The lithium-and-brain evidence only makes sense in context: it spans different forms (trace water, supplements, prescription), different life stages, and study designs of very different strength. Most of it is ecological — comparing regional water lithium with regional rates of suicide, dementia, or autism — which can reveal associations but cannot prove cause. This page lays out the studies by form and design so individual findings are not taken out of context.
This page is population-level epidemiology and early laboratory science, not individual-risk guidance. The suicide rows summarize regional research and say nothing about any individual's risk. If you are in crisis, in the US call or text 988 (Suicide & Crisis Lifeline), available 24/7, or contact your local emergency services.
How strong is the evidence on lithium and the brain?
It is suggestive but, by clinical standards, still early. The water-epidemiology lane is dominated by ecological studies that correlate regional drinking-water lithium with regional rates of suicide, dementia, or autism. Most suicide studies find an inverse association, dementia findings are non-linear and mixed, and prenatal exposure shows a harm signal for autism. Because these are population correlations, they cannot establish individual causation. The only true individual-level cohort (Duthie and colleagues, 2023) was null, low-dose findings are inconsistent, and publication bias has been flagged (Eyre-Watt and colleagues, 2021; DOI).
Separately, a 2025 laboratory lane reframed the question: a Nature study (Aron and colleagues) reported that endogenous lithium is reduced in mild cognitive impairment and that amyloid plaques sequester lithium in the Alzheimer's brain, with lithium depletion accelerating pathology in mice (DOI). That is mouse and human-tissue work, not human treatment. The first randomized human trial followed in 2026 and did not meet its primary outcomes (see the database below). Treat all of these as hypotheses at different stages, and never transfer a result from one lane to another.
Studies database — lithium and brain outcomes, by form
The table below compiles the studies discussed across this hub. The Form column makes the lanes explicit: trace water-lithium, prescription lithium carbonate, lithium orotate (the supplement salt), mouse models, and review or commentary articles. Evidence does not carry across these forms: a carbonate result is not an orotate result, and a mouse result is not a human result.
| Study | Year | Form | Region | Design | Outcome | Sample | Water Li (µg/L) | Direction | Notes |
|---|---|---|---|---|---|---|---|---|---|
| Aron and colleagues (DOI) | 2025 | Mouse / orotate | n/a | Mouse models + human tissue (ROSMAP) | Alzheimer's pathology, memory | Mouse + human-tissue | n/a | Lithium depletion worsened pathology; orotate prevented it (mice) | Endogenous lithium reduced in MCI; amyloid sequesters lithium; orotate chosen for reduced amyloid binding; via GSK-3β. Mouse + human-tissue, not human treatment |
| Bush (DOI) | 2025 | Commentary | n/a | News & Views companion | Expert context on Aron | n/a | n/a | Framed as important but unproven hypothesis | The Nature commentary accompanying Aron 2025 |
| Gildengers and colleagues (DOI) | 2026 | Carbonate | USA (Pittsburgh) | RCT (double-blind, placebo-controlled pilot) | Cognition, hippocampal/cortical volume, BDNF in MCI | 80 adults ≥60, 2 yr | n/a | Null on primaries; borderline verbal-memory signal | First human RCT. Used lithium carbonate — the amyloid-sequestered form. None of 6 co-primary outcomes met significance |
| Moore and colleagues (DOI) | 2026 | Review | n/a | Narrative review (25-year) | Lithium as disease-modifying agent in dementia | n/a | n/a | Multiple resilience pathways; repletion hypothesis awaits replication | GSK-3β, BDNF, Bcl-2, mitochondrial effects; low-dose neurotrophic in preclinical models; supports a trial of low-dose orotate to limit carbonate's kidney/thyroid risk, not as brain-superior chemistry |
| Hajek and colleagues (DOI) | 2026 | Commentary / orotate | n/a | Chemistry critique | Whether lithium orotate is distinct from Li⁺ | n/a | n/a | Skeptical: orotate likely dissociates to ordinary Li⁺ | Argues no stable orotate complex shown in physiological fluids; pharmacokinetics comparable to carbonate. One side of a live debate |
| Schrauzer & Shrestha | 1990 | Trace water | Texas, USA | Ecological (county) | Suicide, homicide, rape | 27 counties, 10 yr | low <12 vs high 70–170 | Inverse (protective at high) | First study in field; crime + suicide |
| Ohgami and colleagues | 2009 | Trace water | Japan (Oita) | Ecological | Suicide (SMR) | 18 municipalities | low (JP range) | Inverse | SMR 2002–2006 |
| Kabacs and colleagues | 2011 | Trace water | England (East) | Ecological | Suicide | 47 subdivisions | ~0–21 | Null | Negative study |
| Kapusta and colleagues | 2011 | Trace water | Austria | Ecological | Suicide | 99 districts | wide | Inverse | — |
| Helbich and colleagues | 2012/2015 | Trace water | Austria | Ecological (spatial) | Suicide | districts | wide | Mixed / spatially sensitive | Depends on spatial model |
| Blüml and colleagues (DOI) | 2013 | Trace water | Texas, USA | Ecological (county) | Suicide | 226 counties | wide | Inverse | Larger replication of Schrauzer; inverse after socioeconomic adjustment |
| Giotakos and colleagues | 2013 | Trace water | Greece | Ecological | Suicide | prefectures | wide | Inverse | — |
| Pompili and colleagues | 2015 | Trace water | Italy | Ecological | Suicide | provinces | wide | Inverse | Sex differences reported |
| Knudsen and colleagues | 2017 | Trace water | Denmark | Ecological | Suicide | national | 0.6–30.7 | Null (after adjustment) | Negative study |
| Liaugaudaite and colleagues | 2017 | Trace water | Lithuania | Ecological | Suicide | municipalities | wide | Inverse (men) | Signal in men |
| Kessing and colleagues (DOI) | 2017 | Trace water | Denmark | Ecological (nested case-control) | Dementia (also AD, vascular) | 73,731 cases / 733,653 controls | ~2–30 (median ~12) | Non-linear: protective >15 (IRR 0.83); higher risk 5.1–10 (IRR 1.22) | Largest dementia study; JAMA Psychiatry |
| Fajardo and colleagues | 2017 | Trace water | Texas, USA | Ecological (county) | Alzheimer's | county-level | protective only >30 | Inverse (above 30 µg/L) | — |
| Parker and colleagues | 2018 | Trace water | USA | Ecological (insurance/claims) | Dementia | claims data | low | Null after adjustment | Crude inverse vanished after adjustment; small counts |
| Muronaga and colleagues | 2022 | Trace water | Japan | Ecological | Alzheimer's dementia | 808 cities/wards (~91% pop) | low (JP range) | Inverse — women only | National database 2010–2014 |
| Rafsanjan district study | 2022 | Trace water | Iran | Ecological | Suicide attempts | 16 areas | mean 47.3 (9.4–141) | Inverse; strongest in women | r=−0.55 overall; −0.73 women |
| Duthie and colleagues | 2023 | Trace water | Scotland | Cohort (individual) | Dementia | ~37,000 | very low (<2) | Null overall; paradoxical higher risk in women | Only true individual cohort; levels likely too low |
| Kugimiya and colleagues | 2023 | Trace water | Japan | Ecological | Suicide | nationwide | — | Inverse (reported) | Among the largest suicide studies |
Reviews and meta-analyses (water-lithium epidemiology)
| Review | Year | Journal | Type | Outcome | Pooled | Summary finding |
|---|---|---|---|---|---|---|
| Del Matto and colleagues (DOI) | 2020 | Neurosci Biobehav Rev | Systematic review | Suicide | 16 ecological | 11 of 16 found higher water Li → lower suicide; means 3.8–46.3 µg/L |
| Memon and colleagues (DOI) | 2020 | Br J Psychiatry | SR + meta-analysis | Suicide | 15 ecological | Inverse; significant for total and female suicide |
| Barjasteh-Askari and colleagues (DOI) | 2020 | J Affect Disord | SR + meta-analysis | Suicide | 13 ecological + 1 cohort (939 regions; 3.7M) | OR 0.42 (0.27–0.67) reduced suicide |
| Eyre-Watt and colleagues (DOI) | 2021 | Aust N Z J Psychiatry | SR + meta-analysis | Neuropsychiatric | 27 studies / 2,678 regions / 113M | Suicide r=−0.19; admissions reduced; dementia unclear; flagged publication bias |
| Fraiha-Pegado and colleagues (DOI) | 2024 | Int J Bipolar Disord | Systematic review | Dementia | 5 studies | Trace Li → lower dementia; replicated across studies and continents; protective ~2–56 µg/L; below ~2 µg/L no effect; women more sensitive |
Developmental and youth evidence
| Study | Year | Country | Life stage | Design | Outcome | Direction |
|---|---|---|---|---|---|---|
| Liew/Ritz and colleagues (DOI) | 2023 | Denmark | Prenatal (in utero) | Ecological case-control | Autism (ASD) | Harm: OR 1.23 per IQR; ~24–26% higher at mid quartiles; 8,842 cases / 43,864 controls; Wnt/β-catenin mechanism; JAMA Pediatrics |
| Danish trace-element cohort | 2021 | Denmark | Childhood | Cohort | ADHD | Lithium-specific direction not established; 284,309 born 1994–2007; exploratory |
| Shimodera and colleagues | 2010 | Japan (Kochi) | Adolescence | Individual cross-sectional | Depression, aggression, suicidal behavior | Protective for depression and interpersonal violence; null for suicidal ideation and self-harm |
| Nishida / Shimodera group | 2018 | Japan (Kochi) | Adolescence | Cross-sectional | Psychotic experiences | First study; specific direction not extracted |
| Youth bipolar lithium review | 2024 | Multi | Children/adolescents (≤25) | Systematic review (clinical, not water) | Suicidality/self-harm in bipolar youth | Generally supports reduced ideation/self-harm; mixed; therapeutic context, not drinking water |
DHA and dementia evidence (companion-nutrient parallel)
DHA is included here because its literature follows the same pattern as trace lithium: strong observational signals, underwhelming randomized trials. It is a cautionary parallel, not lithium evidence.
| Study | Year | Source | Design | Sample | Finding | ApoE4 note |
|---|---|---|---|---|---|---|
| Sala-Vila and colleagues (Framingham Offspring) (DOI) | 2022 | Nutrients | Prospective cohort | 1,490 dementia-free ≥65 | Highest vs lowest RBC DHA: lower incident Alzheimer's | ApoE4 carriers may benefit from extra DHA |
| UK Biobank re-analysis | 2023 | cohort | Prospective cohort | 267,312 | Total omega-3 Q5 vs Q1: all-cause dementia HR 0.79; AD HR 0.87 | Strongest for total/non-DHA omega-3 |
| Omega-3 & dementia meta-analysis | 2023 | cohorts | Meta-analysis | Multiple cohorts | Dietary DHA: lower dementia and AD; erythrocyte DHA RR 0.94; plasma EPA RR 0.88 | — |
| Lázaro and colleagues | 2024 | Alz Dement (DADM) | Cross-sectional + FDG-PET | 320 cognitively unimpaired at-risk | DHA–brain-glucose associations restricted to amyloid+/tau+; ALA stronger in ApoE4 | ALA links stronger in ApoE4 carriers |
| Omega-3 & dementia review | ~2014 | PMC review | Narrative review | — | DHA broadly neuroprotective via multiple mechanisms | Some studies suggest ApoE4 limits protection |
| RCT caveat (multiple trials) | various | — | Randomized controlled trials | — | Limited or null efficacy for slowing decline or preventing AD | Likely timing/dose/too-late issues |
The life-stage arc (the through-line)
Synthesizing across the water-epidemiology lane, the direction of the association appears to change with life stage:
- Prenatal / fetal: harm signal — higher water lithium associated with higher autism risk (Liew/Ritz, 2023), proposed to act by perturbing Wnt/β-catenin during active neurodevelopment.
- Childhood: largely a data gap. One large Danish cohort screened lithium among 17 trace elements against ADHD; a lithium-specific direction was not established.
- Adolescence: leans beneficial for mood and behavior — inverse with depression and interpersonal violence (Shimodera, 2010) — but null for adolescent suicidal ideation and self-harm in that sample.
- Adulthood: benefit signal — ecological inverse association with suicide in most studies (a minority null).
- Older age: benefit signal — lower dementia and Alzheimer's at higher lifelong exposure (Kessing, Fajardo, Muronaga), non-linear in the Danish data.
The harm appears specific to the fetal neurodevelopmental window; by adolescence the direction is already net-positive for mood and aggression. The precise crossover age is unstudied (the early-childhood gap). See lithium across the lifespan for detail.
Key caveats (read before citing any number)
- Design: almost all water studies are ecological except Duthie (cohort) and the adolescent Japanese surveys (individual-level). Population correlations are not individual causation.
- Confounding: region-level confounders (socioeconomic status, geology, ethnicity, urbanicity, longevity) are hard to remove.
- Low-dose inconsistency: the Danish U-shape (apparent harm 5–10 µg/L, benefit above 15 µg/L), the Scotland null/paradox, and two null suicide studies.
- Publication bias: flagged by Eyre-Watt and colleagues (2021); the ecological suicide signal conflicts with some randomized-trial meta-analyses.
- Form separation: the 2026 human RCT (Gildengers) used lithium carbonate, the mouse work (Aron) used lithium orotate, and the epidemiology is about trace water lithium. A null in one form is not a verdict on another.
- DHA parallel: strong observational dementia protection but underwhelming RCTs — the same observational-beats-trial pattern as trace lithium.
- Water context: typical US public-supply water lithium clusters in the low single digits to low tens of µg/L; many municipal supplies are below the few-µg/L floor where an effect is even plausible.
Evidence compiled June 2026. This page is maintained as a reference; new human trials will change the picture.
Frequently asked questions
Is the evidence that lithium protects the brain strong?
No. It is suggestive but early by clinical standards. Most water studies are ecological population correlations, the one individual cohort was null, low-dose findings are inconsistent, and publication bias has been flagged. The 2025 laboratory finding is mouse and human-tissue work, and the first human trial did not meet its primary outcomes. Treat the associations as hypotheses, not proof.
Why do ecological studies count as weak evidence?
Ecological studies compare group-level averages (regional water lithium versus regional disease rates) rather than tracking individuals. They cannot rule out that another regional factor — wealth, geology, age structure — explains the pattern, so they cannot establish individual causation.
Do any randomized trials support lithium for the brain?
The first randomized trial of low-dose lithium in mild cognitive impairment (Gildengers and colleagues, 2026, JAMA Neurology; 80 adults, two years) did not meet any of its six pre-specified primary outcomes, with only a borderline signal on one verbal-memory measure. That trial used lithium carbonate, the form that under the leading 2025 hypothesis is sequestered by amyloid in the Alzheimer's brain, so it tempers certainty about low-dose lithium but does not test the non-sequestered forms (such as orotate) the mouse work used. Earlier, smaller studies were more encouraging: a placebo-controlled amnestic-MCI trial reported slowed decline and reduced tau (reviewed in Forlenza and colleagues, 2012; DOI), and a small randomized study reported a 300 µg/day microdose stabilized cognition in Alzheimer's over 15 months (Nunes and colleagues, 2013; DOI). There are no randomized trials of trace (drinking-water) lithium.
Does lithium orotate work better in the brain than other forms?
Whether lithium orotate behaves differently in the brain than other lithium salts is an open question. There are promising early reasons it might: the 2025 mouse work used orotate specifically because it showed reduced amyloid binding, and orotate has long been proposed to enter cells or cross into the brain more readily. But this is not proven in humans, the human data are extremely early, and some researchers argue on chemistry grounds that orotate likely dissociates to ordinary lithium ions after ingestion (Hajek and colleagues, 2026; DOI). The honest summary: biologically interesting, mechanistically plausible, clinically unproven.
What is the single most important study here?
It depends on the lane. For human treatment, the 2026 Gildengers trial is now the state of the art, and it was null on its primaries. For dementia epidemiology, the largest is Kessing and colleagues (2017), a Danish nested case-control study of 73,731 cases that found a non-linear pattern. For suicide, the 2020 meta-analyses (Memon; Barjasteh-Askari) pool the ecological literature. For mechanism, the 2025 Aron study and the 2026 Moore review are the reference points. Each has the limitations described above.
Does this page give individual medical or risk advice?
No. Everything here is population-level research and early laboratory science compiled for reference. It is not individual-risk information or medical advice, especially regarding suicide. For personal concerns, consult a qualified clinician; in crisis, call or text 988 in the US.