need a physician promptly

Some presentations are a reason to see a physician promptly rather than to keep reading. Thoughts of harming yourself, at any intensity, at any hour; if those thoughts are present now, contact emergency services or call or text 988, the Suicide and Crisis Lifeline. A first depressive episode arriving late in life, particularly alongside any change in memory, walking, speech or continence.

Depression with unexplained weight loss, fever, night sweats or a new physical symptom. Depression that began with or shortly after starting a new medication. A mood that has swung the other way into unusual energy, reduced need for sleep or uncharacteristic risk-taking, which points at a different diagnosis with a different treatment.

One more belongs on that list, because it is specific and time-limited. A team inside the American drug regulator pooled 372 placebo-controlled trials covering 99,231 adults and found the suicidality risk of starting an antidepressant to be strongly age-dependent. Under the age of 25 the risk rose. Between 25 and 64 the net effect looked neutral for suicidal behavior and possibly protective for suicidal thinking, and over 65 both fell.

New or worsening thoughts of self-harm in the first weeks after starting or changing an antidepressant, especially in someone young, is a reason to contact the prescriber the same day. That is a reason to call rather than a reason to stop. Stopping an antidepressant abruptly carries dangers of its own.

Each of those needs a person with a stethoscope and a prescription pad. The broad psychiatric picture is handled elsewhere in this library, at mental health, and the condition that most often travels with depression has its own page at anxiety.

How the chemical imbalance story was built backwards

Almost everyone has been told that depression is a chemical imbalance in the brain. The way that story was built explains why it has been so hard to confirm.

Schildkraut and Coppen reasoned from two drug accidents to a monoamine shortage

It began with two accidents. A drug given for tuberculosis lifted patients' moods. A drug given for high blood pressure appeared to lower them. Both act on the brain's monoamines, a family of signaling molecules that includes serotonin, noradrenaline and dopamine. In 1965 a psychiatrist at the National Institute of Mental Health named Joseph Schildkraut reasoned backwards from the drugs to the disease.

If these compounds move mood by moving monoamines, perhaps depression is a shortage of monoamines. He proposed it as a catecholamine deficiency, catecholamine being the branch of that family holding noradrenaline and dopamine, and he was explicit that his evidence was indirect. His own review gathered the blood-pressure observations together with their problems, and later authors pressed the problems harder, since most people who took that drug never became depressed.

Two years later a British psychiatrist, Alec Coppen, made the parallel case for the indoleamines, the branch holding serotonin. Between them these two reviews set the terms of the next sixty years of argument. Both reasoned from what the drugs did back to what the illness must be. Neither measured anybody's brain.

The depletion experiments

Testing the idea meant finding a way to lower serotonin in a living person on purpose. There is one. The brain cannot build serotonin without tryptophan, an amino acid it can only get from food. Give someone a drink containing every other amino acid and the body sweeps tryptophan out of the blood to keep pace. The psychiatrist Pedro Delgado and his group at Yale did exactly that to patients who had recovered from depression, dropping plasma tryptophan by 70% to 80% in five hours.

The result was striking. Of fifteen patients who had recovered on fluoxetine, which works through serotonin, eight relapsed within hours. Of fifteen who had recovered on desipramine, which does not, one did. Depression, apparently on demand, and gone again once the drink wore off.

That experiment is easy to over-read, and the pooled evidence shows why. A psychiatry group at the Academic Medical Center in Amsterdam gathered every published study that had stripped a monoamine out of a human being. They found 73 tryptophan-depletion studies and could pool 45 of them, along with 8 that removed tyrosine, the food-derived building block for noradrenaline and dopamine.

Draining serotonin did nothing at all to the mood of healthy people. It lowered mood modestly in people with a family history of depression, and more in people who had been depressed and recovered. Their own conclusion was that depletion marks a vulnerability rather than a cause.

That is the shape of the whole literature. The same input landed differently depending on the system it arrived in. Nobody had to be handed a chemical shortage to become depressed. Some systems answered the shortage and others never noticed it.

The serotonin evidence is genuinely contested, in both directions

A group led by the London psychiatrist Joanna Moncrieff did something more systematic, in work published online in 2022. Rather than gather studies, they gathered the reviews, across all six areas where a serotonin deficiency should have surfaced. Serotonin and its breakdown product in body fluids. Receptor binding. Transporter levels. The depletion experiments. The transporter gene alone, and the same gene in combination with stressful life events. Seventeen studies met their criteria, twelve of them systematic reviews or meta-analyses.

Two of those areas need translating. The serotonin transporter is the pump that clears serotonin out of the gap between two nerve cells once its message has been delivered. The gene that builds that pump comes in versions of differing efficiency, and a famous line of research proposed that one version left people more vulnerable to depression after hard events.

No area produced consistent evidence that depressed people have lowered serotonin. One pooled analysis found that lowered plasma serotonin tracked antidepressant use rather than depression itself.

That conclusion is contested. Thirty-five researchers, led by the psychiatrist Sameer Jauhar, replied in the same journal. They argued the conclusion was overstated, citing weaknesses in the review process, selective reporting, oversimplification and errors of interpretation. They worked through the depletion and imaging literatures in detail to make the case. Anyone citing the umbrella review has to carry the reply alongside it.

Two things can be said without picking a side. The umbrella review asked whether serotonin is low in depression. It did not ask whether antidepressants work, and its authors never claimed that it had. Those are separate questions with separate answers. Sixty years of measurement aimed at one molecule, and the answer is still argued over. That is information about the question as much as about the molecule.

The model does not read serotonin as irrelevant. Serotonin is a real signaling system with real effects on sleep, appetite, aggression and mood. The reading concerns where the organization lives. Tone is carried by the relationship among many signals rather than by the level of any one of them. A search aimed at a single note will keep returning ambiguous answers about a chord. That is how a sixty-year ambiguity looks from inside the Unified Model of Tone.

Antidepressants beat placebo, and the margin is argued over

If the deficiency story is in trouble, an obvious question follows. Do the drugs built on it work at all?

The best answer available is a network meta-analysis led by the psychiatrist Andrea Cipriani, and the way it was assembled is the part that matters here. Trials that find nothing are less likely to be published, so any review of the published literature is reading a biased sample. This team wrote to regulators and manufacturers and chased the unpublished trials too.

Across 522 double-blind randomized trials covering 116,477 people, every one of the 21 drugs beat placebo. The odds ratios ran from 2.13 for the strongest down to 1.37 for the weakest. An odds ratio above 1 means the drug produced a response more often than the dummy pill did. These are real differences, produced by the same methodological culture that keeps finding the serotonin account wanting.

A drug can work without the story about why being right. Those are two claims, and only one of them is in trouble.

The publication-bias correction

Two qualifiers belong immediately after. The first came from a psychiatrist who had worked as a reviewer inside the American drug regulator and knew something the journals did not show. The agency holds every trial a company ran, including the ones nobody published.

Erick Turner pulled the files on 12 antidepressants and 12,564 patients, then hunted for the matching journal articles. Read from the published record, 94% of trials looked positive. Read from the regulator's record, 51% were. Pooling the journal versions inflated the apparent benefit by about a third.

The severity question, open on both sides

The second qualifier is the one the rest of the evidence turns on, and it needs a piece of arithmetic first. Effect size states the gap between two groups in units of the spread inside them, so that studies using different measurements can be compared. Around 0.2 counts as small, 0.5 as moderate and 0.8 as large. That scale runs through the rest of the evidence here.

Instead of pooling trial averages, the psychologist Jay Fournier and colleagues asked the original investigators for the raw data on every individual patient, and got it for 718 people across six studies. Then they plotted the drug-minus-placebo benefit against how depressed each person had been at the start, measured on the Hamilton scale, the standard clinician-rated severity score. Below a score of 23 the difference was smaller than the conventional threshold for a small effect. It crossed into clinical significance at 25.

That finding is contested too. A later team pooled individual patient data from six placebo-controlled trials, 2,464 people, and tested the same severity question. The interaction between starting severity and treatment was not significant, and it held up across every sensitivity analysis they ran. Their conclusion was that similar benefit can be expected whether depression is mild, moderate or severe. The question is open.

The model needs neither answer. In both datasets the same molecule at the same dose did different things to different people, and severity is only one candidate for what separated them. An input does not carry an outcome by itself. An input meeting a particular organization produces an outcome, and the organization is the variable neither analysis recorded.

Antidepressants and psychotherapy help a great many people, and relief has real value whatever produced it. Stopping an antidepressant abruptly can be dangerous, and any change to a medication belongs with the prescriber who knows the person.

The cortisol fault in depression is the failure to come down

Load is the first of the three foundations of tone that carry depression's signature, and the question it asks is what holding a state costs.

The idea belongs to Bruce McEwen, a neuroendocrinologist at Rockefeller University who spent his career on what stress chemistry actually does. What the body releases to meet a demand protects it in the short run. Cortisol mobilizes fuel, sharpens attention and quiets systems that can wait.

When the demand never lets up, the same chemistry does damage, and McEwen called the accumulated wear allostatic load. The cost is not paid by the stressor. It is paid by the machinery that keeps answering it, which is the argument the load page works through in full.

Raised cortisol shrinks under scrutiny

The obvious prediction is that depressed people run high cortisol, and for forty years studies reported exactly that. The psychologists Cinnamon Stetler and Gregory Miller added it up properly, across 671 effect sizes from 361 studies covering 18,454 people. Cortisol was raised, at an effect size of 0.60. Then the same authors restricted the analysis to studies meeting minimal methodological standards, and the effect fell by almost half, to 0.33.

What remained concentrated in three groups. Older hospitalized patients. Those with psychotic features, meaning a depression severe enough to carry false beliefs or perceptions. And those with melancholic features, the pattern of early waking and weight loss that the biology splits along. That is a subgroup finding that had been reported as a universal one.

The curve that will not come down

A second meta-analysis asked a different question and got a more interesting answer. A group at the University of California, San Francisco looked at the shape of the cortisol curve rather than its height.

Across seven studies and 196 people, depressed and non-depressed participants started at similar levels and rose to similar peaks under a laboratory stressor. The difference appeared afterwards. More than 25 minutes after the stressor ended, the depressed group was still elevated. The stress response of a depressed nervous system mounts normally and fails to stand down.

Recovery time is a different class of measurement from a resting value, and the model expects it to move first. A snapshot reports where a system is sitting. How long it takes to come back reports what the system can still do. A measure of regulation, if one is ever built for depression, will be built out of measurements like that one.

Two depressions point opposite ways on the same biological axes

If depression is really several conditions under one name, the biology should split along the clinical lines. Twice now it has.

Melancholic depression is the version with early waking, weight loss and mood at its worst in the morning. Atypical depression is the version with heavy sleep, increased appetite and weight gain. A Dutch cohort study sorted 776 people into chronic melancholic depression, chronic atypical depression and healthy controls, then compared cortisol output, inflammatory markers and metabolic measures.

The two groups came apart cleanly. The epidemiologist Femke Lamers and colleagues found the melancholic patients carrying the higher cortisol output, and the atypical patients carrying the higher inflammatory markers, body mass index, waist circumference and triglycerides.

One symptom, two biologies

The second split is sharper, because it turns on a single symptom. One team scanned 23 depressed people whose appetite had increased, 31 whose appetite had decreased, and 42 healthy controls, having drawn blood and saliva first. The two depressed groups matched each other on depression severity, on anxiety, on loss of pleasure and on body mass index. On the biology they diverged completely.

The appetite-loss group ran high cortisol. The appetite-gain group ran high insulin, and needed more of it to do the same job, a state called insulin resistance. They ran high leptin, the hormone fat tissue releases to report how much fuel is stored. They also ran high on two markers of immune activity. C-reactive protein is what the liver raises when inflammatory signaling is running, and interleukin-6 is one of the proteins immune cells use to talk to each other and to the brain.

Same label. Same severity. Opposite directions on the same axes. Average the two groups and the number you get belongs to neither of them.

That average is what most depression trials report. A trial recruits by diagnostic category, delivers one predetermined input to everybody, and reports the mean. If the sample holds two organizations pointing opposite ways, a genuinely large effect in one of them reads as a modest effect overall.

The model states this as a prediction rather than a complaint. Stratify by a measure of regulatory state taken before treatment, specify from that measure who receives what, and the matched arm should outperform the pooled average the literature keeps reporting.

Both studies are cross-sectional, meaning the biology was measured at the same moment as the symptoms. Neither can say the organization came first. What they can say is that one diagnostic word is covering at least two different bodies.

Inflammation is raised in about a quarter of depressed patients

Inflammatory signaling that will not switch off is the same bill as a cortisol curve that will not come down. The immune system is the body's other standing defense, and it has the same failure mode: a response built for episodes, left running.

That an immune state can produce a mood state is not in dispute. An infection brings loss of appetite, loss of interest and the wish to lie still in a dim room, a coordinated withdrawal that has a name, sickness behavior. Robert Dantzer works where the immune system meets behavior.

• He assembled the animal and human evidence showing that the brain produces that state in response to the body's own inflammatory signals rather than to the microbe.
• Andrew Miller and Charles Raison, two psychiatrists who spent their careers at the same border, argue that the resemblance to depression is no accident.
• That loop is worked through on its own page in this library, at inflammation.

The number that matters for depression is a head count. In how many depressed people is inflammation actually raised?

A Toronto pharmacology group led by Yekta Dowlati pooled the 24 studies that had measured cytokines, the immune system's signaling proteins, in patients meeting full diagnostic criteria. Two came out clearly higher in depression: tumor necrosis factor alpha and interleukin-6. The other six they could test showed no difference at all.

Averages still cannot answer the question, because a raised mean can come from a few very high values. A psychiatry group at Cambridge led by Emanuele Felice Osimo counted heads instead of means, across 37 studies covering 13,541 patients and 155,728 controls. About 27% of depressed patients had C-reactive protein above 3 mg/L. About 58% had it above 1. Against matched healthy controls the odds of a CRP above 3 mg/L were 1.46 times higher.

Any account that calls depression an inflammatory illness has just described about a quarter of the people reading it. That subgroup was not explained by inpatient status, by antidepressant use, by age, by body mass index or by ethnicity. It sits there, unexplained, inside the category. Roughly three quarters of depressed people are not inflamed, and any account starting from inflammation owes the reader an answer about which quarter it means.

Switching the immune system on builds a depression; switching it off helps only the inflamed

Medicine cannot deliberately make a healthy person depressed. It comes closest by accident, in a clinic treating something else entirely.

Interferon alfa is a signaling molecule the body makes during viral infection. Given as a drug at high dose it switches the immune system on hard, and it has been used against melanoma. It also makes a large fraction of the patients who receive it depressed, which turns an oncology ward into the nearest thing available to a controlled experiment in causing depression.

A depression assembled in slow motion

The psychiatrist Dominique Musselman and her group at Emory randomized 40 such patients to the antidepressant paroxetine or to placebo, starting two weeks before the interferon began. Within twelve weeks major depression developed in 45% of the placebo group and in 11% of the paroxetine group.

The researcher Lucile Capuron, in the same department, then watched those 40 patients assemble a depression in slow motion, and the sequence is worth having. Loss of appetite, fatigue and pain arrived within two weeks in a large proportion of patients. Depressed mood, anxiety and trouble thinking came later, and mainly in those who went on to meet full criteria.

Paroxetine helped the mood, the anxiety, the thinking and the pain. It did much less for the fatigue and the appetite loss. One diagnosis produced two symptom families with different arrival times and different drug sensitivities, inside 40 people in a single clinic.

Infliximab and the split that reversed

If inflammation drives depression, blocking it should lift mood, and that trial has been run. Charles Raison randomized 60 patients whose depression had resisted treatment to three infusions of infliximab, an antibody that blocks tumor necrosis factor alpha, or to placebo. Overall there was no difference between the groups. By the ordinary reading, the trial failed.

Then the analysis split the sample by how inflamed each person was at the start. Above a CRP of 5 mg/L the drug beat placebo. At or below 5 mg/L the placebo beat the drug. One input met two starting states and traveled in opposite directions, inside a trial that was reported as negative.

That split is an interaction inside 60 people, and the response-rate contrast it rests on did not itself reach significance. It also carries a prediction, and the prediction has since been put to the test. The psychiatrist Roger McIntyre and colleagues ran a separate trial in bipolar depression.

Sixty adults with bipolar I or bipolar II depression were enrolled on evidence of inflammatory activation recorded before treatment, then randomized to adjunctive infliximab or to placebo. The reduction in symptom severity at twelve weeks was not significant. Sorting people in advance on how inflamed they were did not reproduce the result that sorting them afterwards had produced.

That failure has to be read rather than stepped around. Enriching a sample on a single blood marker is a different operation from stratifying it on a measure of regulation, and the model has to say which of the two it claims. It claims the second.

The one signal that did emerge in that trial points the same way. The patients who separated from placebo were those reporting childhood physical or sexual abuse, which is a fact about a developmental history rather than about a blood value on the day.

The wider anti-inflammatory record and the genetic null

The wider anti-inflammatory literature is messier still. A Danish group pooled all 36 randomized trials of anti-inflammatory painkillers, cytokine blockers, cholesterol drugs, steroids, and two agents borrowed from other uses, an antibiotic called minocycline and a diabetes drug called pioglitazone. Added on top of antidepressants the pooled effect was moderate to large. Remission rates roughly doubled, and response rates rose by about three quarters.

The same authors then reported that every included study carried a high risk of bias, and that infections trended upward. In the analysis of these drugs used on their own, the trials disagreed with one another almost completely, a figure reported as 93% heterogeneity. They called explicitly for work identifying which subgroup benefits.

Genetics supplies a clean null. Inherited variants are dealt at conception, before any illness or habit can intervene, which makes them a natural randomization. A group at the Max Planck Institute of Psychiatry in Munich, led by Nils Kappelmann, used genome-wide data from hundreds of thousands of people.

They asked whether variants that raise inflammation also raise depressive symptoms. C-reactive protein was genetically correlated with individual symptoms about as strongly as body mass index was. The causal analysis did not consistently support inflammation causing them. One signal survived every sensitivity check: genetically raised interleukin-6 signaling was associated with suicidality.

That null is a real result. The model reads it this way. A driver that is real in a quarter of a category should not surface as a cause of the whole category. A genetic instrument tests the average person rather than the inflamed one. The same data also sit with inflammation lying downstream of the depressed state instead of upstream of it.

What the model owed the reader was the interaction in the stratified data, and one prospective test of that interaction has come back empty. A second failure of the same kind, in a sample sorted on a regulatory measure rather than on a marker level, would take this section apart.

Heart rate variability runs low in depression, and some antidepressants lower it further

Oscillation is the second foundation carrying depression's signature: the rhythms a system runs and the range it can move through. The fastest one you can feel is the gap between two heartbeats.

That gap is never quite the same twice. The variation records a brake being applied and released, beat by beat. The vagus nerve is the body's main parasympathetic line, running from the brainstem down to the heart, lungs and gut, and its traffic slows the heart. The sympathetic system is the accelerator. Heart rate variability, which has a page of its own in this library, measures how freely the brake is working.

A system with a good brake can change speed quickly in either direction. Depressed people have less of that variability. The psychologist Andrew Kemp pooled 18 studies, 673 patients and 407 controls, and found variability lower and falling further as depression deepened. That is a clean finding with a dirty confound sitting underneath it.

Separating the illness from its treatment

Three drug classes matter for what follows. The tricyclics are the first generation, named for their three-ring chemical shape, and they act on many receptor systems at once. The dual-action agents block both the serotonin pump and the noradrenaline pump. The SSRIs block the serotonin pump alone.

A Dutch cohort of 2,373 people was large enough to separate the illness from its treatment. Both currently and formerly depressed participants had lower variability than controls, and neither lifestyle nor anxiety explained it. Then the analysis accounted for psychoactive medication. One association disappeared and the other was strongly attenuated. The authors concluded that the association was mainly driven by the antidepressants.

Two years later the same cohort was measured again, 2,114 people this time, which let the researchers watch individuals start and stop drugs. Starting a tricyclic or a dual-action antidepressant raised heart rate and lowered vagal control. Starting an SSRI lowered vagal control without raising the heart rate. Stopping, in every case, moved the numbers back toward those of non-users.

Kemp's meta-analysis disagrees about that last class. It found the tricyclics lowering variability, and the newer serotonin agents having no significant effect on it. The two results have not been reconciled, and their designs differ in a way that matters. One watched the same individuals start and stop a drug, reading a change within a person. The other compared treated patients against separate controls. Where they agree is the older agents.

So is the finding the illness, or the medication? The psychologist Celine Koch and colleagues answered the first half of that by restricting a meta-analysis to unmedicated adults alone. Every measure of variability they could extract was still lower in depression.

Both things can be true at once. The illness lowers the brake, and at least some of the drugs lower it further. An account reporting only one half of that is incomplete.

One sleepless night lifts depression in half of patients, and sleep brings it back

The daily rhythm is the slow oscillation, and in depression it comes apart early.

A sleeping brain does not hold one state all night. It cycles between quiet, deep sleep and an active, dreaming state called REM, named for the rapid eye movements that give it away. Reaching that first REM period normally takes about ninety minutes. Sleep laboratories watch the cycle by recording brain waves, eye movements and muscle activity overnight, a procedure called polysomnography. They had been recording psychiatric patients for thirty years before anybody pooled the results.

REM arrives early in mood disorders

The sleep researcher Ruth Benca gathered 177 studies covering 7,151 patients and controls. Most psychiatric groups slept less and less efficiently, with the loss falling on the deep, non-REM half of the night. In mood disorders two things changed. The proportion of the night spent in REM went up, and the delay before the first REM period got shorter. Patients with mood disorders differed from healthy sleepers more often and more strongly than any other group.

No single sleep variable is specific to any one disorder, so none of this diagnoses anything. It says the rhythm is disturbed, and it says the disturbance is not incidental. Broken sleep also runs ahead of the mood. A meta-analysis of 21 longitudinal studies found that people with insomnia and no depression carry about twice the risk of developing it. Sleep physiology has its own page here, at sleep, and chronic insomnia has another, at insomnia.

The result that ought to be famous

Keep a depressed person awake all night and roughly half of them feel better by morning, often dramatically, often within hours. Two psychiatrists at Irvine, Joseph C. Wu and William E. Bunney, collected 61 papers covering more than 1,700 patients and described the effect in 1990. It does not last. Among unmedicated responders, 83% relapsed after a single night of sleep, and even a short nap could bring the depression back.

Twenty-seven years later a team at a Philadelphia veterans' hospital put a modern number on it across 66 independent studies. The response rate was 45% in studies with a randomized control group and 50% in those without. It did not vary by the kind of deprivation, the kind of patient, medication status, age or sex.

Nobody proposes staying awake as a treatment. Blinding is impossible in a study of this kind, and no mechanism is established. The phenomenon is solid and the explanation is missing.

A depression that lifts in eight hours and comes back with one night of sleep is behaving like a state the system can be knocked out of, and falls back into. A chemical shortage does not refill itself over a nap. That is what a stable configuration looks like from the inside. The system sits in one setting, a large enough perturbation moves it out, and once the perturbation is gone it settles back.

A claim like that has consequences a reader can check. Improvement should arrive in steps rather than in proportion to the size of the input, which is what an overnight lift and a same-day relapse are. The reading is in trouble if entrenched depressions turn out to resolve smoothly and proportionally when one variable is corrected.

Moving the daily rhythm moves depressed mood

If the rhythm is a variable the mood rides on, then moving the rhythm should move the mood. Three lines of evidence bear on that, and none of them is clean.

Light against fluoxetine

Light boxes had been accepted for winter depression and dismissed for the rest of the year. A Canadian trial led by the psychiatrist Raymond Lam randomized 122 adults with non-seasonal depression into four arms: morning light plus a placebo pill, fluoxetine plus a sham device, both active, or neither. The sham was an inactive negative ion generator. Over eight weeks, light alone and the combination both beat placebo. Fluoxetine alone did not.

That is one trial of 122 people, and no sham device blinds anybody to whether a bright light is switched on. The fluoxetine arm was too small to detect a drug effect reliably, so the trial is evidence about light rather than evidence against a medication.

The wider literature says the same thing more cautiously. A review from McMaster pooled 20 randomized trials and 881 participants with non-seasonal depression and found a standardized mean difference of 0.41 favoring light. The trials disagreed with one another substantially, at 60% heterogeneity, and the risk of bias in them was high enough that the reviewers graded the overall quality of the evidence as poor. The summary is a real effect inside a weak literature.

Making the overnight effect hold

The psychiatrist Francesco Benedetti and his group in Milan asked whether the overnight effect could be made to hold by resetting the rhythm instead of merely interrupting it. Sixty inpatients with bipolar depression received repeated total sleep deprivation combined with light therapy over one week, on top of their existing medication.

Response was 70% in patients whose depression had not previously resisted drugs and 44% in those whose had. Nine months later 57% of the responders in the first group and 17% of the responders in the second were still well.

Two things belong with that result. It is an open trial with no control group, in bipolar rather than unipolar depression, in patients who all stayed on their medication. And every one of them was an inpatient under medical supervision, which is not incidental. Deliberately going without sleep can tip a person with bipolar disorder into mania or a mixed state, the mood swing that is itself a reason to see a physician promptly. It is not something for anybody to attempt alone.

What the trial does show is that the overnight effect can hold when the rhythm is moved along with it.

The rhythm on 91,105 wrists

The third line is the largest. Asking people whether their sleep is regular gives you their opinion. UK Biobank strapped accelerometers to 91,105 wrists for a week and got the rhythm itself. From those recordings a team at the Institute of Health and Wellbeing in Glasgow, led by Laura Lyall, derived relative amplitude.

That is a measure of how sharply a person's most active hours separate from their least active ones. A high amplitude means clear days and quiet nights. A low one means the two have blurred together.

Flatter rhythms tracked lifetime major depression, bipolar disorder, mood instability, loneliness, lower happiness and slower reaction times. The associations held independently of how much activity a person did in total, and independently of childhood trauma.

The effect sizes are small, about a 6% change in odds per step down the amplitude scale, and the design is cross-sectional, so nobody can say which came first. What it establishes is that the shape of a rhythm carries information the amount of activity does not, which is the case the oscillation page makes at length.

Helplessness is the default state; control is what gets learned

Prediction is the third foundation carrying depression's signature: how far a system is acting on its own model of the world rather than on the world. It has a page of its own here. The best story psychology has about it was told backwards for fifty years, then corrected by the people who told it.

In 1967 two young researchers at the University of Pennsylvania, Martin Seligman and Steven Maier, were running conditioning experiments when the animals started behaving wrongly. Dogs that had first received shocks they could do nothing about later failed to escape shocks they could easily have escaped. Dogs given control the first time escaped normally. The experimenters concluded that the animals had learned that nothing they did mattered, and called it learned helplessness. It became one of the most influential animal models of depression ever proposed.

The circuit that registers having a say

By 2005 Maier's laboratory in Colorado could see where in the brain control was being registered. Two structures matter. The dorsal raphe is a small nucleus in the brainstem that supplies much of the brain with serotonin. The medial prefrontal cortex sits at the front, and among its jobs is deciding what a situation means.

Uncontrollable stress drove the dorsal raphe into overdrive, and that drove the passivity. When the same stress was controllable, the prefrontal region reached down and held the raphe quiet. Then came the experiment that settles it. Jose Amat and colleagues in Maier's psychology department silenced the prefrontal region with a drug, and controllable stress behaved exactly like uncontrollable stress. The animals were not registering the shock. They were registering whether they had any say in it.

Helplessness was never the thing being acquired. Control was.

In 2016, fifty years after the original experiment, Maier and Seligman published a paper saying they had read their own result backwards. Passivity under prolonged aversive events is not learned at all. It is the default, unlearned response, wired through the dorsal raphe. What an animal learns, when it gets the chance, is that it has control, and that learning runs through the prefrontal cortex and switches the default off.

This is animal neuroscience, and the authors label the clinical implications as speculation. The shape is what carries over. A default state, and a circuit that inhibits it once action is seen to change outcome, describes a regulator and the value it defends.

That is a very different sentence from the one the original experiment produced. The circuit that registers having a say has gone quiet. A quiet circuit is not a choice anybody made.

Dopamine carries surprise, and in depression the surprise signal is contested

The other half of the prediction story explains the symptom people find hardest to put into words.

Anhedonia is the loss of pleasure, and the standard account used to be that dopamine is the pleasure chemical and depressed people have less of it. That account died in a monkey laboratory. Wolfram Schultz, a neurophysiologist, was recording from single dopamine cells with fine electrodes while monkeys learned to expect a drop of juice.

The cells fired hard when the juice arrived unexpectedly. When the same juice arrived exactly as predicted, they did not fire at all. When a predicted juice failed to arrive, they fell silent below their own baseline.

That makes no sense for a pleasure signal, because the juice tasted the same every time. Working with two computational scientists, Schultz showed the cells were reporting the difference between the reward received and the reward predicted. Dopamine carries surprise, and surprise is how a model of the world gets updated. A world that keeps matching your expectations produces no signal at all, because there is nothing left to learn.

Dopamine reports the size of the correction the brain has to make to its own forecast, whatever the juice tastes like.

Gradin found blunted prediction errors in depression and Rutledge found them intact

Then the prediction that account makes about anhedonia. If dopamine carries surprise, a person who cannot feel pleasure should show a blunted surprise signal. A neuroscience group at Dundee led by Victoria Gradin scanned 15 depressed patients, 14 patients with schizophrenia and 17 controls during a reward-learning task. A computational model of expectation was fitted to each person's brain activity moment by moment.

In depression the signal was blunted in two places. One was the midbrain, the small structure at the top of the brainstem where the dopamine cells Schultz recorded from sit. The other was the striatum, a deep structure that handles reward. The more blunted the signal, the more severe that person's anhedonia. In schizophrenia the blunting appeared in different regions. Fifteen patients per group, and the result is a correlation.

A larger study using a different task did not find it. The computational neuroscientist Robb Rutledge and colleagues in London scanned 32 people with moderate major depression and 20 controls, using a task built to remove the learning-related confounds of the earlier work. Prediction-error signals in the ventral striatum were intact and did not differ significantly from controls. The effect of those signals on moment-to-moment happiness was intact too. Their conclusion was that the neural and emotional impact of reward prediction errors survives major depression.

So the blunting finding sits in the same contested category as serotonin and exercise. The disagreement is also readable. One study measured the signal during learning and the other deliberately took learning out of the task, which is a difference in what was asked of the system. A signal can be generated intact and still fail to move anything downstream of itself, which is what the London authors proposed.

The psychologists Michael Treadway and David Zald supply the sentence that makes this humane. Anhedonia had been used for two different things: not enjoying what you get, and not being willing to work for it. The animal evidence ties dopamine to the second. Many depressed people can still feel pleasure once it arrives. What has gone is the willingness to spend effort to reach it, which is a different clinical problem with a different explanation.

Put the two halves together and prediction has a mechanism under it. A brain whose forecasts stop being corrected by good news will keep producing the forecast that produced the mood. Where that correction fails is exactly what the two scanner studies disagree about.

Depression changes how the brain reads the body

Load and prediction meet in one place, and that place has an address in the brain.

An American neuroanatomist spent a career tracing the pathway that carries the body's own condition upward. Temperature, ache, itch, the burn in a working muscle, the state of the gut, the feeling of the heart. That stream, which physiology calls interoception, arrives in a fold of cortex called the insula.

As it travels forward through that fold it is rebuilt at each step into something more integrated, and what emerges is the felt sense of how you are. His claim, which remains contested, is that all subjective feeling is built on that rebuilding. The fidelity of the incoming signal has its own page here, at input quality.

Twenty unmedicated depressed adults and twenty healthy ones were asked, inside a scanner, to attend to sensations from their own gut, heart and bladder. The neuroscientist Jason Avery and colleagues in Tulsa found a specific strip of insula less active during that attention in the depressed group. The quieter it was, the worse the depression, and the worse that person's physical symptoms. Attending to the body is what that strip is for, and in depression it was doing less of it.

The same region behaved differently at rest. It was more tightly coupled to the amygdala and to the subgenual prefrontal cortex, two regions heavily involved in threat and mood. The tighter that coupling, the more severe the depression. A region that should be reading the body was reading it less and talking to the threat machinery more. Twenty per group, and cross-sectional, so the altered reading could be cause, consequence or neither.

The heartbeat-counting literature

The behavioral version of the question is less settled still. The usual test asks someone to count their own heartbeats over a timed interval without taking a pulse, then scores how close the count came. A German systematic review led by Michael Eggart screened 389 records and found only six studies that had done it properly in major depression.

The pattern was not even a straight line. Moderately depressed people were the least accurate, and accuracy appeared to return toward normal in severe depression. A curve that goes down and then back up is either a real non-linearity or an artifact, and six studies cannot say which. The results are confounded by sex, anxiety, panic disorder and antidepressant use. The task itself has been criticized for measuring what people believe about their heart rate as much as what they perceive. That literature is unsettled.

Two theoretical accounts join the body to the mood, and neither reports a measurement. One reads depression as a disorder of the body's energy budget, so that a brain locked onto its own model keeps reporting a deficit the senses cannot correct.

The other adds a second step, in which a brain whose predictions about the body keep failing draws a conclusion from the repeated failure, that it cannot regulate its own body. Its authors call that belief low allostatic self-efficacy. They are cited because mainstream computational psychiatry arrived independently at the shape this model describes. A system whose information about itself has degraded will run on its model instead.

Exercise moves depression, and the trials disagree by a factor of six

Exercise is the most studied non-drug input in depression, and the literature disagrees with itself by a factor of six. That disagreement is more informative than either number.

The pooling measure is the standardized mean difference, which is the same effect-size scale applied across trials that used different questionnaires.

0.18, 0.62, or 1.11

Cochrane reviewers pooled 35 trials of exercise against no treatment or a control condition, 1,356 participants in all, and found a standardized mean difference of 0.62 favoring exercise. Then they did the thing that makes Cochrane reviews uncomfortable reading.

They repeated the analysis using only the six trials that had concealed which arm a participant was heading into, analyzed everyone they randomized, and used assessors who did not know who had exercised. In those six trials, 464 people, the effect was 0.18 and not statistically significant.

The exercise scientist Felipe Schuch and colleagues looked at the same literature and reached the opposite conclusion. Applying a statistical correction for missing studies to 25 randomized trials, they argued that publication bias had been understating the effect and reported an adjusted standardized mean difference of 1.11. Two competent teams read one body of evidence and came out six times apart.

The largest attempt to settle it gathered 218 trials, 495 arms and 14,170 participants. The psychologist Michael Noetel and a team ranked exercise types against each other instead of lumping them together. Walking or jogging, yoga, strength training, mixed aerobic exercise and tai chi all beat active controls, with the effect scaling with prescribed intensity.

They also reported that only one of the 218 studies met the criteria for low risk of bias, and graded their own confidence as low for the best-supported type and very low for the rest.

That intensity finding is worth a sentence of reading, because on its face it says more is better, and the model says dose curves in living systems are not straight.

A dose can fall below the level at which a system registers it, land inside the window where it integrates, or exceed what it can tolerate and provoke a defense. A curve averaged across people who were never sorted reports where most of them happened to be sitting, which in an untrained depressed sample is below the window.

Nobody can be blinded to whether they are exercising. That single fact sets a ceiling on how clean this literature can ever be.

Direction of cause, and what the spread can and cannot show

The psychological analogue is behavioral activation: schedule the activities the depression has stopped, and do them whether or not you feel like it. A trial run from the Exeter medical school and led by David Richards randomized 440 adults to behavioral activation delivered by junior mental health workers, or to full cognitive behavioral therapy from psychologists. At twelve months the simpler treatment performed no worse than the more expensive one, which is what a non-inferiority trial sets out to test.

Then the causal direction, because active people are less depressed and depression also stops people moving. The epidemiologist Karmel Choi and colleagues used inherited variants across 611,583 adults as a natural experiment. Variants predicting higher accelerometer-measured activity predicted lower odds of major depression, the odds falling by about a quarter for each standard deviation of activity. A standard deviation is roughly the ordinary spread across a population. Variants predicting depression did not predict lower activity, and self-reported activity showed nothing in either direction.

Part of the gap between 0.18 and 1.11 is expectancy. The model reads the remainder as what happens when a trial gives everyone the same prescribed dose of the same modality across a sample that was never sorted by anything. A well-matched input and a mismatched one get averaged into a mean describing neither.

That reading has to be offered carefully, because the spread of outcomes inside an exercise trial is not by itself evidence for it. Measure anyone on a bad day and the next measurement will usually look better whether or not anything was done, because an unusually extreme reading is partly chance, and chance does not repeat.

Statisticians call that regression to the mean. It manufactures the appearance of responders and non-responders in data where no real difference between them exists, and ordinary measurement noise within a person does the same. The claim here rests on sorting people before a trial rather than on the spread found afterwards.

What this literature does supply is that exercise types rank differently, and that measured activity carries a signal reported activity does not. Movement has its own page here, at movement.

The range collapsed to one setting: depression as tone

Everything above is established science. It is also divided among psychiatry, endocrinology, immunology, cardiology and sleep medicine, and no clinic sees all five readings on one patient. The Unified Model of Tone reads them as one thing.

Tone is the integrated organization of the body's interacting state, the way its mechanical, electrical, chemical, fluid and neural processes are related at a given instant. It is one bound state rather than a list of parts. Oscillation is the carrier, in the way notes carry a chord. The chord is the relationship among them, and the chord is what a body actually reads.

Health, on this account, is the range. A healthy system can mount a cortisol response and switch it off. It can inflame and resolve. It can go deeply asleep and come fully awake. It can be surprised by good news and revise its forecast. The breadth of that range is what health is. Tone that has drifted or narrowed out of that range, holding one setting whatever the moment asks, is what manifests as illness, and in this condition it manifests as depression.

This reading applies once the findable drivers have been looked for. A thyroid, a new medication, or a late-onset organic cause is a different problem, and it is treated on its own terms.

Depression is that range collapsed into one costly setting the system can no longer leave. The parts are not failing, because a setting is not a part.

Load, oscillation and prediction do most of the work here, and each earns its place from a specific result. Load is the reading of a cortisol curve that rises normally and then will not come down, and of inflammatory signaling with no off switch in a quarter of patients.

Both say the same thing in two mediums: a continuous demand answered by machinery built for episodes. Oscillation is the reading of a vagal brake that runs low, a first REM period arriving too early, and a day-and-night amplitude that has flattened. Prediction is the reading of a forecast that no longer updates on good news.

Why depression has no lesion

Take the hardest puzzle the category presents and run it through that. Depression has no lesion, which for a century has been read as a gap in the instruments. The model reads it as a fact about what kind of thing depression is. Every instrument medicine owns was built to photograph a part: a slice of tissue, a blood value, a signal from one region.

An organization across coupled systems has no photograph, because it is not sitting anywhere. Two people can carry opposite cortisol curves, opposite appetite changes and opposite immunometabolic profiles and arrive at the same word, because the word was never a pathway. That is why a century of looking for the lesion has not found one, and why it will not end by finding one.

Why so many unlike inputs move it

The second puzzle is why so many unlike things move this condition. Medication reaches it through receptor coupling. Light works on the daily clock. Exercise arrives as metabolic and proprioceptive input. Behavioral activation and psychotherapy work on prediction and meaning. Sleep manipulation moves the rhythm itself.

Anti-inflammatory agents act on immune signaling, and the one prospective test of that route in a pre-selected inflamed sample came back empty. A vagus nerve stimulator, a device implanted under the collarbone with a lead on the vagus nerve in the neck, delivers electrical pulses on a fixed cycle to the same brake described above.

A loop with many access points can be interrupted at any of them, which is why inputs with nothing in common all report effects. It is also why none of them works for everybody. The doorway has to match the person, and no doorway on that list is the preferred one.

The same logic covers what starts a depression. Infection starts some. Grief starts some. Chronic insomnia roughly doubles the risk. High-dose interferon produced a major depression inside twelve weeks in nearly half of the patients who received no antidepressant alongside it. Those four have nothing in common as substances or events. They have one thing in common as inputs. Each arrives at an organization already holding a particular shape, and the shape decides what the input becomes.

Not a personal failing

One thing has to be said plainly, because the model is easy to misread here. None of this makes depression a personal failing or a shortage of effort. The animal work is the clearest evidence on that point, since the passivity is a default state rather than a lesson somebody learned. A system whose regulation has narrowed did not choose the narrowing, and trying harder is not the variable that widens it.

Restoring a regulator against moving a number

Relief has real value, and both aims here are legitimate.

A person who can sleep, work and stay alive because of a medication has been given something worth having, and the model grants that without qualification. Surgery is the same principle at a larger magnitude: change a constraint directly, because the system cannot. The distinction is about what an input is aiming at.

That line runs across the professions rather than between them. Every field contains both aims. A conversation can be delivered as reassurance that moves a number for an afternoon, or as the thing that changes how a person predicts.

A medication can mask, and a medication can hold a system steady long enough for its own regulation to return. Light timed to where a particular person's rhythm actually sits is aimed at a regulator. The same light box handed to everyone with the diagnosis on a fixed schedule is aiming at the number again.

When symptom recovery and regulatory recovery come apart

Depression supplies an unusually clean illustration, and it comes from the cardiac work above. In one meta-analysis the newer antidepressants left heart rate variability unchanged in patients whose symptoms had resolved. Symptom recovery and regulatory recovery came apart in the same people. That is the claim being used here, and it is separate from the unsettled argument about which drug classes lower variability.

That gap is also how the model reads a relapse. A well-executed input that resolves a complaint which then returns unchanged is not a failed input. It is a signal that the organization it was aimed at sits one layer further back than where it was delivered.

The presenting pattern in a long depression usually rides on a deeper one, and the rising relapse rate across successive STAR*D steps is what that looks like from here. The discipline that goes with the claim is strict. A clinician who insists every case is secretly one thing has relocated the first layer rather than found the deepest one.

The same reading applies in the other direction, which is what keeps the claim symmetrical. An input that moved variability, cortisol recovery and rhythm amplitude while the mood stayed exactly where it was would be an equally incomplete result, and the model would have to say so.

What follows practically is a matter for a clinician who knows the person. What follows scientifically is a test.

The trial design that would settle the tone reading of depression

The model's sharpest claim is a prediction that can fail, and depression is a good place to run it.

These patients are not uniform on the measurable variables, even though they are uniform on the complaint. Cortisol output runs high in one clinical group and lower in another. Appetite runs in both directions, and the two directions carry opposite immunometabolic profiles. Sleep goes both up and down inside the same category. Same diagnosis, opposite deviations.

So the prediction is stated in advance. Take a mixed group of depressed people, measure a regulatory variable before anything is done, and specify from that measure what each person receives. If a genuine restoration of tone is what gets delivered, the high values should come down and the low values should come up.

The group average should barely move while the spread around the middle narrows. A one-directional agent cannot produce that pattern. Neither can regression to the mean, which does not produce coordinated movement across several coupled measures in step with symptom change. A sham arm is not optional, because sorting people by starting side produces some convergence on its own.

No trial has been designed that way, because nobody stratifies by a regulatory measure recorded before the outcome is known. The nearest thing to a test came from the other direction and did not go well.

Sorting after the fact, infliximab beat placebo above a CRP of 5 mg/L and lost to it below. Sorting in advance on inflammatory activation, the same antibody did nothing. What is claimed here is a measure of regulation rather than a marker level, and that version of the claim remains untested.

Three results that cut against the reading

Three results cut against the reading and belong in the body rather than in a footnote. The severity moderation is contradicted by a later pooled analysis of individual patients, which found no interaction between starting severity and treatment. The causal test of the inflammation story returned a null, with genetic variants that raise inflammation failing to consistently raise depressive symptoms.

And a sham-controlled trial of implanted vagus nerve stimulation in treatment-resistant depression missed significance at ten weeks, with its authors saying plainly that the trial did not yield definitive evidence of short-term efficacy. That last one bites here specifically. The device stimulates the same nerve the vagal argument above is built on, which means driving a regulator is not the same operation as restoring one. What the device evidence does and does not show is worked through on the mental health page.

What would have to show up in the data

Start with the discrimination that matters most in treatment. A measure of regulatory state recorded before any input should carry information about who responds to what, and a measure that carries it confirms the reading here. An input that restores regulation moves people who start above the healthy band and people who start below it toward the same middle. An input that moves the whole sample one fixed direction is pushing the output, which is masking a symptom rather than restoring a regulator.

A second test takes up the larger claim, that these findings are one organization read at many sites. Cortisol recovery time, heart rate variability, day-and-night amplitude, reward prediction error and insular response to interoceptive attention would have to be recorded together in the same patients. When they load on a shared underlying factor, they are windows onto one variable rather than five separate faults sharing a name, and that shared factor confirms this reading.

A third is simpler. Symptoms that move in step with those measures establish the coupling claimed between the organization and the complaint. Each of these experiments is runnable with instruments that exist today.

Early warning in the dynamics

That leaves the measure, and its ingredients have already been named. The cortisol finding that matters is the failure to come down rather than the height of the peak. The cardiac finding that matters is the variability rather than the mean. The rhythm finding that matters is the amplitude rather than the amount of activity. Each is a dynamic measure, and the model predicts that dynamic measures move before static ones do, which is the case the time course page makes in full.

One study has already watched that happen, and it came from an unlikely direction. Ecologists know that a lake about to flip from clear to green recovers more slowly from small disturbances before it flips. The slowing appears in the data before the change appears in the water. The ecologist Ingrid van de Leemput and a Dutch team applied the same mathematics to mood, using thousands of momentary self-reports collected from healthy people and patients over months.

Three things changed before a shift into depression, and the same three changed before a shift out of it. Each moment predicted the next more strongly than before, which is the slowing. The swings got wider. And the separate emotions moved together more tightly, which is coupling rising as a system loses its independent degrees of freedom. The dynamics moved before the state did, in both directions.

That is what a system approaching a transition looks like, and it is measurable today with a phone and a questionnaire. Someone could be handed a warning before an episode arrives rather than a diagnosis after it. The samples are modest, and no prospective clinical validation exists yet.

How depression relates to the rest of the library

Depression is a condition page, and the readings it rests on each have a page of their own. The claims below are the specific connections, one per page.

• Tone is the pillar this page reads through: the integrated organization the body holds across its systems at once, and the range that organization can move through.
• Load carries the cortisol curve that will not come down and the inflammatory signaling that will not resolve; depression is one of its clearest clinical expressions.
• Oscillation owns the rhythm findings here, from the beat-to-beat brake to the flattened day-and-night amplitude on 91,105 wrists.
• Prediction explains why a brain whose forecasts stop updating on good news keeps producing the mood its model expects.

• Mental health holds the wider psychiatric picture, including what the vagus nerve stimulation evidence does and does not show.
• Anxiety is the condition that most often travels with depression, and its biology runs high where depression's runs flat.
• Inflammation works through the sickness-behavior loop that makes a quarter of depressed patients an immunological story.
• Insomnia roughly doubles the risk of a later depression, which makes it a leading input rather than a side effect.
• Sleep holds the architecture the early REM finding comes from.
• Heart rate variability is the instrument behind the vagal brake findings, and its page explains what the number can and cannot certify.
• Movement carries the exercise literature this page compressed into one disagreement.
• Input quality is the fidelity of the body's self-report, the variable the insula findings put on the table.
• And time course explains why an entrenched depression answers inputs differently than a fresh one, which is the reading behind the rising relapse rates across STAR*D steps.

Frequently asked