Introduction

Lithium was identified as a psychoactive drug in the mid-twentieth century,1 and remains the primary therapy for bipolar affective disorder2 (although it is often used in other psychiatric disorders also). Unfortunately, it has a narrow therapeutic window limited by adverse effects,3, 4 and neither the mechanism of action nor its associated toxicity are well understood. Lithium modulates multiple key cellular signaling molecules,3, 5 including inositol monophosphatase (IMPase) and glycogen synthase kinase-3 (GSK-3), considered likely to be therapeutic targets.3 Parkinson-like hand tremor is a common adverse effect of lithium therapy, and limited cross-sectional studies and several case reports have raised concern that lithium may increase the risk for extrapyramidal syndromes (for example, dyskinesia) not only in overdose, but possibly also at appropriate maintenance levels.6, 7, 8, 9, 10, 11, 12, 13, 14, 15 Recently, long-term lithium use was reported as associated with an increased incidence of anti-parkinson drug use or Parkinson’s disease (PD) diagnosis.16 Whether lithium could induce extrapyramidal syndromes by neurotransmitter imbalance (like first-generation antipsychotics such as haloperidol) or by damage to the basal ganglia is not known.

Although a previous study demonstrated the neurotoxic potential of lithium in mice treated with a dose considered to be within the therapeutic range (2.55 g kg−1 in food),17 lithium has been more extensively investigated for a host of neuroprotective properties in various neurological and psychiatric models including cerebral ischemia/reperfusion injury18 and other neurotoxic insults.19 A recent study19 of adolescents at ultrahigh risk for psychosis20 found that low-dose lithium treatment over 4 months lowered hippocampal T2 relaxation time (T2*), and this was taken to signify neurochemical improvements that could underlie potential neuroprotective benefits of low-dose lithium as an approach to prevent the onset of psychosis.21, 22 However, a reduction of T2* can also be caused by an elevation of brain iron, and has been used to map such changes in diseases where iron levels rise, such as cortical tissue in Alzheimer’s disease (AD).23

Lithium has attracted considerable interest as a therapeutic candidate in AD research. In neuronal cultures, lithium is reported to decrease tau phosphorylation24, 25, 26, 27 and to inhibit Aβ generation.28, 29, 30 In Aβ-overexpressing transgenic models for AD, lithium treatment consistently decreases Aβ levels, GSK-3 activity and tau phosphorylation.29, 31, 32, 33, 34 In such models, lithium is reported to prevent Aβ toxicity,32 preserve dendritic structure,31 promote neurogenesis34 and rescue Aβ-induced cognitive impairment.31, 33, 34 Similarly, in transgenic mice that overexpress pathogenic mutant tau, lithium treatment reduced levels of hyperphosphorylation.35, 36, 37, 38, 39 Despite these promising preclinical findings, lithium failed to show therapeutic efficacy for AD in a 10-week multicenter, randomized, single-blind, placebo-controlled trial, with no significant effects on cerebrospinal fluid levels of Aβ and phospho-tau or cognitive end points.40

Nevertheless, lowering total tau expression is still being explored as a potential treatment for AD, and lithium is still being considered as a treatment option as it lowers both tau protein and mRNA levels in cultured cortical neurons.41, 42 Supporting this concept, long-term (5 months) lithium treatment of PrP T44 tau transgenic mice resolved tau aggregates by promoting ubiquitination and degradation of tau, rather than altering GSK-3 activity.36

We, and others, have reported concerns that lowering levels of brain tau could induce age-dependent neurodegeneration.43, 44, 45 In tauopathies (such as AD and PD), insoluble tau aggregates are accompanied by a fall in soluble tau levels in affected tissue.43, 46, 47, 48, 49, 50, 51 We have argued that this will lead to a loss of tau function that could cause neurodegeneration, as tau functions to lower neuronal iron levels by promoting the presentation of the amyloid protein precursor (APP) to the neuronal surface,43 where it promotes the efflux of iron.52, 53, 54 Without soluble tau, neurons accumulate iron to toxic levels, and hence tau-knockout mice develop a parkinsonism with impaired-cognition phenotype that is rescued by iron chelation.43, 44 Iron accumulation in the substantia nigra (SN) is a robust feature of PD, and is a sufficient cause for parkinsonism (reviewed in Ayton and Lei55). There has been one case report of such SN iron elevation with extrapyramidal signs following long-term lithium treatment.56 We hypothesized that as lithium lowers neuronal tau, lithium could engender SN iron accumulation and induce parkinsonian neurodegeneration, explaining a cryptogenic aspect of its adverse effect profile.

Materials and methods

Reagents

Reagents were purchased from Sigma-Aldrich (Castle Hill, NSW, Australia), unless specified.

The lithium clinical imaging study

The lithium clinical imaging study has been previously described.19 The pilot study was conducted as an open-label, nonrandomized, controlled, single-center study. Participants were consecutive referrals to the Personal Assessment and Crisis Evaluation (PACE) Clinic, a specialist clinic for young people at ultrahigh risk (UHR) for psychosis.20 PACE is a subprogram of ORYGEN Youth Health, a specialist service for young people with emerging mental disorders. Inclusion criteria for PACE were: (1) age between 14 and 30 inclusive years, (2) resident of the Melbourne metropolitan area or the ability to travel to the PACE clinic at least once a week and (3) meeting PACE UHR criteria. UHR criteria were determined using the Comprehensive Assessment of the At Risk Mental State (CAARMS).57 Rationale and validity for these criteria have been fully described elsewhere.58 Exclusion criteria were: (1) medical or neurological conditions that may diminish functioning or that may account for some of the symptoms leading to the initial referral, for example, epilepsy; (2) clinically relevant biochemical or hematological abnormalities; (3) serious coexisting illnesses such as liver or renal insufficiency, significant cardiac, vascular, pulmonary, gastrointestinal, endocrine, neurological or metabolic disturbances (including patients known to be HIV positive); (4) history of previous acute psychotic episodes either treated or untreated; (5) a current score of 5 or 6 on the CAARMS Mania item or a history of a previous manic episode; (6) any previous use of neuroleptic or mood-stabilizing medications; (7) history of severe drug allergy or hypersensitivity; (8) history of intellectual disability (IQ < 70); and (9) inability to understand or communicate in English. All UHR subjects received PACE treatment as usual throughout the course of the study including psychological and medical treatment. Use of antidepressants was allowed if UHR patients met criteria for major depression. UHR subjects in the low-dose lithium group took one slow-release 450 mg tablet of lithium carbonate each night for the entire study period (3 months), intended to achieve serum lithium levels of 0.3–0.5 mM, in addition to treatment as usual.

Baseline magnetic resonance imaging scans and clinical assessments were conducted before commencement of the low-dose lithium treatment, and within 2 weeks of admission to the study. Monitoring of psychopathology and tolerability measures occurred at monthly intervals, and follow-up scanning was conducted as close as possible to 3 months after baseline (median for the control group was 3.0 months, range 0.8–27 months; for the lithium group 3.8 months, range 1.6–5.9 months, with some subjects in both groups not completing the full treatment period). Extrapyramidal side effects were not systematically assessed. The local research and ethics committee approved the study protocol. All participants gave written informed consent. Additional written informed consent was obtained from parents or guardians for participants aged 14–18 years. Participants were made aware that they could withdraw from the study at any time and that withdrawal would not compromise access to any clinical services.

Brain imaging and SN T2 relaxometry

All scans were performed on a 3T GE LX Horizon scanner (GE Healthcare, Milwaukee, WI, USA) at the Brain Research Institute (Melbourne, Victoria, Australia). Eight tilted, coronal T2-weighted images (perpendicular to the long axis of the hippocampus) were obtained over a range of echo times (29 to 231 ms; TR=4 s, slice thickness= 6 mm with 1.5 mm gap, matrix 256 × 256, field of view 24 cm). Single exponentials were then fitted to the image data of corresponding voxels from these eight echoes using iBrain (Brain Research Institute). This created a series of T2 maps, one for each coronal slice. The brightness of each individual voxel on this map represents its calculated T2 relaxation time. The images were then analyzed in various regions of interest with reference to a standard neuroanatomical atlas by two raters who were blinded to treatment group, and quantified using ANALYZE 7.2 software (Mayo Clinic, Rochester, MN, USA), as previously described.19

Primary neuron culture

Primary neurons were prepared as previously described.43 Briefly, cortices of wild-type and tau-knockout embryos (day 14) were dissected and dissociated in 0.025% (w/v) trypsin. The neurons were plated (Nunc, Mt Waverley, VIC, Australia) at a density of 600 000 cells per cm2 in plating medium (Dulbecco’s modified Eagle’s medium with 10% fetal calf serum, 5% house serum and 10 mg l–1 gentamycin sulfate; Invitrogen, Mt Waverley, VIC, Australia). After 2 h of incubation, the neurons were changed to Neurobasal supplemented medium (with B27, 500 μM glutaMAX and 10 μg ml–1 gentamycin sulfate). Experiments were performed 7 days later. Confluent cells were incubated in Neurobasal medium solution before treatment with various concentrations of lithium chloride for 18 h with or without radioactive 59Fe or 65Zn, or various compounds (L690,330; 6-bromoindirubin-3'-oxime (BIO)) at concentrations indicated. The cells were then harvested, with half the batch lysed using lysis buffer (1% Triton X-100, 50 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA, pH 7.4, together with 1:50 proteinase inhibitor and 1:1000 kinase inhibitors I and II), and the other half retained for metal analysis (see below), or assayed for radioactive tracers by γ-counter.

59Fe-loaded transferrin and 65Zn preparation

Purified human apo-transferrin (Tf) was loaded with 59Fe (Perkin Elmer, Glen Waverley, VIC, Australia) to form Tf(59Fe)2 as previously described.43, 52 Neurons were incubated with 1.0 × 10-6 M Tf(59Fe)2 in serum-free Neurobasal supplemented medium for periods indicated. At the conclusions of the treatments, neurons were washed thoroughly with phosphate-buffered saline (PBS) and harvested with trypsin. All media and cell lysates were measured by γ-counter (Wizard 3, Perkin Elmer) for counts per min. For iron efflux assay, cells were incubated at 37 °C and medium was removed at select time points. A comparable time-course control study was performed at 4 °C to account for nonspecific binding of 59Fe to the membrane surface.

The 65Zn experiments were adapted from a previous publication.59 Briefly, serum-free Neurobasal supplemented medium was mixed with 0.04 MBq 65Zn (Oak Ridge Laboratory, Oak Ridge, TN, USA) and unlabeled ZnCl2 (20 μM). After 18 h of treatment, the reaction was stopped by washing cell monolayers three times with ice-cold nonlabeled Hanks’ balanced salt solution (with 2 mM L-histidine and 5 mM EDTA). The cells were then harvested and measured for 65Zn retention by γ-counter (1282 CompuGamma; LKB Wallac, Mount Waverley, VIC, Australia).

Mouse tissue preparation

All mice were housed in a conventional animal facility according to standard animal care protocols and fed standard laboratory chow (Meat Free Rat and Mouse Diet, Specialty Feeds, Glen Forrest, WA, Australia) and tap water ad libitum. All animal procedures were approved by the Florey Institute animal ethics committee (10-017) and were performed in accordance with the National Health and Medical Research Council guidelines. Tau-knockout,60 APP-knockout52 and background C57Bl6/SV129 mice stocks raised under the same conditions were maintained homozygously, with backcrossing to the parental inbred strain every three generations. The mouse was killed with an overdose of sodium pentobarbitone (Lethabarb, Clifford Hallam Healthcare, Dandenong South, VIC, Australia, 100 mg kg−1) and perfused with ice-cold saline before tissues were collected. Body weight and total brain wet weight were recorded. Once extracted, the right brain hemisphere was microdissected and stored at −80 °C until required. The left brain hemisphere was fixed in 4% paraformaldehyde for 24 h, and then transferred to 30% sucrose+PBS (pH 7.4) and kept at 4 °C overnight for tyrosine hydroxylase(TH) immunohistochemistry and brain section analysis.

Lithium treatment to mice

The 3-month-old male Bl6/C57J mice (used in Figure 2) as well as age-matched tau-knockout, APP-knockout and background control (BL6/129sv) mice (used in Figure 4) were randomly assigned (by Excel) and orally gavaged with lithium chloride (3.6 mg kg−1 per day, freshly dissolved in 0.9% NaCl, 0.5% Na-carboxymethylcellulose, 0.5% benzyl alcohol and 0.4% Tween-80, with the use of an oroesophageal needle, continuously for 21 days) or the solution alone. Behavioral tests were performed at days 18–21, 2 h after the daily lithium dose. To control for effects of potential acute sedation, drug-naive Bl6/C57J mice were treated with a single dose of diazepam (3 mg kg−1) or lithium (3.6 mg kg−1) and the performance tests carried out 2 h later.

Behavioral measurements

Pole test assessment was performed as previously described.43 Briefly, mice were placed vertically on a 30 cm vertical, 1 cm diameter pole placed within in their home cage, where mice make an 180° turn and return to the base of the pole. On the day before testing (day 1), the animals were habituated to the pole and allowed five consecutive trials. Animals were then recorded via digital video on the test day (day 2). The interval time for the mouse to turn toward the ground (‘time to turn’) and to reach the ground (‘time to finish’) were both recorded. Each mouse underwent five trials and the average was used in analysis. When mice were unable to turn and move down the pole, the trial was determined as incomplete. Rotarod assessment was performed as previously described.43 Briefly, mice were assessed using a Panlab Rotarod apparatus (Barcelona, Spain) in an accelerating model with triplicate measurements (maximum time of 2.5 min; speed increases every 8 s). The time on the rod as well as the final speed of the rod was recorded and the triplicate averaged for analysis. Open field test: Spontaneous motor activity of mice was measured using a photo-beam activity system (Truscan 2.0, Coulbourn Instruments, Holliston, MA, USA) as previously described.43 The test area was 25.4 cm wide by 25.4 cm deep by 40.6 cm high. The mouse was placed in the chamber for 1 h to allow acclimatization to its surroundings. Parameters of movements were calculated from the interception of beams to provide an XY coordinate. For Y-maze test, a Y-shaped gray-painted timber with arms 29.5 cm long × 7.5 cm wide × 15.5 cm high was used as previously described.43 All mice were subjected to a two-trial Y-maze test separated by a 1-h intertrial interval to assess spatial recognition memory, with all testing performed during the light phase of the circadian cycle. The three identical arms were randomly designated start arm, novel arm and other arm. Visual cues were placed on the walls of the maze. The first trial (training) was for 10 min, and the mice were allowed to explore only 2 arms (starting arm and other arm). For the second trial (retention), mice were placed back in the maze in the same starting arm, and allowed to explore for 5 min with free access to all 3 arms. Behaviors were recorded on video during a 5-min trial and the Ethovision video-tracking system (Noldus, Wageningen, The Netherlands) was used for analysis. Data are expressed as the percentage of frequency and duration for novel arm entries made during the 5-min second trial. All behavioral tests were performed by an investigator blinded to the experimental group.

Anatomical morphology from brain slice

The anatomical morphology was examined as previously described.43 Briefly, caudate/putamen (CPu) and cerebellum sections from mice were sectioned using a Leica Cryostat (North Ryde, NSW, Australia) set at 50 μm thickness and areas of interest were measured using Image-J (v1.49b, NIH, Bethesda, MD, USA), using landmarks (anterior commissure for CPu and flocculus for cerebellum) to identify the depth of coronal sectioning (bregma +0.26±0.01 mm and bregma −6.12±0.01 mm). Two sections per mouse per area were analyzed. CPu area was defined by the boundaries of corpus callosum, lateral ventricle and anterior commissure. Corpus callosum, neocortical and cerebellar cortical thicknesses were averaged from five measurements. All quantifications were blinded.

TH immunohistochemistry

TH immunohistochemistry was performed as previously described.43 On a calibrated Leica Cryostat, 30 μm sections (1:3 series) were collected through the SN pars compacta (anteroposterior −2.92 to −3.64 mm from bregma, Mouse Atlas Figures 55–61), generating 8 sections per mouse (the second of the three sections was analyzed). After a further brief fixation (4% paraformaldehyde for 30 s), sections were blocked in 3% normal goat serum (Millipore, Macquarie Park, NSW, Australia) and incubated with primary anti-TH rabbit polyclonal (1:3000, Millipore) overnight. The sections were then incubated with goat anti-rabbit secondary horseradish peroxidase-conjugated antibody for 3 h (Millipore), followed by diaminobenzidine visualization (1% w/v in PBS+1% w/v CoCl2, 1% w/v NiSO4)+3% w/v hydrogen peroxide. TH immunostained sections were counterstained with Neutral Red to visualize the Nissl substance in all neurons before mounting on Superfrost-Plus slides (Trajan Scientific and Medical, Ringwood, VIC, Australia).

Stereological estimation of nigra neurons

SN neuron number was estimated as previously described.43 Briefly, both TH-positive and TH-negative Nissl-positive neurons were scored according to the optical fractionator rules by an investigator blinded to the experimental group. Using an unbiased counting frame (x=35 μm, y=45 μm) at regular intervals on a sampling grid (x=140 μm, y=140 μm), viewed through a 60 × 1.3 NA oil objective (DMLB Leica Microscope), cells were detected with a morphometry and design-based stereology software package (Stereo Investigator 10.04, Microbrightfield, Colchester, VT, USA). The coefficients of error and coefficients of variance were calculated as estimates of precision and only values of <0.1 were accepted.

Dopamine and DOPAC measurement

Dopamine metabolites were measured as previously described.43 Briefly, CPu tissues were homogenized in high-performance liquid chromatography (HPLC) sample buffer (0.4 M perchloric acid, 0.15% sodium metabisulfite and 0.05% EDTA) before centrifugation at 10 000 g at 4 °C for 10 min. Supernatants were used for dopamine measurement by a HPLC system (ESA Biosciences, Chelmsford, MA, USA; model 584) coupled to an electrochemical detector (ESA Biosciences; Coulochem III detector) (E1: −150 mV, E2: +220 mV, and guard cell: +250 mV). Then, 50 μl was injected onto a MD-150 reverse phase C18 column (ESA Biosciences) and elution was performed at a flow rate of 0.6 ml min−1 in the mobile phase (75 mM sodium dihydrogen phosphate, 1.7 mM 1-octanesulfonic acid sodium salt, 100 ml l−1 triethylamine, 25 mM EDTA, 10% acetonitrile, pH 3). Peaks were identified by retention times set to known standards. Data were normalized to wet weight tissue.

Metal analysis

Metal content was measured as previously described.43 Briefly, samples from each experimental condition were freeze-dried and then resuspended in 69% nitric acid (ultraclean grade, Aristar, VWR Chemicals, Radnor, PA, USA) overnight. The samples were then heated for 20 min at 90 °C, and an equivalent volume of hydrogen peroxide (30%, Merck, South Granville, NSW, Australia) was added for a further 15-min incubation at 70 °C. The samples were diluted in double-distilled water and assayed by inductively coupled plasma mass spectrometer (Ultramass 700, Varian, Palo Alto, CA, USA). Each sample was measured in triplicate and the concentrations determined from the standard curve were normalized to tissue wet weight.

Western blot

Samples from each experiment were homogenized in PBS (pH 7.4) with EDTA-free protease inhibitor cocktail (1:50, Roche, Dee Why, NSW, Australia)+phosphatase inhibitors I and II (1:1000) and centrifuged at 100 000 g for 30 min. Protein concentration of the supernatant following centrifugation was determined by BCA protein assay (Pierce, Mt Waverley, VIC, Australia). Aliquots of soluble tissue fraction with equal protein concentrations were separated in 4–12% Bis-Tris gels with NuPAGE MES running buffer (Invitrogen), and transferred to nitrocellulose membranes by iBlot (Invitrogen). The membranes were blocked with milk (10% w/v) and probed with appropriate primary and secondary IgG-horseradish peroxidase-conjugated antibodies (Dako, North Sydney, NSW, Australia). Enhanced chemiluminescence detection system (GE Healthcare) with the Fujifilm LAS-3000 (Campbellfield, VIC, Australia) was used for visualization. Densitometric quantification of immunoreactive signals was performed by ImageJ (1.49b, NIH) and normalized to the relative amount of β-actin and expressed as a percentage of the mean of the control group. The following primary antibodies were used in the current study: anti-β-actin (Sigma), anti-tau (Dako) and anti-pTau396 (Invitrogen).

Statistics

Statistical analysis was carried out in Prism 6 (GraphPad Software, La Jolla, CA, USA) and power calculations were performed using PASS 13 (trial version, NCSS Statistical Software, Kaysville, UT, USA) based on the variance of results in our previous publication10 and in pilot experiments. Equivalence of variance between experimental groups was confirmed using Levene’s test, when applicable. Normal distribution was assumed for all statistical analyses. All tests were two tailed, with the level of significance set at 0.05. Specific types of test used for each experiment are described in the figure legends.

Results

Low-dose lithium treatment induces reversible T2* changes in the SN consistent with iron elevation

To determine whether lithium treatment has the potential to raise nigral iron levels, we analyzed this nucleus in scans from a previous clinical trial.19 The impact of low-dose lithium treatment on hippocampal tissue integrity in individuals at UHR for psychosis was assessed by T2 relaxometry in an open-labeled, nonrandomized, controlled, single-center pilot study. Participants were treated with lithium or as usual for 3 months, and repeated 3T magnetic resonance images were taken from participants before commencement of the study (scan one) and after drug withdrawal (scan two). Lithium treatment induced a decrease in hippocampal T2 relaxation time19 that could not only indicate an improvement in various neuropathologies (for example, edema), but also reflect an increase in tissue iron.61, 62, 63, 64, 65

As iron elevation in the SN could contribute to Parkinson-like side effects such as the commonly observed hand tremor, we revisited this study and measured T2 relaxation in this region, as well as other comparison regions in the same coronal planes (as marked in Supplementary Figure 1a, with reference to a standard neuroanatomical atlas). We found that the average T2 relaxation time in the SN of the lithium group at 3 months (scan two) was significantly reduced from baseline (scan one) (−5%, P=0.007), and significantly lower than the control (treatment as usual) group at the same time point (−7%, P<0.001; Figures 1a and b), consistent with iron accumulation or edema. The pattern of changes was similar in both hemispheres. Other regions analyzed, caudate and lenticular nucleus, showed no change in T2 relaxation time (P=0.18 for caudate and P=0.29 for lenticular nucleus; Supplementary Figure 1b) compared with the baseline scan. The lateral ventricular cross-sectional area was also unchanged from baseline (P=0.177; Supplementary Figure 1b).

Figure 1
figure 1

T2 relaxation time reductions in the substantia nigra (SN) of participants treated with lithium. (a) Representative magnetic resonance imaging (MRI) images from treatment as usual (top) and lithium (bottom) ultrahigh-risk groups. Arrow indicates the SN. (b) Significant reduction in T2 relaxation time was found in lithium-treated subjects (P<0.001 compared with treatment as usual controls; P=0.007 compared with pretreated scan, two-tailed t-test). Dotted line indicates values of baseline scan; n (treatment as usual)=9, n (lithium)=11. Mean±s.e.m. are shown. ***P<0.001 (versus treatment as usual), ##P<0.01 (versus prescan).

PowerPoint slide

Full size image

Iron accumulation and parkinsonism in a mouse model of lithium therapy

To test whether lithium-induced relaxometry changes could be due to brain iron accumulation, 3-month-old C57/Bl6J mice were orally administered lithium chloride at a low therapeutic dose40 (3.6 mg kg−1 body weight per day) for 21 days. The dose was well tolerated and body weight was unaltered during the treatment period (P=0.188; Figure 2a). Lithium levels were elevated in all tested tissues (P<0.001 for cortex, SN, cerebellum and liver), with the greatest proportional elevation in cortex (+263%) and SN (+152%; Figure 2b). Iron levels were increased in cortex (+40%, P=0.008) and SN (+19%, P=0.039), but not cerebellum, liver or plasma (Figure 2c). The effect of lithium on biological metals was specific to iron, with copper and zinc unaltered in all tested tissues (Supplementary Figures 2a and b). Lithium levels significantly correlated with iron levels in SN (P=0.007, R2=0.62; Figure 2d) and cortex (P=0.026, R2=0.439; Supplementary Figure 3a) in lithium-treated mice, consistent with elevated lithium causing an elevation in iron. The correlation was seen neither in sham-treated SN (R2=0.02; Figure 2d) nor cortex (R2=0.03; Supplementary Figure 3a), nor observed in the cerebellum of lithium- (R2=0.01) or sham-treated (R2=0.05) animals, a region that did not accumulate iron with lithium treatment (Supplementary Figure 3b). Neither copper nor zinc correlated with lithium in these tissues (Supplementary Figures 3c and d).

Figure 2
figure 2

Lithium (Li) induces brain iron elevation and parkinsonism in mice. (a) Li treatment did not induce a change in mouse weight, monitored daily as an index of animal health. (b) Li treatment elevated lithium levels in mouse brain and liver (P<0.001 for cortex (Ctx), substantia nigra (SN), cerebellum (Cereb) or liver, two-tailed t-test). (c) Li treatment induced iron elevation in cerebral regions (P=0.008 for Ctx and P=0.039 for SN; two-tailed t-test) but not Cereb, liver or plasma. (d) Iron levels in SN correlate with lithium levels in Li-treated mice (R2=0.62, P=0.007, linear regression) but not sham-treated mice (R2=0.02, P=0.702, linear regression). (e) Quantification of western blots showed reduction of tau in cortex (P=0.032, two-tailed t-test) and SN (P<0.001, two-tailed t-test) of Li-treated mice compared with sham-treated mice. (f) Following 21 days of exposure, Li-treated mice took longer to turn (P=0.008, two-tailed t-test) in the Pole test. (g) Li-treated mice maintained less time on the rod compared with sham-treated mice (P=0.004, two-tailed t-test); n=12 per treatment group. (h, i) Representative tyrosine hydroxylase (TH) staining of sham-treated (top) and Li-treated (bottom) mice. Quantitative data are shown in (i). Significant reduction was found in TH-immunoreactive (TH-ir) neurons in the SN of Li-treated mice (P<0.001, two-tailed t-test); n=5 per treatment group. (j) Significant reductions were found in the striatal dopamine levels (P=0.023, two-tailed t-test), measured by high-performance liquid chromatography (HPLC) in Li-treated mice; n=12 per treatment group. Mean±s.e.m. are shown. *P<0.05, **P<0.01, ***P<0.001.

PowerPoint slide

Full size image

We assessed the brains of lithium-treated mice to determine whether a drop in soluble tau could explain the iron elevation.43 As expected,36, 38, 39, 66 the fraction of brain tau phosphorylated at Ser396 was suppressed by lithium treatment (cortex: −40%, P=0.034; SN: −43%, P=0.039; Supplementary Figure 4). We also discovered that lithium administration lowered PBS-soluble total tau levels in cortex and SN (cortex: −31%, P=0.032; SN: −42%, P<0.001; Figure 2e).

As tau loss causes iron-mediated neurodegeneration with parkinsonism,43, 44, 45 we investigated whether lithium treatment impaired motor function in these mice. Lithium-treated mice compared with sham-treated mice exhibited a significant increase in time to turn (P=0.008; Figure 2f) and to finish (P=0.041; Supplementary Figure 5a) on the Pole test, and significantly impaired performance on the accelerated Rotarod test as evidenced by reduced latency to fall (P=0.004; Figure 2g) and lower average speed to fall (at speed 6, P<0.001; at speed 7, P=0.015; Supplementary Figure 5b). Rotarod impairment has previously been noted in 10-month-old mice treated for 4 weeks with lithium, but the cause was not ascertained.39 Lithium-treated mice also exhibited reduced locomotion distance (P=0.022; Supplementary Figure 5c), reduced velocity (P=0.023; Supplementary Figure 5d) and reduced average distance per movement (P=0.018; Supplementary Figure 5e), without alterations in movement duration or time spent in the corners (data not shown) in the open field test.

Control studies were performed to ascertain whether short-term sedation4, 67 could have contributed to the motor impairment induced by lithium. Diazepam, a sedative with well-characterized effects on motor behaviors,68, 69, 70, 71 was compared with lithium within 2 h after dose for acute motor behavioral effects. Single-dose lithium or diazepam did not impair mouse Rotarod performance at this interval (Supplementary Figure 6a). Diazepam, but not lithium, induced a sudden drop in spontaneous locomotion 20 min after dosing in the open field test (P=0.006; Supplementary Figure 6b). These data indicate that motor impairments induced by lithium were not due to sedation.

To determine whether the motor deficits induced by lithium could be because of Parkinson-like neuropathology, the TH-immunoreactive (dopaminergic) neurons within the SN were stereologically counted. The dopaminergic neurons were significantly decreased in mice treated with lithium for 21 days (−30%, P<0.001; Figures 2h and i), but TH-negative, Neutral Red-positive neurons were unaltered (Supplementary Figure 7a). Accompanying this loss, dopamine (−40%, P=0.023; Figure 2j) and 3,4-dihydroxyphenylacetic acid (DOPAC; −37%, P=0.024; Supplementary Figure 7b) levels in the striatum were also decreased in lithium-treated mice. Excess dopamine induces hyperactivity, a characteristic of mania,72 and is attenuated by lithium treatment.73 Lithium decreases dopaminergic release74 and suppresses levels of dopamine and its metabolites in the striatal projection field and nucleus accumbens.75 Our findings indicate that suppression of the SN–striatal dopaminergic pathway may contribute to the motor deficits induced by lithium as well as, potentially, to its long-term antimanic and antipsychosis properties.

Lithium induces tau-mediated iron accumulation in neuronal culture

Having found that by inducing nigral tau suppression lithium treatment of normal mice recapitulated a series of abnormalities that appear in tau-knockout mice (nigral iron elevation with dopaminergic neurodegeneration and parkinsonism), we investigated the mechanism in mouse primary cortical neuronal cultures. Lithium (at nontoxic concentrations29) induced significant iron accumulation within 18 h in a dose-dependent manner (5 mM Li: +48%, P=0.005; 10 mM Li: +56%, P<0.001; Figure 3a), but did not alter copper or zinc levels (Figure 3b), consistent with the changes in brain levels of these metals in mice treated with lithium (Figure 2c). Lithium (10 mM) significantly inhibited the efflux of iron (chased after 59Fe loading) into the media (P<0.001; Figure 3c), explaining the retention of iron. As expected, lithium treatment (10 mM) had no effect on zinc retention in primary neurons treated with radio-labeled 65Zn (P=0.176, Supplementary Figure 8).

Figure 3
figure 3

Lithium (Li)-induced iron elevation is caused by tau reduction in primary neuronal culture. (a) Li treatment to neurons for 18 h induced dose-dependent 59Fe retention (P=0.005 for 5 mM Li, P<0.001 for 10 mM Li, one-way analysis of variance (ANOVA), n=4 per treatment). (b) Li treatment of neurons for 18 h induced specific iron elevation (P<0.001, two-tailed t-test, n=6 per treatment), without altering copper or zinc. (c) Li treatment (10 mM) inhibited iron export (P<0.001, nonlinear regression with extra sum-of-squares F test, n=6 per group). (d) Li treatment reduced neuronal tau levels (representative blots shown beneath) at 5 mM (P=0.007) and 10 mM (P=0.034, one-way ANOVA) after 18 h; n=4 per group. (e) Tau-knockout neurons retained more 59Fe iron (P=0.032) (18 h) but, in contrast to wild-type neurons, were resistant to Li-induced iron accumulation (P=0.038, two-way ANOVA); n=4 per group. Each experiment was independently repeated three times. Mean±s.e.m. are shown. *P<0.05, **P<0.01, ***P<0.001.

PowerPoint slide

Full size image

The activities of both IMPase and GSK-3 (ref. 5) are considered molecular targets for lithium’s therapeutic effect in bipolar disorder.3 L-690,330, a potent and more specific inhibitor of IMPase,76 had no effect on cellular iron level during the 18-h incubation (P=0.106; Supplementary Figure 9a), thus eliminating IMPase in the mechanism of iron accumulation induced by lithium. However, BIO, a potent and specific inhibitor of GSK-3,77 caused iron retention similar to the effect of lithium (+59%, P=0.002, Supplementary Figure 9b).

Both lithium and BIO are reported to decrease total tau protein levels in addition to their inhibitory effects of GSK in cultured neurons.42 In agreement with previous reports,41, 42 we found that lithium (⩾5 mM, 18 h) suppresses tau protein levels in primary neurons (5 mM Li: −48%, P=0.007; 10 mM Li: −35%, P=0.034; Figure 3d) and in SH-SY5Y cells (5 mM Li: −25%, P=0.003; 10 mM Li: −25%, P=0.002; Supplementary Figures 10a and b). BIO (18 h) also suppressed tau protein levels in primary neuronal culture (1 μM BIO: −25%, P=0.043; 2 μM BIO: −26%, P=0.048; Supplementary Figures 10c and d) that may be the mechanism for its ability to increase neuronal iron levels.

To demonstrate the requirement for tau in the mechanism of lithium-induced iron accumulation in neurons, lithium (10 mM, 18 h) was administered to tau-knockout primary neurons and compared with background-matched wild-type neurons (Figure 3e). As previously shown,43 tau-knockout neurons retained more iron (+20%, P=0.032) compared with wild-type neurons. As expected, lithium treatment (10 mM, 18 h) of wild-type neurons induced a significant accumulation of iron (+20%, P=0.038), but lithium treatment of tau-knockout neurons induced no further iron accumulation. Thus, lithium-induced neuronal iron accumulation requires the expression of tau protein.

Tau-knockout and APP-knockout mice are resistant to neurodegeneration caused by chronic lithium administration

Tau mediates neuronal iron efflux by facilitating the trafficking of APP to the cell surface, where it stabilizes the iron export channel ferroportin.43, 52, 53, 54 To test whether lithium-induced iron accumulation requires APP, we applied the same lithium treatment to tau-knockout mice, APP-knockout mice and their background-matched wild-type mice. We studied mice at 3 months of age, because the knockouts have yet to develop the age-dependent elevation in brain iron levels.43, 44

All three strains showed no weight reduction during 21 days of lithium treatment (Figure 4a), even though APP knockouts were lighter in weight (as reported previously78). Lithium treatment predictably caused an elevation in mouse plasma lithium levels (wild type: +6.6-fold, P<0.001; tau knockout: +6.7-fold, P<0.001; APP knockout: +4.3-fold, P=0.007; Figure 4b), although the levels in all three strains remained well below the lower limit of the therapeutic concentration range.40 As with the Bl6/C57J mice (Figure 2c), lithium treatment induced iron elevation in both the cortex and SN of the littermate Bl6/129Sv mice (cortex: +15%, P=0.031; SN: +20%, P=0.041), but despite lithium elevations in brain regions being commensurate to those in Bl6/C57J mice (Figure 2b), tau- and APP-knockout mice were unaffected (Figures 4c and d), consistent with the neuronal culture findings (Figure 3e). Brain copper or zinc levels were again unaffected (Supplementary Figures 11a–d), and iron levels in the cerebellum and liver were unaltered (Supplementary Figures 11e and f), by lithium treatment in all three strains.

Figure 4
figure 4

Lithium (Li)-induced parkinsonism is mediated by tau and amyloid protein precursor (APP). (a) Daily mean weights of the mouse cohorts revealed no significant impact of lithium treatment within each genotype group during the course of study. APP-knockout mice were lighter, as reported previously.78 (b) Li treatment significantly elevated plasma lithium levels of 3-month-old wild-type (P<0.001, two-way analysis of variance (ANOVA) with post hoc Dunnett’s test), tau-knockout (P<0.001) and APP-knockout (P=0.007) mice. These concentrations are well below the usual therapeutic range of clinical lithium therapy. (c, d) Elevated iron levels were observed in the cortex (c, P=0.031, two-way ANOVA with post hoc Dunnett’s test) and substantia nigra (SN; d, P=0.041) of Li-treated wild-type mice, but loss of tau or APP abolished the elevation. (e, f) Li treatment induced tau reduction in cortex and SN of both wild-type (e, P=0.041 for cortex, P<0.001 for SN, two-tailed t-test) and APP-knockout mice (f, P=0.006 for cortex, P=0.041 for SN, two-tailed t-test). (g, h) Li-treated wild-type mice required longer time to turn in the Pole test (g, P=0.020, two-way ANOVA with post hoc Dunnett’s test) and showed significantly reduced average distance per movement in the open field test (h, P<0.001), but loss of tau or APP protected against Li-induced motor impairment. (i) Li treatment caused SN dopaminergic neuron loss (P<0.001, two-way ANOVA with post hoc Dunnett’s test), but did not cause similar neuronal loss in tau-knockout or APP-knockout mice. (j) Li treatment impaired cognitive function of wild-type mice, evidenced by significantly reduced novel arm duration in the Y-maze test (P=0.022, two-tailed t-test). Such impairment was not evident in tau-knockout and APP-knockout mice. (k) Representative coronal section of cerebrum of wild-type mice treated with sham (left) or Li (right). Arrow indicates the landmark used to identify the section. Lines indicate the area quantified. (l) Significant enlargement of lateral ventricular (LV) was found in Li-treated wild-type mice (P=0.023, two-way ANOVA with post hoc Dunnett’s test), but not in Li-treated tau-knockout or APP-knockout mice; n=9–11 per treatment group. Mean±s.e.m. are shown. *P<0.05, **P<0.01, ***P<0.001.

PowerPoint slide

Full size image

Ablation of APP abolished lithium-induced iron accumulation (Figures 4c and d), despite soluble tau protein levels in cortex and SN of APP-knockout mice treated with lithium being suppressed comparably to wild type (cortex: −19%, P=0.041 for wild type and −41%, P=0.006 for APP knockout; SN: −42%, P<0.001 for wild type and 21%, P=0.041 for APP knockout; Figures 4e and f and Supplementary Figure 12). These results confirm our prediction that lithium-induced brain iron accumulation requires both tau and APP expression, with tau being upstream in the APP-associated iron homeostatic pathway.

As lithium-induced Parkinson-like symptoms such as hand tremor may be mediated by iron accumulation in the brain, disabling lithium-induced SN iron accumulation in tau- and APP-knockout mice (Figure 4d) should also prevent motor deficits and SN neuronal loss. Indeed, lithium-induced motor deficits were abolished in tau- and APP-knockout mice, as evidenced by unaffected time to turn (wild type, P=0.020; Figure 4g) and time to finish (wild type, P=0.048; Supplementary Figure 13a) in the Pole test, and unaltered average distance per movement (wild type, P=0.049; Figure 4h), distance of locomotion (wild type, P=0.020; Supplementary Figure 13b) and velocity (wild type, P=0.048; Supplementary Figure 13c) in the open field test. Untreated APP-knockout mice did show baseline motor deficits (Figures 4g and h and Supplementary Figures 13a–c), as previously reported,78 but these were not worsened by lithium. Tau or APP ablation also negated SN dopaminergic neuron loss induced by lithium treatment (wild type, P<0.001; Figure 4i and Supplementary Figure 14a). Together, these results support the conclusion that lithium-induced parkinsonism is mediated by iron accumulation that which is tau and APP dependent.

As lithium treatment also increases cortical iron levels (Figures 2c and 4c), we investigated its effect on cognitive performance and related brain anatomy. In the Y-maze test, we found the memory performance of wild-type mice was markedly impaired by 21 days of lithium treatment (P=0.022) but, as with the Pole test (Figure 4g), tau- and APP-knockout mice were unaffected by lithium (Figure 4j). Accordingly, lithium treatment caused brain atrophy in wild-type mice, evidenced by enlarged lateral ventricular area (+72%, P=0.023; Figures 4k and l), shrinkage of CPu area (−13%, P=0.011; Supplementary Figure 14b) and reduced cortical thickness (−7%, P=0.040; Supplementary Figure 14c), without affecting corpus callosum width (Supplementary Figure 14d). Again, lithium did not induce brain atrophy in tau- or APP-knockout mice (Figure 4l and Supplementary Figures 14c and d). These toxic consequences of lithium administration to cortical thickness are in agreement with a previous report that found cortical neuronal apoptosis in mice upon chronic lithium treatment.17

Discussion

Lithium has an established record of efficacy in bipolar affective disorder, despite a narrow therapeutic window limited by various toxicities of uncertain mechanism. Here we found that lithium may contribute to nigral and cortical degeneration, with motor and cognitive impairment in wild-type mice, recapitulating the neurophysiological and neurochemical phenotypes of aged tau-knockout mice.43, 44 Lithium suppresses tau protein levels in cultured neurons and in mice, causing iron accumulation through a tau- and APP-dependent interaction. The T2 relaxation time changes in the human lithium trial are consistent with iron elevation in hippocampus and SN. Although homeostasis keeps brain iron levels stringently stable in health, and geared to resist depletion,79 rodent studies have shown that nearly 10% of brain iron is in relatively rapid exchange with dietary iron,52, 80 indicating that in order to prevent iron accumulation, iron efflux from the brain must be constantly active. It is this mechanism that lithium inhibits by, at least in part, suppressing tau expression. Tau-knockout mice develop iron-induced neurodegenerative changes at 12 months of age43, 44 but lithium-induced toxic iron accumulation occurs within weeks. This is probably because the tau-knockout mice in youth have the benefit of upregulated compensatory responses from other microtubule-associated proteins45 in contrast to the more acute suppression of tau expression by lithium treatment. Our findings add to the evidence that lithium at therapeutic (and eventually even at subtherapeutic) chronic doses can cause neurodegeneration,17 and provide a possible mechanism for adverse neurological effects in susceptible individuals.

Parkinsonism is an uncommon adverse effect of lithium treatment, but hand tremor is common and may be age and dose dependent.81, 82, 83 In addition, lithium use was found to be associated with an increased incidence of anti-parkinson drug use or a PD diagnosis.16 The mechanisms underlying these motor abnormalities are not known, but could be because of persistent or transient extrapyramidal syndromes caused by damage to the basal ganglia and SN. Iron may play a critical role in promoting SN pathology in PD,55, 84 and our findings demonstrate how lithium could worsen iron-mediated SN damage. SN iron accumulation in trial subjects may have occurred after only 3 months of low-dose lithium treatment (Figure 1b), although no conspicuous motor abnormalities were reported in this group. In addition, our findings in mice support the possibility that lithium-induced parkinsonism can become irreversible in a proportion of patients.6, 7, 8, 11, 15

Unexpectedly, we found that lithium treatment at doses inducing blood levels well below the human therapeutic range caused cognitive impairment and brain atrophy in mice after 3 weeks (Figures 4j–l). It has been shown previously that lithium treatment at therapeutic doses can cause cortical neuronal apoptosis,17 and this may explain our findings. Although lithium treatment was previously reported to correct the smaller hippocampal volumes in patients with bipolar disorder,85 there is also evidence that long-term lithium treatment is associated with cognitive impairment86, 87, 88, 89 and brain atrophy.89 Neurodegeneration in mice administered lithium to achieve blood levels that are easily tolerated by humans may reflect different pharmacodynamics in the species, or that humans are less sensitive than mice to lithium-induced neurotoxicity.

Lithium-induced motor deficits in mice are reported to involve the calcineurin/nuclear factor of activated T cells (NFAT)/Fas ligand signaling pathway, and prevented by inhibition of calcineurin or knockout of Fas.17 Iron modulates calcineurin activity90, 91 and can thereby impact the NFAT/Fas signaling pathway. Treatment with iron dramatically elevates NFAT activity,92, 93 whereas administering the iron chelator deferoxamine abolishes iron-induced NFAT activation.92, 93 Together with our current data, a proposed model pathway emerges (Figure 5) where lithium suppresses tau protein levels, thus preventing APP trafficking to the cell surface and resulting in neuronal iron retention. In turn, this activates calcineurin/NFAT/Fas signaling induces cortical and nigral neuronal apoptosis and engenders motor and cognitive disability.

Figure 5
figure 5

A model of affected pathways for lithium-induced neurodegeneration. Lithium (LI) treatment suppresses soluble tau protein levels that prevent APP trafficking to the neuronal surface. Lack of surface amyloid protein precursor (APP) destabilizes ferroportin (Fpn), resulting in iron accumulation. Elevated iron activates calcineurin/nuclear factor of activated T cells (NFAT)/Fas signaling that then induces neuronal apoptosis, engendering motor and cognitive disability.

PowerPoint slide

Full size image

Lithium was an attractive drug candidate for AD as its target, GSK-3, regulates both Aβ production and tau phosphorylation.5 However, it is neither a potent nor specific inhibitor of GSK-3.94 In fact, neuronal deficits have been reported to result from both genetic reduction and specific inhibition of GSK-3,95, 96 highlighting the risk of inhibiting this enzyme.97 Lithium arrests neuropathology in some AD rodent models,36, 38, 39 but these are aggressive models driven by transgene overexpression, where the benefits of lithium inhibiting a dominant pathological target (for example, tau hyperphosphorylation) outweigh lithium-induced neurotoxicity. The lithium-induced neurotoxicity that we and others17, 39 observed might only be detected in normal mice or mice that do not express a neuropathology that is inhibited by lithium.

Our current study might help explain the failure of trials of lithium for AD40, 98 and amyotrophic lateral sclerosis,99, 100, 101 both diseases where elevated iron in the affected tissue has been implicated in pathogenesis.52, 102, 103, 104 Despite the neurotoxicity in mice that our studies have elaborated, clinical lithium pharmacotherapy has predominantly neurotrophic benefits in the treatment of bipolar disorder. It is an intriguing hypothesis to consider that these benefits might be related to the correction of iron deficiency in certain brain regions.