Part 2: The Inverted-U Is Not One Unique Thing — Four Ways a Drug Can Stop Working When You Raise the Dose
If you have ME/CFS, you know the experience: a drug helps at a low dose, and then your doctor — following the logic that has governed medicine since Paracelsus — escalates. “Let’s see if we can get more benefit.” You don’t. You get worse. You go back to the low dose. The drug works again.
This pattern — the inverted-U, the biphasic curve, the hormetic window — has been documented for at least eighteen medications in ME/CFS: LDN, LDA, modafinil, duloxetine, beta-blockers, guanfacine, gabapentinoids, rapamycin, corticosteroids, DORAs, H1 antihistamines, allopregnanolone, NAC, ketotifen, quercetin, sulforaphane, lithium, melatonin. The list is long enough that coincidence is implausible. Something systematic is happening.
The temptation is to unify them under a single explanation — “hormesis,” the idea that low-dose stress triggers compensatory adaptive upregulation. That explanation is correct for some of them. It is wrong for most of them.
The eighteen drugs share a curve shape. They do not share a mechanism. The inverted-U in ME/CFS is at least four pharmacologically distinct phenomena that happen to produce the same dose-response pattern. Collapsing them into one concept obscures the clinical implications, which differ by category.
1 Category 1: Nrf2-mediated hormesis (the real hormesis)
Drugs: LDN (at low dose), low-dose lithium, melatonin, sulforaphane, quercetin. NAC is listed here as a concurrent Nrf2 activator but also classified under Category 4 (subtype 4c) for its biphasic thiol biochemistry — it may operate through both mechanisms at different dose ranges.
What it is. Hormesis in the strict pharmacological sense: a low-dose stressor triggers compensatory upregulation of protective pathways, and that upregulation — not the drug itself — produces the benefit. The drug is the signal. The cell’s response is the treatment.
The mechanism. Mild oxidative or inflammatory stress activates the Keap1-Nrf2-ARE pathway. Under basal conditions, Keap1 binds Nrf2 and targets it for degradation. When stress signals accumulate — modest TLR4 blockade, mild ROS elevation, partial enzyme inhibition — Keap1 releases Nrf2, which translocates to the nucleus and activates transcription of over 200 genes: antioxidant enzymes, anti-inflammatory proteins, mitochondrial biogenesis factors, proteasome subunits, autophagy components. The cell upgrades its entire stress-response infrastructure because it detected a threat. This is the principle behind exercise in healthy people: exercise-induced ROS triggers Nrf2-mediated mitochondrial biogenesis, making you fitter. The stress is the signal for adaptation.
Why the low dose is essential. Too much stress and the adaptive response is overwhelmed — the cell switches from compensation to damage control. Too little stress and Keap1 doesn’t release Nrf2 — the signal never fires. The therapeutic window is the Goldilocks zone where stress is sufficient to trigger Nrf2 but insufficient to cause damage.
Why higher doses make you worse. For LDN specifically: at 0.5–1.5 mg, partial TLR4 blockade triggers Nrf2-mediated M1→M2 microglial shift. At 3.0 mg and above, TLR4 is blocked too completely — the cell no longer detects the stress signal, Nrf2 stays bound to Keap1, and the anti-inflammatory program collapses. The drug is still binding the receptor — it just stopped producing the adaptive response that made it therapeutic. Note that this is a failure-to-trigger (too little stress signal) rather than an overwhelmed response (too much stress). The hormetic curve has an inverted-U for opposite reasons on each arm: the left arm fails from insufficient signal, the right arm fails from excessive damage that overwhelms repair capacity. LDN at higher doses is on the left arm.
The clinical implication. Nrf2 capacity varies between patients. A patient who loses LDN benefit at 1.0 mg has narrow Nrf2 reserve — their cells cannot sustain the compensatory program beyond a narrow stress window. A patient who maintains benefit to 3.0 mg has broad Nrf2 reserve. The width of the hormetic window is not a dosing artefact — it is a readout of the patient’s stress-response capacity. It predicts response to every other Nrf2-activating drug in the cluster.
Why hormetic thresholds differ between individuals. The position and width of the hormetic window are not fixed properties of the drug — they are properties of the organism. Three determinants modulate where a given patient’s hormetic curve sits on the dose axis:
Redox state. The Keap1-Nrf2 sensor is itself redox-sensitive. In a cell with high basal oxidative stress, Keap1 cysteines are already partially oxidised — the sensor is pre-loaded, and a smaller perturbation triggers Nrf2 release. In a cell with low basal oxidative stress, the sensor is quiescent, and a larger perturbation is needed. This means the same micro-dose of LDN can trigger Nrf2 in a patient with high oxidative tone and fail to trigger it in a patient with low oxidative tone.
Allostatic load. The cumulative burden of chronic stress — cytokine exposure, sleep disruption, metabolic strain, pain signalling — depletes the transcriptional machinery that Nrf2 needs to execute its program. Nrf2 can translocate to the nucleus, but if the cell’s biosynthetic capacity is exhausted, the anti-inflammatory gene products don’t get made. High allostatic load narrows the hormetic window from both sides: it raises the threshold for triggering Nrf2 (because the sensor is desensitised) and lowers the ceiling for the adaptive response (because downstream resources are depleted).
Baseline cellular reserve. Mitochondrial density, NAD⁺ availability, proteasome capacity, and sirtuin activity all influence whether a cell can sustain an Nrf2-mediated transcriptional program. Low reserve means the adaptive response is fragile — even a correctly triggered Nrf2 signal may fizzle because the cell lacks the energy or machinery to execute it. This is why the same drug at the same dose produces a therapeutic lift in one patient and a stress-like flare in another: the drug is identical, but the cellular infrastructure receiving the signal is not.
This means the hormetic curve is not a pharmacological constant. It is a systems-biology readout of the patient’s cellular condition at the time of dosing. A patient’s hormetic window can widen or narrow over the course of their illness — it is dynamic, not static. The clinical task is not to find a single permanent dose but to titrate against a moving physiological target.
The surprise in this category — melatonin. Melatonin is the drug where the “more is better” assumption fails in the least obvious way, because the drug has two effects that respond completely differently to dose. The first is the chronobiotic signal — the message to the brain’s clock that day is ending and sleep should begin. This signal does not get stronger with a higher dose. It is a phase cue, not a sedative: once it has told the clock the time, more of the molecule adds nothing to that message. The second effect — anti-inflammatory and antioxidant action through MT1/MT2 receptors that feed into the Nrf2 system — is hormetic. It works at physiological doses (roughly 0.3–1 mg) and inverts as the dose rises into the supraphysiological range, because sustained high occupancy desensitises the receptors that carried the protective signal.
The trap is that the two effects are invisible to the person taking the pill. Someone who sleeps a little on 0.3 mg reasonably assumes 3 mg will make them sleep more — but the sleep-onset benefit they already have was the chronobiotic phase cue, which does not scale, and the anti-inflammatory benefit they did not notice is quietly being extinguished by receptor desensitisation. What looks like “taking more of the same helpful drug” is actually two changes at once: the scaling effect that was never there to begin with, plus the loss of a hormetic benefit that was real. This is why a patient can report that melatonin “stopped working” or started causing a daytime grogginess “hangover” and depression-like mood at 3–5 mg. Those are signatures of sustained supraphysiological exposure — the hangover from a prolonged high plasma level, and the depression-like flattening that comes from a high dose suppressing daytime dopaminergic tone. The drug has not “worn off.” The dose was not protecting a benefit; it was erasing it.
2 Category 2: Partial-agonist inverted-U
Drugs: LDA/aripiprazole (single member).
What it is. A receptor-occupancy phenomenon. The drug activates the receptor with lower efficacy than the natural ligand. At low occupancy, it supplements the deficient endogenous signal — net agonism. At high occupancy, it displaces the more effective natural ligand — net antagonism.
The mechanism. Aripiprazole has approximately 25% intrinsic activity at D2 receptors. Dopamine has 100% intrinsic activity — it is the full agonist. When few receptors are occupied by aripiprazole, the net effect on the system is additional activation: the partial agonist adds to whatever endogenous dopamine is present, raising net tone. When most receptors are occupied, aripiprazole blocks dopamine from binding — and since aripiprazole activates the receptor only 25% as effectively as dopamine, net tone drops. The inversion point is approximately 50% occupancy.
Why it’s not hormesis. The cell is not mounting an adaptive response. There is no Nrf2, no transcriptional program, no compensatory upregulation. The curve is a direct, instantaneous consequence of receptor-ligand binding kinetics — a law-of-mass-action property, not a biological adaptation. It applies identically in a test tube, a healthy volunteer, and an ME/CFS patient. It is not specific to the disease.
The clinical implication. The inversion point reveals D2 receptor reserve. If benefit inverts at 1 mg, the patient has few functional D2 receptors — each one’s contribution to net tone is large, and blocking a modest fraction is enough to invert. If benefit persists to 2 mg, receptor reserve is broad. The diagnostic signal from LDA’s inverted-U is about dopamine system integrity, not about stress-response capacity.
3 Category 3: Catecholamine inverted-U at prefrontal cortex
Drugs: Modafinil, duloxetine, beta-blockers, guanfacine, gabapentinoids.
What it is. A canonical neuroscience principle: moderate catecholamine tone at prefrontal D1 and α2A receptors optimises signal-to-noise ratio in working memory circuits; excessive tone collapses network tuning.
The mechanism. Prefrontal cortex pyramidal neurons require a specific concentration range of dopamine and norepinephrine to maintain “tuned” firing — responding selectively to relevant inputs while suppressing noise. Below the optimum, signal is too weak; neurons don’t fire reliably. Above the optimum, D1 and α2A overactivation collapses the selectivity — neurons fire to everything, including noise, and working memory representations degrade.
This is not a pharmacological artefact. It is how the prefrontal cortex evolved to operate. The inverted-U is the native operating curve of the system. Drugs that raise catecholamine levels — DAT inhibitors (modafinil), NET inhibitors (duloxetine at higher doses), α2A agonists (guanfacine) — push the system along its existing curve. Too little drug and you’re below the optimum. Too much and you overshoot.
Why it’s not hormesis. No stress signal. No compensatory upregulation. The neuron’s response to dopamine is the same at every concentration — it’s the circuit-level computation that inverts. The mechanism is network dynamics, not cellular adaptation.
The clinical implication. The inverted-U identifies the patient’s baseline catecholamine tone. A patient who loses modafinil benefit at 50 mg has low baseline dopamine — a small DAT blockade pushes them past the optimum. A patient who benefits to 200 mg and beyond has high baseline tone — it takes substantial additional dopamine to overshoot. The inversion point is a functional assay of prefrontal catecholamine status, and it predicts the optimal dose for every drug in the cluster.
The surprise in this category — duloxetine. Duloxetine quietly contains two different drugs, and the dose separates them. It blocks the serotonin transporter (SERT) at low occupancy and the norepinephrine transporter (NET) only once occupancy rises. These two actions have opposite dose-response shapes. The serotonergic action — the pain, sleep, and anxiety relief — is monotonic: it saturates and stays, with no inverted-U. The noradrenergic action — the energy, motivation, and drive — is a true catecholamine inverted-U: it climbs into benefit as NET engages and then collapses as norepinephrine overshoots the prefrontal optimum.
The consequence is the most disorienting drug behaviour in the group: a patient can raise the dose and lose the energy benefit while keeping the pain relief, and interpret that as the drug “stopping working” — when in fact half of it is working exactly as well as before. The pain signal is serotonergic and never inverted; the energy signal crossed its own peak. Disentangling the two is itself diagnostic. If raising the dose removes the energy lift but spares pain relief, the norepinephrine system is sitting on the right arm of its curve — the correct move is to drop back to the dose that held energy benefit, not to conclude the drug has failed, and not to abandon it, because the pain relief was never the part that turned. The two mechanisms can be separated by dose because they occupy different transporters at different concentrations — a built-in within-patient experiment no single drug class usually offers.
The surprise in this category — beta-blockers. Every drug so far raises catecholamine tone and pushes the system toward its peak from the left. Beta-blockers are the exception that completes the picture: they lower catecholamine action, pushing from the right arm toward the peak. Guanfacine does this too, by quieting the locus coeruleus. So the category contains drugs that move tone up and drugs that move it down, and both directions can be therapeutic.
The surprise is what a patient who benefits from both directions reveals. If a low-dose elevator (modafinil) helps you and a low-dose reducer (a beta-blocker, or guanfacine) helps you, you have not contradicted yourself — you have exposed that your baseline catecholamine tone is usable across a wide band, because both a small addition and a small subtraction land inside the tolerable range. That wide peak is itself information: it means your catecholamine system is not the brittle bottleneck. The converse is the diagnostic edge case. A patient who is helped by modafinil but cognitively dulled by even a low-dose beta-blocker is sitting on the left arm with genuinely low baseline tone. For them, the reducer is not a neutral experiment: it pushes an already-low system below the floor, and combining beta-blockers with guanfacine would be actively counterproductive. The direction of the drug you respond to is a probe of where you sit on the curve, and the category’s real power is that it tests both directions at once.
4 Category 4: Concentration-dependent target selection and biphasic biochemistry
Drugs: Rapamycin, corticosteroids, allopregnanolone, DORAs, H1 antihistamines, NAC, ketotifen, quercetin.
What it is. Three distinct subtypes share a common consequence — benefit at low concentration, harm or loss of benefit at higher concentration — but arise from different mechanisms. (4a) Two-target selectivity: the drug has a high-affinity therapeutic target and a low-affinity target with opposing effect. At low concentration, only the therapeutic target engages. At higher concentration, spillover to the second target inverts benefit. (4b) Pharmacokinetic threshold: the drug’s therapeutic target is peripheral; at higher doses, CNS penetration engages brain targets that produce sedation or impairment. (4c) Biphasic biochemistry: the drug’s metabolic pathway has opposing net effects at different concentrations.
Rapamycin (mTORC1 vs mTORC2) — subtype 4a. mTOR exists in two complexes. mTORC1 regulates autophagy, protein synthesis, and metabolism. mTORC2 regulates cell survival, cytoskeleton, and insulin signalling. Rapamycin binds mTORC1 with higher affinity than mTORC2. At low intermittent doses (1–3 mg/week), rapamycin partially inhibits mTORC1 — autophagy is restored, mitochondrial quality control improves, and senescent-cell SASP (the inflammatory secretion profile of ageing or damaged cells) is suppressed. At higher daily dosing (5–10 mg/day), rapamycin also inhibits mTORC2 — insulin resistance develops, immunosuppression occurs, and the metabolic benefit inverts. This is not hormesis. It is differential binding affinity across two structurally related complexes.
Corticosteroids — subtype 4a. Physiological replacement (5–10 mg prednisone) provides the anti-inflammatory signal the HPA axis normally supplies. Supraphysiological doses suppress the HPA axis itself — the body stops producing its own cortisol, and the taper produces rebound inflammation worse than baseline.
The surprise in this category — corticosteroids. This is the most dangerous surprise in the entire catalogue, and the one where the dose curve is most easily misread. Corticosteroids are not a single drug with a dose-response curve. They are two drugs separated by a threshold, and crossing that threshold inverts not just the benefit but the physiology you were trying to help.
At a physiological dose, a corticosteroid is not really a treatment — it is a replacement. In a patient whose own HPA axis produces too little cortisol (a state worth suspecting in ME/CFS, though not assumed), low-dose hydrocortisone or prednisone simply supplies the anti-inflammatory glucocorticoid signal the body should have been making. The benefit comes from restoring a deficit. This is why a dramatic response at replacement dose is itself a diagnostic prompt to run an ACTH/cortisol workup — the drug only produces a striking effect if something endogenous was genuinely missing.
Above that physiological window, the drug stops being a replacement and becomes a suppressor of the very axis that makes the hormone. Supraphysiological dosing tells the pituitary and adrenal glands to stand down — the body reduces its own cortisol production to match what it now receives from outside. The problem is what happens on the way down. Because the axis has been silenced, stopping or tapering the drug removes the outside cortisol before the body has resumed making its own — leaving a window with no cortisol at all. The rebound inflammation that follows is frequently worse than the original illness, and it is not a sign the drug stopped working; it is the withdrawal of a hormone the body had been told not to make. This is why corticosteroids must be tapered, never stopped abruptly, and why the taper can be difficult even at modest doses if the axis has been suppressed.
The clinical rule this yields is counterintuitive and worth stating plainly: with corticosteroids, the dose that helps is the smallest one that restores a deficit, and raising it past the replacement window does not add benefit — it converts the drug into something that endangers the system it was supporting. The “more is better” assumption is not merely ineffective here; it is the specific mechanism by which a helpful replacement becomes a harmful suppressor with a rebound trap on the way out. Because of the real harm potential, any corticosteroid use in ME/CFS — at any dose, replacement or higher — belongs under a clinician’s guidance, and the surprise being described is a caution about mechanism, not a recommendation to try it.
DORAs and H1 antihistamines — subtype 4b (pharmacokinetic threshold). DORAs (dual orexin receptor antagonists): partial orexin blockade at low dose improves sleep; complete blockade at higher dose causes sleep paralysis and daytime hypersomnia — the drug overshoots the sleep-promoting window. H1 antihistamines (e.g., cetirizine): peripheral H1 blockade at 5–10 mg relieves MCAS symptoms; CNS penetration at ≥20 mg produces sedation — the drug crosses the BBB and engages a brain target that was not contributing to peripheral benefit. The inversion is a concentration-gradient phenomenon, not a receptor-level mechanism.
Allopregnanolone (GABA-A biphasic) — subtype 4c. Allopregnanolone is a positive allosteric modulator of GABA-A receptors — it enhances the effect of GABA. At low concentrations (1–5 nM, typical of the luteal phase), it paradoxically increases anxiety in some women. At higher concentrations (>10 nM, pregnancy-level), it produces sedation and anxiolysis. The mechanism likely involves two binding sites on the GABA-A receptor with opposing concentration-response curves. Not hormesis, not network dynamics — two binding sites on the same protein.
NAC — subtype 4c. At low dose (600 mg), NAC can produce paradoxical worsening — the thiol-radical burst from cysteine oxidation exhausts glutathione before net synthesis kicks in. At higher dose (1,200 mg), the synthetic pathway saturates and net glutathione rises. The inversion is at the level of cysteine biochemistry, not receptor pharmacology. NAC also appears in Category 1 as an Nrf2 activator — it may operate through both mechanisms at different dose ranges.
The surprise in this category — NAC. NAC is the one drug in the whole catalogue where the curve runs backwards from intuition, and it is the drug patients are most likely to abandon on the strength of a misleading single trial. NAC is an antioxidant precursor: the body uses it to make glutathione, the cell’s main redox buffer. The mistake is to think of NAC itself as the antioxidant. It is not. It is raw material, and the first thing the body does with a load of it is process it — and that processing step itself produces a burst of reactive oxidant intermediates before the glutathione is built.
At 600 mg, that processing burst can exhaust existing glutathione faster than new glutathione is made — the cell is briefly more oxidatively stressed, not less. In roughly a fifth to a third of patients this shows up as a paradoxical worsening: flu-like symptoms, oxidative malaise, feeling worse within hours of the pill. The natural conclusion — “NAC makes me worse, it is not for me” — is wrong in a specific way. The worsening is not the drug’s steady-state effect; it is a transient overshoot during the synthesis ramp. At 1,200 mg, the synthetic pathway has saturated and net glutathione is rising, and many of the same patients who felt worse on 600 mg feel better on 1,200. The dose that hurts is lower than the dose that helps.
This is why a single failed NAC trial at 600 mg is not a verdict. The clinical moves that respect the curve are to start low enough to avoid the burst and build past it (splitting the dose, for example, so the body is never asked to process a large single bolus), or to recognise that a patient who worsens on 600 mg but improves on 1,200 mg has a glutathione pathway that is intact but capacity-limited — a finding that carries its own meaning, since it predicts that other Nrf2-cluster drugs working through Keap1 (sulforaphane) or GSK-3β (lithium) may succeed where a single low NAC dose seemed to fail.
Ketotifen — subtype 4b. MCAS benefit at low dose from mast-cell stabilisation; sedation above the H1-antihistamine threshold — dose-dependent CNS penetration inverts the net benefit.
Quercetin — subtype 4c. COMT inhibition at low dose can optimise prefrontal catecholamine tone (similar to Category 3); at higher doses, excessive COMT inhibition overshoots the catecholamine inverted-U and impairs cognition.
The clinical implication. These drugs do not share a common mechanism with each other, let alone with the other categories. Their inverted-U curves are drug-specific properties — the dose at which rapamycin loses mTORC1 selectivity tells you nothing about the dose at which modafinil overshoots prefrontal D1, and neither tells you anything about the dose at which LDN’s Nrf2 priming collapses. Within this category, the three subtypes carry different diagnostic value: subtype 4a (two-target selectivity) reveals receptor reserve and target affinity ratios; subtype 4b (pharmacokinetic threshold) reveals BBB integrity and CNS penetration sensitivity; subtype 4c (biphasic biochemistry) reveals metabolic pathway capacity. They are grouped here because they complete the catalogue — these are the remaining non-monotonic drugs in the paper — not because they cluster mechanistically.
5 Dose-band reference tables
The following tables give, for each of the eighteen medications, a row per dose band that has some activity — a range where the drug produces a benefit or a notable effect. All dose bands are the paper’s mechanism-derived hypotheses (low certainty, ~0.30), cross-checked against the standard clinical doses where they exist. No within-range prospective dose-response trial confirms any of them. Units are as given (mg oral dose unless stated; nM for concentrations).
5.1 Category 1 — Nrf2 hormesis drugs
LDN (low-dose naltrexone)
| Dose band | Activity / effect | Mechanism |
|---|---|---|
| 0.25–0.5 mg | Micro-dose; below the Nrf2 trigger threshold for most; transient sleep disruption | Opioid engagement detectable at micro-dose |
| 0.5–1.5 mg | TLR4/Nrf2 hormetic priming — anti-inflammatory M1→M2 microglial shift; the main window | Partial TLR4 blockade → Nrf2 upregulation |
| 1.5–3.0 mg | Endorphin-rebound plateau; benefit preserved | Brief overnight opioid blockade → compensatory endorphin production |
| 3.0–4.5 mg | TRPM3 restoration; orexin disinhibition (“wired”); TLR4 window closed | TRPM3 Ca²⁺ flux; orexin disinhibition tracks TLR4 dose-response |
| >4.5 mg | No therapeutic mechanism; all past optima | None therapeutic |
| 50 mg | Full mu-opioid antagonism — harm | Complete opioid blockade, no hormetic benefit |
Low-dose lithium
| Dose band | Activity / effect | Mechanism |
|---|---|---|
| 0.3–1.0 mg | Nrf2 activation; predicts broad LDN/sulforaphane windows | Partial GSK-3β inhibition derepresses Nrf2 |
| 1.0–5.0 mg | Standard hormetic range; ~2 mg proposed daily probe | GSK-3β inhibition; IMPase inhibition begins at upper end |
| 5–15 mg | Benefit only here → mechanism not Nrf2/GSK-3β; cognitive dulling, thyroid change | IMPase inhibition / inositol depletion |
| 900–1200 mg/day | Standard psychiatric mood stabilization; serum 0.6–1.2 mEq/L | Full GSK-3β + IMPase + neurotransmitter modulation |
Melatonin
| Dose band | Activity / effect | Mechanism |
|---|---|---|
| 0.1–0.3 mg | Reduced dose for CYP1A2 slow metabolizers | Avoid supraphysiological exposure |
| 0.3–1.0 mg | Physiological replacement; phase advance; 1 mg reduced physical fatigue in ME/CFS | MT1/MT2 Nrf2 activation; chronobiotic |
| 1.0–3.0 mg | Benefit above 1 mg → direct antioxidant effects; vivid dreams, headache | Direct radical scavenging + MT1/MT2 |
| 3.0–5.0 mg | Benefit only here → not Nrf2/MT; daytime sedation | Scavenging or sedation |
| 5–20 mg | No additional mechanism; window closed; depression-like symptoms | Suppression of daytime dopaminergic tone |
Sulforaphane
| Dose band | Activity / effect | Mechanism |
|---|---|---|
| 10–30 mg | Strongest, purest Nrf2 probe; highly sensitive | Direct Keap1 cysteine modification → Nrf2 |
| 30–60 mg | Standard Nrf2 reserve; sustained Keap1 modification; possible GI distress | Keap1 → Nrf2 |
| 60–100 mg | Benefit only at high dose → very high Nrf2 degradation rate; plateau; GI intolerance | Keap1 saturation |
Quercetin
| Dose band | Activity / effect | Mechanism |
|---|---|---|
| 250–500 mg | Cognitive focus via COMT inhibition; anti-inflammatory via Nrf2 | COMT inhibition + Nrf2 |
| 500–1000 mg | Both mechanisms contributing; 500 mg BID as mast-cell stabilizer | COMT + Nrf2 |
| 1000–2000 mg | No new benefit; dual-target inversion; hypertension, tachycardia | COMT overshoot + pro-oxidant catecholamine auto-oxidation |
NAC (N-acetylcysteine)
| Dose band | Activity / effect | Mechanism |
|---|---|---|
| 600 mg | Glutathione synthesis; Nrf2 activation — but ~20–30% paradoxical worsening | Thiol burst vs Nrf2 |
| 1200 mg | Net glutathione synthesis established; GI side effects → split | Glutathione + Nrf2 |
| 1800–2400 mg | No additional Nrf2 benefit (plateau); dose-limiting GI toxicity | Saturation |
5.2 Category 2 — Partial-agonist inverted-U
LDA (low-dose aripiprazole)
| Dose band | Activity / effect | Mechanism |
|---|---|---|
| 0.2–0.5 mg | Micro-dose; akathisia here = severe DA deficiency + supersensitive D2 | Microglial D2 partial agonism |
| 0.5–1.0 mg | Standard window: microglial D2 + mesocorticolimbic D2 + 5-HT1A | Partial D2 agonism; 5-HT1A |
| 1.0–2.0 mg | Cognition/motivation + ANS stabilization; inversion point if benefit lost here | Postsynaptic/mesocorticolimbic D2 |
| >2 mg (to ~4 mg) | All therapeutic mechanisms extinguished; net antagonism; sedation, weight gain | D2 occupancy >50% → net antagonism |
| 5–30 mg | Psychiatric range; active harm | D2 occupancy >80% full antagonism |
5.3 Category 3 — Catecholamine inverted-U at prefrontal cortex
Modafinil
| Dose band | Activity / effect | Mechanism |
|---|---|---|
| 25–50 mg | Cognitive gain in low-tone patients; transient headache | Mild DAT blockade on left arm of inverted-U |
| 50–100 mg | Standard optimal window for low-normal baseline DA | DAT blockade at D1; NET at α2A |
| 100–200 mg | FDA-approved dose (narcolepsy/OSA); peak benefit; inversion above 200 mg | DAT blockade; wake-promoting |
| 200–300 mg | Benefit only here → severely low baseline DA; anxiety, tachycardia | Far right of inverted-U; D1/α2A overactivation |
| ≥400 mg | No additional mechanism; catecholamine toxicity | Far right of inverted-U |
Duloxetine
| Dose band | Activity / effect | Mechanism |
|---|---|---|
| 10–20 mg | Predominantly serotonergic; pain/sleep benefit, not catecholamine; nausea | SERT inhibition |
| 20–40 mg | Dual SERT + beginning NET; energy/motivation added | SERT + NET |
| 40–60 mg | Both optimal; standard FDA dose (60 mg) | Full SERT + NET |
| 60–90 mg | Unlikely new benefit; noradrenergic excess → hypertension, agitation | NET past inverted-U peak |
| 90–120 mg | High-dose; dose-dependent BP rise | NET-driven BP elevation |
Guanfacine
| Dose band | Activity / effect | Mechanism |
|---|---|---|
| 0.5–1.0 mg | Optimal PFC α2A tone; working memory, focus; sedation first 1–2 weeks | Postsynaptic α2A agonism |
| 1.0–2.0 mg | Optimal cognitive benefit at α2A peak; somnolence, hypotension | Postsynaptic α2A at peak |
| 2.0–4.0 mg | No new cognitive benefit; anxiolysis/sedation; hypotension | Presynaptic α2A suppresses locus coeruleus |
Beta-blockers (propranolol reference; atenolol/metoprolol equivalent)
| Dose band | Activity / effect | Mechanism |
|---|---|---|
| Propranolol 5–10 mg | Orthostatic control without cognitive cost; mild fatigue | Peripheral β1/β2 blockade |
| Propranolol 10–20 mg | Orthostatic control + cognitive improvement; bradycardia, hypotension | Peripheral + early central β-blockade |
| Propranolol 20–40 mg | No new benefit; cognitive decline; fatigue, depression | Central β-AR pulls tone below optimum |
| Propranolol 40–160 mg/day | Standard antihypertensive dose; not the ME/CFS low-dose strategy | Peripheral β1 blockade |
Gabapentinoids (gabapentin / pregabalin)
| Dose band | Activity / effect | Mechanism |
|---|---|---|
| Gab 100–300 mg / Preg 25–75 mg | Cognitive clarity, reduced sensory hypersensitivity, improved sleep | α2δ subunit block |
| Gab 300–900 mg / Preg 75–150 mg | Symptom control optimized; “brain fog” if noradrenergic suppression excessive | α2δ block |
| Gab ~300 mg (single dose) | Worsened OSA (AHI ↑); side-effect band | GABAergic upper-airway collapsibility |
| Gab 900–3600 mg / Preg 150–600 mg | Standard neuropathic-pain range; severe sedation, dependence risk | α2δ block at full occupancy |
5.4 Category 4 — Target selection and biphasic biochemistry
Rapamycin
| Dose band | Activity / effect | Mechanism |
|---|---|---|
| 0.5–2 mg/week | Restores autophagy, suppresses SASP; predicts benefit from intermittent fasting | mTORC1 inhibition (high affinity) |
| 2–3 mg/week | mTORC1 benefit maximized; metabolic side effects begin | mTORC1 inhibition |
| 3–6 mg/week | mTORC1 benefit offset by mTORC2 metabolic costs; net negative | mTORC2 inhibition begins |
| ≥5 mg/day | No rationale; expected net harm | Full mTORC2 toxicity |
Corticosteroids (prednisone reference)
| Dose band | Activity / effect | Mechanism |
|---|---|---|
| Prednisone 2.5–5 mg/day | Physiological replacement; restores anti-inflammatory tone; dramatic response → rule out adrenal insufficiency | Physiological GR activation |
| Prednisone 5–15 mg/day | Benefit → significant steroid-responsive inflammation | Supraphysiological GR activation |
| Prednisone 15–40 mg/day | Benefit only here → corticosteroid-resistant inflammation; HPA suppression dominant | GR saturated; HPA suppression |
| ≥40 mg/day | No rationale in ME/CFS; net harm | Cushing’s syndrome, avascular necrosis |
Allopregnanolone
| Dose band | Activity / effect | Mechanism |
|---|---|---|
| ~1–5 nM (luteal level) | Paradoxical anxiogenic effect; skip this range | High-affinity GABA-A site alone |
| >10 nM (pregnancy level) | Anxiolysis, sedation; therapeutic | Both GABA-A sites engaged |
| Brexanolone IV 60–90 µg/kg/h | Proof-of-concept (postpartum depression) | GABA-A positive allosteric modulation |
DORAs (dual orexin receptor antagonists)
| Dose band | Activity / effect | Mechanism |
|---|---|---|
| Daridorexant 25 mg | Partial blockade sufficient; minimal side effects | Partial OX1R/OX2R blockade |
| Daridorexant 50 mg | Sleep benefit maximized; possible next-day somnolence (FDA max) | Near-complete occupancy |
| Daridorexant 100 mg | No additional benefit; narcolepsy-like state; not approved | Complete blockade overshooting the window |
| Suvorexant 10 mg | Partial blockade; starting dose | Partial OX1R/OX2R blockade |
| Suvorexant 15–20 mg | Sleep benefit maximized; next-day somnolence risk | Near-complete blockade |
| Lemborexant 5 mg → 10 mg | Recommended → maximum dose | Dual OX1R/OX2R antagonism |
H1 antihistamines
| Dose band | Activity / effect | Mechanism |
|---|---|---|
| Cetirizine 5 mg / Loratadine 10 mg / Fexofenadine 180 mg | Peripheral H1 blockade; minimal side effects | Peripheral H1 |
| Cetirizine 10–20 mg / Loratadine 20 mg | Additional peripheral benefit; ≥20 mg cetirizine → CNS penetration → sedation | CNS H1 blockade |
| Cetirizine 10 mg (pre-exercise PEM-prophylaxis probe) | H1 blockade to reduce PEM severity (proposed) | Blocks exercise-induced histamine amplification |
Ketotifen
| Dose band | Activity / effect | Mechanism |
|---|---|---|
| 0.25–1.0 mg | MCAS benefit without sedation | Mast-cell stabilization |
| 1.0–2.0 mg | Mast-cell benefit maximized; CNS sedation may add sleep benefit | Mast-cell + CNS H1 |
| 2.0–4.0 mg | Net benefit inverts; sedation exceeds MCAS gain; weight gain | CNS H1 sedation dominates |
6 Practical implication
The therapeutic implication of an inverted-U depends entirely on which category the drug belongs to.
For Nrf2 hormesis drugs, the width of the window measures stress-response capacity. A patient with a narrow LDN window is predicted to have narrow windows for sulforaphane, melatonin, and lithium — because all converge on the same Keap1-Nrf2 pathway. The prediction is testable.
For the catecholamine inverted-U, the inversion point measures baseline prefrontal catecholamine tone. A patient with a narrow modafinil window is predicted to have narrow windows for duloxetine and guanfacine — because all converge on the same D1/α2A circuit. The prediction is testable, and it is mechanistically independent of the Nrf2 prediction.
For the partial-agonist inverted-U, the inversion point measures receptor reserve — a structural property of the dopamine system, not the stress-response system.
For the Category 4 drugs (concentration-dependent target selection and biphasic biochemistry), the inversion point is a drug-specific property, and it carries no cross-drug predictive value whatsoever.
If you treat the inverted-U as one phenomenon, you predict that all eighteen drugs will correlate within patients — narrow LDN window, narrow modafinil window, narrow rapamycin window. That prediction is probably false. The mechanistic decomposition predicts a more interesting result: correlation within clusters, no correlation across clusters. Nrf2 drugs should track each other. Catecholamine drugs should track each other. LDA should be borderline — it shares a dopamine substrate with the catecholamine cluster but operates through a different mechanism (partial agonism vs reuptake inhibition). Rapamycin and allopregnanolone should be islands.
7 The falsifiable prediction
If hormetic reserve is a single cross-system trait — a general property of ME/CFS physiology — then inversion points should correlate across all eighteen drugs. The orthogonal-mechanism test par excellence is LDN vs modafinil: completely distinct receptors (TLR4 vs DAT/NET), cell types (microglia vs presynaptic terminals), and anatomical targets. If these two correlate (r ≥ 0.4), hormetic reserve is a genuine systems property. If they don’t (r < 0.2), the inverted-U is a collection of drug-specific artefacts and the single-trait model is false.
This test is specified in the paper as the HIP-B trial: six drugs (LDN, sulforaphane, duloxetine, modafinil, rapamycin, allopregnanolone) at four dose levels, within-patient crossover, n ≥ 80 for definitive testing. Until HIP-B or an equivalent is run, the unity of the inverted-U in ME/CFS is an empirical question, not an established fact.
8 Certainty estimate
| Claim | Certainty |
|---|---|
| Nrf2-mediated hormesis is the mechanism for the Nrf2-cluster drugs | Moderate — Nrf2 hormesis is well-established in toxicology (Calabrese corpus); direct evidence in LDN specifically is absent |
| The partial-agonist inverted-U applies to LDA | High — receptor-occupancy property, documented for all partial agonists |
| The catecholamine inverted-U applies to prefrontal D1/α2A drugs | High — canonical neuroscience, replicated across species and methods (Arnsten, Cools) |
| mTORC1/C2 selectivity explains rapamycin’s inverted-U | High — established in mouse and human cell lines (Sarbassov, Lamming) |
| Allopregnanolone’s biphasic curve is a two-binding-site phenomenon | Moderate — biphasic concentration-response documented; two-site mechanism inferred |
| Inversion points correlate within mechanistic clusters | Low — never tested for any cluster |
| Inversion points do not correlate across clusters | Low — never tested |
| HIP-B would distinguish single-trait from multi-trait hormetic reserve | Moderate — the statistical power calculation is sound (n ≥ 80 for r ≥ 0.4 vs r < 0.2); the trial does not exist |
This post draws on the hormesis framework developed in Loth 2026, synthesising the Calabrese hormesis corpus, the Arnsten/Cools catecholamine inverted-U literature, Sabatini/Lamming mTORC1/C2 dose selectivity, and Andreen allopregnanolone biphasic pharmacology. The framework was triggered by Kevin Lee’s clinical observation that higher-dose LDN often shows less benefit than lower-dose LDN, and expanded by Yannick L.’s observation that the same logic applies to LDA and every other low-dose or high-dose medication in the paper — a pattern that the standard single-mechanism TLR4 model cannot explain. The inter-individual threshold variance framework — redox state, allostatic load, and baseline cellular reserve as determinants of hormetic window position — draws on Kevin Lee’s articulation of why hormetic dose-response curves are organism-dependent rather than pharmacologically fixed.
Prev: Why More Isn’t Better: The Dose-Response Paradox in LDN and LDA · Next: Your LDN Dose Is a Diagnostic · Series: Part 4: The Pharmacopoeia · Part 5: How to Run a Dose-Finding Trial