Behind the Scenes: Lithium Removal with Household Water Purification Devices

Lithium is an element, atomic number 3. It is a soft, light, highly reactive metal with a variety of uses. Among other things, it’s often found as a trace mineral in drinking water. Small amounts of lithium are naturally present in many water sources, but levels of lithium in American drinking water have been increasing for the past 60 years.

In 1964, the US Department of the Interior published a report called Public water supplies of the 100 largest cities in the United States, which found a median lithium concentration of only 2.0 µg/L in US drinking water. The highest level they recorded was 170 µg/L. 

In 2021, the USGS released a report that found a median level in US groundwater of 6.9 µg/L. This is almost four times the median level in the 1960s, but looking at nothing but the average obscures the fact that many people are getting exposed to even more. For comparison, the maximum level they found in groundwater was 1700 µg/L, ten times the maximum recorded in 1964. 

The USGS also found that about 45% of public-supply wells and about 37% of domestic-supply wells contain concentrations of lithium “that could present a potential human-health risk per the current EPA guidelines”. Here’s how they describe it in the paper:

Lithium concentrations in untreated groundwater from 1464 public-supply wells and 1676 domestic-supply wells distributed across 33 principal aquifers in the United States were evaluated for spatial variations and possible explanatory factors. Concentrations nationwide ranged from <1 to 396 μg/L (median of 8.1) for public supply wells and <1 to 1700 μg/L (median of 6 μg/L) for domestic supply wells. For context, lithium concentrations were compared to a Health Based Screening Level (HBSL, 10 μg/L) and a drinking-water only threshold (60 μg/L). These thresholds were exceeded in 45% and 9% of samples from public-supply wells and in 37% and 6% from domestic-supply wells, respectively.

Levels in drinking water seem to have increased due to a number of related factors, including the use of drilled wells to tap deeper aquifers, higher levels of fossil fuel prospecting and pollution, and the fact that worldwide lithium extraction and use in industrial applications has increased in general, multiplying the opportunities for accidental exposure and pollution. 

Lithium is not currently regulated in drinking water, and water quality reports don’t regularly include it. Most water treatment plants do not track lithium or attempt to reduce it. But the EPA and other government agencies are becoming more concerned about lithium exposure, even at the trace levels found in drinking water: 

Just this January, lithium was added to the EPA’s proposed Unregulated Contaminant Monitoring Rule. The Rule is used by the EPA to collect data for contaminants that are suspected to be present in drinking water and that do not have health-based standards set under the Safe Drinking Water Act.

Although useful for treating mental health disorders, pharmaceutical use of lithium at all therapeutic dosages can cause adverse health effects—primarily impaired thyroid and kidney function. Presently lithium is not regulated in drinking water in the U.S. The USGS, in collaboration with the EPA, calculated a nonregulatory Health-Based Screening Level (HBSL) for drinking water of 10 micrograms per liter (µg/L) or parts per billion to provide context for evaluating lithium concentrations in groundwater. A second “drinking-water-only” lithium benchmark of 60 µg/L can be used when it is assumed that the only source of lithium exposure is from drinking water (other sources of lithium include eggs, dairy products, and beverages such as soft drinks and beer); this higher benchmark was exceeded in 9% of samples from public-supply wells and in 6% of samples from domestic-supply wells.

Lithium is well-known to have psychoactive effects, which is why lithium salts are often prescribed as a psychiatric medication. In particular, lithium tends to make people less manic and less suicidal. Less charitably, it is sometimes described as a sedative. 

But these effects may not always require psychiatric doses. A long-running literature of epidemiological research (meta-analysis, meta-analysis, meta-analysis) suggests that long-term exposure to trace levels of lithium commonly found in drinking water can also have psychiatric effects. Specifically, trace levels in drinking water are often found to be associated with decreased crime, reduced suicide rates, and/or decreased mental hospital admissions. 

Finally, here at Slime Mold Time Mold we suspect that lithium exposure may contribute to the obesity epidemic. Lithium often causes weight gain at psychiatric doses, and while there’s no smoking gun yet, there’s some evidence that there might be a connection between long-term trace lithium exposure and obesity. People who are exposed to more lithium, especially at their jobs, tend to be more overweight. Cities with higher rates of obesity tend to have more exposure to lithium. And a group of Native Americans (the Pima) who had unusually high levels of lithium in their water also had unusually high levels of obesity, all the way back in the 1970s. Food levels may also be a possible vector (though it’s complicated).

So people often ask us, how can I get lithium out of my tap water? 

For a long time, we weren’t able to answer this question. Until very recently, no one was concerned about lithium levels in drinking water, so there isn’t much research on how to get it out. Heck, back in 2014 the NYT ran an opinion piece arguing that maybe we should start putting lithium in our drinking water. How times have changed.

This is further complicated by the fact that lithium is pretty weird. At an atomic number of only 3, it is the third-lightest and third-smallest element. In some ways it is more like the gasses hydrogen and helium than it is like the metals iron, lead, or mercury, which are much larger and much heavier. This makes it hard to predict whether techniques that can remove other metals would also remove lithium, which is present in solution as an especially tiny ion. 

(A favorite “Whoaahhh” fact about Li+ is that it is so small, a bit of electrical energy can make it can creep into the crystal lattices of other compounds and basically just hang out there indefinitely, usually with a bit of swelling of the “host” crystal. It’s kind of like pouring sand into a jar of marbles — lithium is so tiny it can sneak into very small spaces, which is virtually impossible with any other metal ion. The technical term is that it “intercalates” into these materials. Lithium intercalating back and forth between cobalt oxide and graphite, for instance, is the basis of the lithium ion batteries that power virtually every phone and laptop and electric vehicle. There’s an entire field of research focused on making lithium creep into and out of various materials to store energy. People have been trying for a long time to make Na+ do this, since Na+ is so much cheaper and more abundant than Li+, but it’s still way too hard to make any kind of useful battery with an ion as big as Na+.)

To answer the question of how to get lithium out of your drinking water, we set up a project with research nonprofit Whylome to test several commercially-available water filters, the kinds of things you might actually buy for your home, and see how good they are at removing lithium. It’s taken a couple of months of planning, testing, and analysis, but those results are finally ready to share with the world.

This project was funded by generous donations to Whylome from individuals who have asked to remain anonymous. Further support for the research was provided by The Tiny Foundation, which allowed us to expedite several aspects of the research. Special thanks to our funders, Sarah C. Jantzi at the Plasma Chemistry Laboratory at the Center for Applied Isotope Studies UGA for analytical support, and to Whylome for providing general support. 

The full report is here, the raw data are here, and the analysis script is here. Those documents give all the technical details. For a more narrative look, read on. 

TABLE OF CONTENTS

  1. Methods
  2. Results
  3. Complications
  4. Conclusions

1. Methods

The basic idea of the study is pretty simple.

You buy a bunch of normal water filtration devices (henceforth “filters”, even though they’re technically not all filters) from a store, like Home Depot, or online, from places like Amazon. Or online from Home Depot.

You spike large quantities of water with specific amounts of lithium, to get water containing known levels of lithium.

Then, you run the lithium-spiked water through the filters and take samples of the water that comes out the other end. 

Finally, you submit that water to chemical analysis and find out how much lithium was removed by each of the filters. 

This is basically the perfect garage experiment — except that in this case, filters were tested in the laundry room, not in a garage.

1.1 Water Filtration Devices

To get a sense of the different options available on the market, we elected to test three different types of devices: carbon filters (which are what most people think of when they think of at-home filters); reverse osmosis devices; and electric water distillation stills. 

We chose devices from brands that most people have heard of, and models that people tend to buy. If you click through the links below, you’ll see that many of these devices are best-sellers.

We settled on the following mix of carbon filters: two pitchers, the Brita UltraMax Filtered Water 18-Cup Pitcher and the PUR Ultimate Filtration Water Filter Pitcher, 7 Cup; three on-tap systems, the Brita 7540545 On Tap Faucet Water Filter, the PUR PLUS Faucet Mount PFM350V, and the Culligan Faucet Mount FM-15A; and two under-sink systems, the Waterdrop 15UA and the Brondell Coral UC300.

We settled on two reverse osmosis devices, the GE GXRQ18NBN Reverse Osmosis Filtration System and the APEC ROES-50 5-stage Reverse Osmosis System.

We also tested two distillation machines, the Megahome 580W Countertop Water Distiller and the Vevor 750W Water Distiller

Devices were purchased off of Amazon, from the Home Depot, or from their manufacturer, depending on availability. For each device, we also purchased as many extra filters as needed, so that each test could start with clean filters (see the report for more detail).

Carbon Filters — We came into this pretty confident that carbon filters would perform very poorly for lithium removal, despite some nonsense to the contrary floating around the internet (for example, here and here). Carbon has a low affinity for Li+, so we didn’t expect it would pull very much out of the water. Carbon can remove some metals, like lead, by ion exchange — the same principle used in water softeners. But the metals it is good at removing are multivalent (having a charge of +2 or +3 or +4), not +1 like Li+.

Carbon is also known for having noticeable variation between individual filters, because the carbon in question is made from plant material (often coconut). There will be minor variations in the carbon properties between batches, depending on how fast the coconuts were growing that month and minutiae like that. So we went into this expecting that there might be some differences between different filters, even within the same brand and/or model. 

Since we expected that carbon filters would probably all suck based on the mechanism of action, and because we expected that there might be noticeable variation, we decided to test several different brands of carbon filters, in multiple configurations (pitcher, on-tap, and under-sink). This is why we tested so many devices and why we got a relatively wide mix of brands and configurations.

This way, if carbon filters are all equally ineffective, it should be very obvious. But if we’re wrong and they’re great, or some are much better than others, we have a good chance of noticing. Carbon filters are also the cheapest and most commonly used filters, another reason to test more of them.

We expected less variation in the other two kinds of devices, so we decided to test two models of each. 

Distillation — We expected that distillation machines would work well, but we didn’t know if that was 80% well, 90% well, or 99.9% well. Lithium salts have zero volatility, so when water evaporates and condenses, the lithium should be left behind. The main risk is that droplets of liquid could get caught in the condenser, which could result in some of the original liquid getting into the clean distillate. So a well-designed distillation machine should perform well, but we didn’t know how reliable or well-designed small at-home countertop models would be.

Reverse Osmosis — We were the most uncertain about reverse osmosis. Reverse osmosis is very good at removing divalent metal ions (like Ca2+ and Mg2+), and pretty effective at removing monovalent metal ions from tap water (like Na+ and K+), but it wasn’t clear if this pattern would extend to lithium. In some ways Na and K are very similar to lithium — all three are present in water as single-charge positive ions, and all three are the same chemical group, the alkali metals. But lithium is much smaller and lighter than other elements. Na has an atomic number of 11, and K has an atomic number of 19, while Li has an atomic number of only 3. 

As a result, we weren’t sure if reverse osmosis would be anywhere near as effective at removing lithium as it is at removing these other contaminants. Maybe reverse osmosis would pull lithium out of the water just like any other ion. Maybe it would miss lithium entirely, because the ion is so small. Or maybe something in between. So we went into this expecting that reverse osmosis might be anywhere from 0% to 100% effective. 

1.2 Lithium Spiked Water

For realism, we worked with actual American tap water. In this case, we used tap water from the town of Golden, Colorado. Despite the fact that it was indeed part of the Colorado Gold Rush, Golden, CO is not named after the gold rush or even after gold itself; it is named after some guy named Tom Golden.

Samples of the tap water were spiked with known quantities of “ultra dry” lithium chloride salt to create spiked water samples of known lithium concentration.

We ended up testing four concentrations of lithium: 40, 110, 170, and 1500 µg/L Li+. This covers a range from “starts to be concerning” to “around the highest levels reported in US drinking water”. There’s also a bit more history to these numbers, but we’ll talk about that below. 

1.3 Testing 

Each filter was tested at each concentration, and at two timepoints (realistically these are “volumepoints”, but that’s not really a word). The carbon filters and the RO devices were each tested after 10 liters and after 20 liters. The distillation machines were tested at 2 liters and again at 4 liters, since they take a really long time to run. 

The testing setup looked roughly like this: 

1.4 Analysis

At the start of the project, we sent the same samples to a couple different testing labs, so we could shop around and compare. All the labs we tried were pretty reliable, but the Plasma Chemistry Laboratory at the Center for Applied Isotope Studies, University of Georgia stood out as the best, so we sent all subsequent samples to them. 

Analysis was performed by ICP-OES. The instrument used was a Perkin Elmer 8300 ICP-OES, and the limit of detection was 1 µg/L. All analyses were done in triplicate and were submitted in a random order.

 

2. Results

The following figure gives an overview of the results. This figure only includes performance at a concentration of 110 µg/L after the first timepoint (2L for distillation, 10L for the others) but the same general pattern holds across pretty much everything: 

2.1 Carbon Filters

Carbon filters are lousy at removing lithium, but probably not 0% effective. Most of the time, water contained slightly less lithium coming out of the filter than it did going in. But the carbon filters didn’t do much, and there wasn’t a huge amount of variation between them.

2.2 Reverse Osmosis

Reverse osmosis was shockingly good at removing lithium. Removal was reliably high for all systems, more than 80% for the GE system and consistently above 95% for the APEC system. The result is unequivocal: reverse osmosis works. Reverse osmosis does not, however, drive these concentrations close to zero. RO is good, but if you start with 100 µg/L in your tap water, you might still end up drinking 10 µg/L even after filtration. 

In many cases you do end up with less than 10 µg/L after filtration, but if you start with a high concentration, you are still generally getting more lithium than was in the median American water source in 1964 (2 µg/L). The lower your starting lithium, the lower the lithium concentration you are getting out of your RO filter.

2.3 Distillation

Finally, distillation machines are nearly perfect at removing lithium. Lithium levels after distillation were undetectable (<1 µg/L) in most cases, and removal was still >99.5% for the highest concentration (1500 µg/L). Distillation reliably drives any levels you would expect to see in American tap water below the level of detection. 

2.4 Long-Term Reverse Osmosis Test

We also decided to do one long-term test of a single system, to check if it kept performing well over a longer period of time, and to see if anything weird happened. We expected that systems would get slightly worse over time, but there might also be a discontinuity, where a system keeps doing well for a while and then suddenly craps out and does much worse. We wanted to see how much decline happened with more use, and check if there was any discontinuity or sudden point of failure. 

Carbon filters don’t work very well even straight out of the box, so obviously we didn’t test one of those. RO doesn’t remove lithium from water quite as well as distillation, but it’s faster, cheaper, and much easier to install. Because RO sits at this sweet spot, we decided to test the GE RO device up to 100 liters. 

We tested the GE RO device against a concentration of 170 µg/L, and the device continued to do a good job removing lithium even up to 100 L. Performance went down slightly over time, but not enormously. At 10 L, the device removed about 98% of the lithium in the water, and by 100 L, it removed about 89%. We don’t know how well it would perform beyond 100 L, but this finding suggests it would keep doing pretty well but progressively worse over time. 

This would be a good topic for further study — run a few RO devices to 1000 L and see what happens. Alternately, you could install a RO device in the home of someone whose tap water is already high in lithium, test its effectiveness once a month, and get a sense of how these devices would perform in a real-world scenario. 

3. Complications

The conclusions from this study are, fortunately, pretty straightforward. But on the way to those conclusions, there were a few complications.

3.1 PUR Pitcher

In addition to the six carbon filters mentioned above, we also tested the “PUR Ultimate Filtration 7-Cup Pitcher”. When we ran it through the same procedure as the other filters, we found there was more lithium in the filtered water than in the original water, at all concentrations. Basically it seemed like the PUR pitcher was adding lithium to the water instead of taking it away. 

This was confusing and seemed like it might be wrong, so we tried the same pitcher again with a different set of filters. This time we didn’t get the weird result — lithium levels went down when we ran water through the filter, just like normal. 

We’re not totally sure why this happened. One possibility is that some of the water evaporated during testing, but letting the water sit for a few days didn’t make a substantial difference compared to filtering rapidly, so this appears unlikely. Another possibility is that there’s meaningful batch-to-batch variation in the lithium content of the filter cartridges. Activated carbon comes from plants (usually coconuts), so conceivably there could be more lithium in some coconuts than in others. If you got unlucky, the carbon might contain a lot of lithium and you would end up adding lithium to the water instead of taking it away. 

In any case, this was strange and inconclusive enough that we ended up removing it from the main analysis, but we’re reporting it here just in case. Good cautionary tale about how even a simple measurement is never simple. 

3.2 Concentration Complication

We originally planned to test lithium concentrations of 10, 60, 100, and 1000 µg/L.

The reasoning was that 10 and 60 µg/L were the EPA thresholds of interest, and that testing 100 and 1000 µg/L covered two further orders of magnitude while still being realistic — according to the USGS, 4% of groundwater wells in the US contain more than 100 µg/L lithium, and the maximum recorded contained 1700 µg/L.

But two things happened to screw that up. 

First, the tap water in Golden gave us a bit of a surprise. Golden is a city in Colorado, and most tap water in Colorado comes from dazzlingly clean snowmelt. Snowmelt should contain almost no lithium (it’s basically been distilled), so we expected that the tap water in Golden would also contain almost no lithium. This assumption was backed up by water quality reports from nearby Denver, CO, which find no lithium in Denver’s water. 

But to our surprise, when we started testing samples, we found that they contained more lithium than we spiked them with. We circled back and tested the unspiked tap water, and found that it contained around 20-25 µg/L, an amount that was reliable across several months. If there are seasonal changes, our January-March sampling window wasn’t big enough to detect them.

The local water treatment plant is fed by Clear Creek, so we collected and tested a sample from the creek about 2 miles upstream from the water treatment plant. The creek there has a concentration of 27 µg/L, very similar to the tap water. It appears that water enters the Golden, CO treatment plant at around 25 µg/L, and the treatment process has very little impact on lithium concentration.

At this point we were questioning our assumptions about water sources, so we collected some local snow and tested that too. The snowmelt had barely detectable lithium, less than or equal to 1 µg/L. This confirms our earlier belief that precipitation is generally very low in lithium (at least in Golden, CO).

If it’s not in the snowmelt, the lithium must be coming from somewhere else. This is speculation, but the Clear Creek watershed does include many abandoned mines, some dating way back to the early gold and silver rushes from the 1800s, and there is at least one Superfund site, so old mine tailings are one possibility (see in particular here). One of the towns upstream (Idaho Springs) has natural hot springs with some geothermal activity, so another possibility is that these springs add lithium to Clear Creek along the way. We didn’t find an obvious link for Idaho Springs, but other hot springs in Colorado definitely brag about the lithium content of their water (Denver Post on Orvis Hot Springs: “The resort’s seven pools are laden with lithium…”), so this seems quite plausible.

This suggests that our original assumptions were mostly correct — snowmelt contains little to no lithium, so most drinking water in Colorado should be quite pure. But in this specific case, looking at water drawn from Clear Creek, we ended up with more than we expected. Water coming from one of Colorado’s snowmelt reservoirs, rather from a well or stream, would probably contain a lot less.

In the end, the lithium levels in Golden’s tap water raised the lithium level of all of our samples by about 25 µg/L. We were already halfway through testing when we discovered this, so we decided to continue with these slightly higher concentrations. If anything, it’s a stricter test of the filters.

Clear Creek, circa 1868

Second, the lithium salt we used was substantially more potent than the stated strength (i.e. much stronger than expected), which also increased the concentrations we tested. 

We used lithium chloride from Fisher Scientific as the lithium spike for all our samples. According to the certificate of analysis, the salt contained a lot of water. But apparently this was not the case. As far as we can tell, the salt appears to have very little water content, so it contains a lot more lithium per weight than expected (about 30% stronger than expected). This caused us to underestimate the amount of lithium in the salt, and as a result, we added more than we meant to. This is why we ended up testing up to 1500 µg/L.

Again, we were already halfway through testing when we discovered this, and decided to forge ahead. Because this error was propagated across all the samples we had submitted, the analysis was still internally consistent. Even though these weren’t the numbers we had set out to study, it doesn’t really matter. Those numbers were arbitrary to begin with; we chose them because we live in a base-10 world. We were still able to compare between filters at realistic concentrations.  

Together, these two factors inflated the concentrations we tested, from 10, 60, 100, and 1000 µg/L to 40, 110, 170, and 1500 µg/L. First, the tap water from Golden added 25 µg/L to all the samples. Then, the unusually dry lithium salt inflated the amount added to each sample by around 30%.

Fortunately, this does not seriously impact our results. Filters were consistent across all concentrations, and in the end we covered a very similar range, 60-1500 µg/L instead of 10-1000 µg/L. We’re only really missing an analysis of how well the filters would work at low levels, around 10 µg/L. But RO devices that drive 40 µg/L to around 1 µg/L can also be expected to drive 10 µg/L way down low.

The only thing we would want to revisit in future studies is to test carbon filters at levels close to 10 µg/L; but our best bet is that they don’t do much at those levels either.

3.3 Doubles

We also caught one other problem. During analysis, we found that we made a mistake when mixing four of the concentrations. Twice as much lithium chloride as intended was added to the solutions for the PUR faucet mount at concentrations of 110 and 170 µg/L, and also for the Culligan faucet mount at 110 and 170 µg/L. As a result, these two filters were actually tested against ~210 µg/L and ~325 µg/L instead of the intended 110 and 170 µg/L. You can easily see this error if you look at the tables in the report. 

This is unfortunate and does complicate the data, but again it doesn’t seriously change the conclusions. Carbon filters don’t get much lithium out of tap water at any concentration, whether it’s 110, 170, 210, or 325 µg/L. There’s no reason to expect that the PUR and Cullighan faucet mounts would perform differently at these concentrations than at the intended ones — these results fit the overall result, which is that carbon filters aren’t good at removing lithium. 

3.4 Why’d You Have To Go And Make Things So Complicated? 

You may not be used to seeing scientific papers talk about mistakes the research team made, or the incorrect assumptions that showed up halfway through the project, or the weird random anomaly that doesn’t have an easy explanation. But the truth is that this is just what research looks like.

Academic researchers are expected to pretend like everything went perfectly and nothing weird happened, but this is not how actual research projects work. In real projects, especially where you’re trying to advance the frontiers of knowledge, you have to take chances, make mistakes, and yes, even get messy.

There are always going to be some accidents in any research project, and instead of sweeping them under the rug and pretending we never make mistakes, we’re going to talk about them. This not only is virtuous, it also puts you (readers) in a better position to form your own opinion about our results. It gives you a better sense of what to expect if you want to replicate or extend our results. And if we didn’t tell you about all the SNAFUs, we’d be giving you the wrong idea about what research is really like. 

And of course, it’s possible there are other mistakes we haven’t caught yet! We know that the best way to troubleshoot is to get as many eyes on the project as possible, which is why we put all our data and code online for you to see.

Obviously we want to avoid mistakes when we can, which is why we use techniques like randomizing sample order and including control samples to help prevent and diagnose mistakes. But this sort of thing happens, and it’s in everyone’s best interest to just publicly say “whoops, our bad”.

4. Conclusions

If you have the time and money, distillation is the best way to get lithium out of your water. The catch is that distillation is slow: distillation machines usually run at less than 1 liter per hour, a small fraction of the speed of other devices, and consume a lot of energy to get there. Distilling all of your cooking and drinking water with one of these machines would be very slow or very expensive or both.

For the average consumer, reverse osmosis is a much better choice. It’s cheaper and faster, and it works nearly as well as distillation does. For the average American, a RO system will ensure that you end up with less than 10 µg/L in your water, probably much less. 

Both of the RO systems we tested were under-sink units, meaning they go under your sink (duh) and create a stream of purified water that is separate from the actual tap. That way you wash your dishes with the high flow rate you’re accustomed to from a faucet, but fill your glass or make pasta with the separate stream of RO-filtered water.

You could also spring for a professional-grade household system that filters all the water that comes into your house, but there are a few complications. First off, while it should work basically the same as these under-sink units, we didn’t actually test a household system. Second, it’s got a much higher upfront cost and it would be more of a pain to install and maintain. Also keep in mind that typically only 20-50% of the water entering the RO unit actually leaves as clean, filtered water; the rest never makes it through the filter membrane and goes down the drain. Throwing away that much water for things like showering or washing your car would mean a lot of wasted water.

Finally, a whole-house RO system typically needs to be accompanied by a water softener, and we’re not sure if water softeners contain lithium or not. Water softeners operate by ion exchange, exchanging one Ca2+ or Mg2+ ion for two Na+ ions. You “regenerate” the system every so often by dumping a big bag of rock salt (NaCl or occasionally KCl) into the “brine tank”, which displaces the Ca/Mg off of the ion exchanger. If the salt being used for regeneration contains lithium, it would make its way into the drinking water just as readily as Na+. We haven’t tested any water-softening salt yet (though we might at some point), but we did test table salt as part of another project, and that definitely contains some lithium. 

Because of this, it’s not clear whether you’d end up drinking more or less lithium if you install a household RO system with a water softener. If you’re using a water softener without a RO system, you’re probably adding some lithium to your water, though we’re not sure how much. 

If you purchase water that was treated by RO or distillation (as many bottled waters are), it’s probably very low in lithium. But the catch here is that many companies put minerals back in, because pure water actually tastes kind of flat and metallic. Aquafina, for example, is first purified through RO before putting a pinch of salt back in for taste. If the pinch of salt contains lithium, you’re back to square one.


Thanks again to our anonymous donors, the Tiny Foundation, Sarah Jantzi, and Whylome for supporting this research. Finally, thank you for reading!

Total Diet Studies and the Mystery of ICP-MS

After our recent post on Lithium in Food, several readers pointed us to a literature on “Total Diet Studies”, or TDS for short.

The TDS approach is pretty intuitive: if you want to study contaminants or residues that people are maybe exposed to through their food, one way to do that is to drive around to a bunch of actual grocery stores and supermarkets, buy the kinds of foods people actually buy and eat, prepare the foods like they’re actually prepared in people’s homes, and then test your samples for whatever contaminants or residues you’re concerned about. 

Or in the words of a review paper on the Total Diet Study approach from 2014:

A Total Diet Study (TDS) generally consists of selecting, collecting and analysing commonly consumed food purchased at retail level on the basis of food consumption data to represent a large portion of the typical diet, processing the food as for consumption, pooling the prepared food items into representative food groups, homogenizing the pooled samples, and analysing them for harmful and/or beneficial chemical substances (EFSA, 2011a). From a public health point of view, a TDS can be a valuable and cost effective complementary approach to food surveillance and monitoring programs to assess the presence of chemical substances in the population diet and to provide reliable data in order to perform risk assessments by estimating dietary exposure.

These papers include measurements of trace elements in various foods, and some of them include measurements for lithium. We didn’t find these papers while writing our first review of the levels of lithium in food and drink because these papers aren’t looking for lithium specifically — they’re looking at all sorts of different contaminants and minerals, and lithium just happens to sometimes make the cut.

Some Total Diet Studies, like this one from the US in 1996, this one from Egypt in 1998, this one from Chile in 2005, this one from Cameroon in 2013, and this one from China in 2020, don’t measure lithium. In fact the USDA has been doing a Total Diet Study since 1961, and haven’t ever measured lithium. 

But anyways, several of these papers do include measurements of lithium in various national food supplies, and they’re strange, because unlike every other source we’ve seen, which all routinely find some foods with more than 1 mg/kg lithium, they find less than 0.5 mg/kg lithium in every single food. 

TDS with Li

The oldest TDS study we’ve seen that includes lithium is from 1999 in the United Kingdom, reporting on the UK 1994 Total Diet Study and comparing those results to data from previous UK Total Diet Studies. (The UK TDS has been “carried out on a continuous annual basis since 1966” but it seems like they only started including lithium in their analysis in the 1990s.) They report the mean concentrations of 30 elements (aluminium, antimony, arsenic, barium, bismuth, boron, cadmium, calcium, chromium, cobalt, copper, germanium, gold, iridium, iron, lead, lithium, manganese, mercury, molybdenum, nickel, palladium, platinum, rhodium, ruthenium, selenium, strontium, thallium, tin, and zinc) in 119 categories of foods, combined into 20 groups of similar foods for analysis.

The highest mean concentration of lithium they found in the food categories they examined was an average of 0.06 mg/kg (fresh weight) in fish. They estimated a total exposure of 0.016 mg lithium a day, and an upper limit of 0.029 mg a day, in the British diet at the time. This appears to be substantially less than the amount found in a 1991 sample, which gave an estimate of 0.040 mg lithium a day in the British diet. They explicitly indicate there is no data on lithium in foods (in their datasets) from before 1991.

France conducted a TDS in 2000, and a report all about it was published in 2005. They looked at levels of 18 elements (arsenic, lead, cadmium, aluminium, mercury, antimony, chrome, calcium, manganese, magnesium, nickel, copper, zinc, lithium, sodium, molybdenum, cobalt and selenium) in samples of 338 food items.

The highest mean concentration of lithium they found in the food categories they examined was an average of 0.123 mg/kg in shellfish (fresh matter) and 0.100 mg/L in drinking water. They estimated an average daily exposure of 0.028 mg for adults, with a 97.5th percentile daily exposure of 0.144 mg. They specifically mention, “drinking waters and soups are the vectors contributing most (respectively 25–41% and 14–15%) to the exposure of the populations; other vectors contribute less than 10% of the total food exposure.”

France did another TDS in 2006, with a report published in 2012. This time they looked at Li, Cr, Mn, Co, Ni, Cu, Zn, Se and Mo in 1319 samples of foods typically consumed by the French population.

Similar to the first French TDS, the highest mean concentration of lithium they found in the food categories they examined was an average of 0.066 mg/kg (fresh weight) in shellfish. But the highest individual measurements were found in two samples of sparkling water, with 0.612 mg/kg and 0.320 mg/kg.

New Zealand seems to run a Total Diet Study programme every 4–5 years since 1975, but we’ve only been able to find lithium measurements from this project in a paper from 2019, looking at data from the 2016 New Zealand Total Diet Study. Maybe, like some of the other TDS projects, they only started including lithium testing later on. Anyways, in this paper they looked at 10 elements (antimony, barium, beryllium, boron, bromine, lithium, nickel, strontium, thallium and uranium) in eight composite samples each of 132 food types.

This paper is a little strange, and unlike most of these papers, doesn’t give much detail. They summarize the main findings for lithium as, “the reported concentrations ranged from 0.0007 mg/kg in tap water to 0.54 mg/kg in mussels” and say that the mean overall intake of lithium in New Zealand adults is 0.020–0.029 mg/day.

The most recent TDS that looked at lithium seems to be this 2020 paper, which looks at food collected between October 2016 and February 2017 in the Emilia-Romagna Region in Italy. They looked at levels of fifteen trace elements (antimony, barium, beryllium, boron, cobalt, lithium, molybdenum, nickel, silver, strontium, tellurium, thallium, titanium, uranium, and vanadium) in 908 food and beverage samples from local markets, supermarkets, grocery stores, and community canteens.

The highest concentration of lithium they found in the food categories they examined was in fish and seafood (50th percentile 0.019 mg/kg, IQR 0.010–0.038 mg/kg), and legumes (50th percentile 0.015 mg/kg, IQR 0.006–0.035 mg/kg). They estimate a dietary lithium intake for the region of 0.018 mg/day (IQR 0.007–0.029 mg/day).

So overall, these papers report that lithium levels in foods and beverages never break 0.612 mg/kg, and almost universally keep below 0.1 mg/kg.

How About Those Numbers

We’re skeptical of these numbers for a couple of reasons.

For starters, these five papers disagree with basically every other measurement we’ve ever seen for lithium in food.

The TDS papers say that all foods and beverages contain less than 1 mg/kg lithium, and that people’s lithium intake is well below 1 mg a day. But this is up against sources like the following, which all find much higher levels (not an exhaustive list):

  • Bertrand (1943), “found that the green parts of lettuce contained 7.9 [mg/kg] of lithium”
  • Borovik-Romanova (1965) “reported the Li concentration in many plants from the Soviet Union to range from 0.15 to 5 [mg/kg] in dry material”, in particular listing the levels (mg/kg) in tomato, 0.4; rye, 0.17; oats, 0.55; wheat, 0.85; and rice, 9.8.
  • Hullin, Kapel, and Drinkall (1969) found more than 1 mg/kg in salt and lettuce, and up to 148 mg/kg in tobacco ash.
  • Duke (1970) found more than 1 mg/kg in some foods in the Chocó rain forest, in particular 3 mg/kg in breadfruit and 1.5 mg/kg in cacao. 
  • Sievers & Cannon (1973) found up to 1,120 mg/kg lithium in wolfberries.
  • Magalhães et al. (1990) found up to 6.6 mg/kg in watercress at the local market.
  • Ammari et al. (2011), looked at lithium in plant leaves, including spinach, lettuce, etc. and found concentrations in leaves from 2 to 27 mg/kg DM.
  • Manfred Anke and his collaborators found more than 1 mg/kg in a wide variety of foods, in multiple studies across multiple years, up to 7.3 mg/kg on average for eggs.
  • Schnauzer (2002) reviewed a number of other sources finding average intakes across several locations from 0.348 to 1.560 mg a day.
  • Five Polish sources from 1995 that a reader recently sent us reported finding (as examples) 6.2 mg/kg in chard, 18 mg/kg in dandelions, up to 470.8 mg/kg in pasture plants in the Low Beskids in Poland, up to 25.6 mg/kg in dairy cow skeletal muscle, and more than 40 mg/kg in cabbage under certain conditions. (These papers aren’t available online but we plan to review them soon.)   

It seems like either the measurements from the TDS papers are right, and all foods contain less than 1 mg/kg lithium, or all the rest of the literature is right, and many plants and foods regularly contain more than 1 mg/kg lithium. The alternative, that both of them are right, would mean that the same foods consistently contain less than 1 mg/kg in France and New Zealand while containing more than 1 mg/kg in Germany and Brazil. This seems like the most far-fetched possibility.

There are three strikes against the TDS numbers. First, they’re strictly outnumbered. When five papers from four sources (two of those papers are from France) say one thing and the rest of the literature clearly says another, it’s not a sure thing, but the side with more evidence… well it has more evidence for it.

Second, the TDS studies have a divided focus. They’re not really interested in lithium at all; they’re interested in the local food supply, and lithium just happens to be one of between 9 and 30 different elements they’re testing for. In comparison, pretty much all the other papers are looking at lithium in particular. If we had to guess which kind of team is more likely to mess up this kind of analysis, the team interested in this one particular element, or the team that randomly included the element in the list of several elements they’re testing for, we know which we’d pick. It’s hard to imagine that every team looking for lithium chose the wrong analysis or screwed it up in the same way somehow. It’s easy to imagine that the TDS studies, which measured lithium incidentally, might get some part of the analysis wrong.

It’s kind of like clothing. Ready-made sizes will fit most elements, but if you have an unusual body type (really long arms, really thick neck, etc.) you may have to go to a tailor. And lithium has the most unusual body type of all the solid elements. It wouldn’t be at all surprising if off-the-rack clothes didn’t fit poor little lithium.

uhhhhh spectral analysis

The third thing that’s strange is that there seem to be some internal contradictions within the studies. For example, in the first French TDS study, the lithium levels in water are much higher than lithium levels in things that are made out of water, which seems impossible. The mean lithium level in drinking water is 0.100 mg/kg, but the lithium levels in things that are mostly water are much lower: 0.038 mg/kg in soups, 0.006 mg/kg in coffee, 0.004 in non-alcoholic beverages, 0.003 in alcoholic beverages, and 0.002 in hot beverages. Soup is maybe a little different, but coffee and beverages are mostly water. How can there be fifty times more lithium in plain water than in hot beverages, which are (we assume) mostly water? 

For that matter, how can drinking water be the category with the second-most lithium (after shellfish)? Water is the main ingredient in beverages, but it’s also a major ingredient of pretty much every food. Fruits, salads, milk, vegetables, etc. etc. all contain lots of water. Unless there’s some major, universal filtering going on, there should be more lithium in at least some foods than there is in water. 

And that’s what you see if you look at the other elements in this first French paper — more in foods than in water. For example, the average level of manganese in drinking water in these data is 0.19 mg/kg, and the mean levels in beverages are all 0.30 mg/kg or higher; the mean level in soup is 0.97 mg/kg; the mean level in fruits is 2.05 mg/kg, much higher. Same for zinc. The mean level in drinking water is 0.05 mg/kg, which is the lowest mean level of zinc of any food category. Other elements, at least, tend to have higher concentrations in some foods than in water.

In the second French TDS study, the same thing happens. The highest concentration of lithium they found in any food was in water, 0.612 mg/kg. The mean for water this time around was only 0.035 mg/kg, but that’s still higher than the means for most beverages and the mean for almost every food. 

(The other TDS papers don’t give mean lithium measurements for water, so we can’t do the same comparison with them.)

This doesn’t make much sense. Water is a major component of many foods and it would be shocking if lithium didn’t find its way from water into food (and more obviously into beer and tea). But all of the fruits and vegetables have less lithium than the water that would presumably be used to irrigate them. 

There’s a rich literature of hydroponics experiments that shows that all sorts of plants accumulate lithium. When you grow them in a lithium solution under controlled conditions, or in soil spiked with lithium, the plants end up containing a higher concentration of lithium than the solution/soil they were grown in.

These spikes are much larger than the levels of lithium plants are normally exposed to in the environment, but they’re experimental evidence that lithium accumulates, even to enormous degrees. You should reliably expect to see more lithium in plants than in the water they’re grown with. There might be some plants that don’t accumulate, but water shouldn’t universally contain the highest amounts.

We didn’t really include these sources in our original review because that was a review of lithium in food, and these hydroponically-grown experimental plants aren’t in the actual food supply. But they’re pretty informative, so here’s a selection of the studies: 

  • Magalhães et al. (1990) grew radish, lettuce and watercress in a hydroponic system, with solution containing lithium levels of 0.7, 6.8 and 13.6 mg/L. These are all somewhat high, but exposure to 0.7 mg/L in water isn’t totally unrealistic. Plants were collected thirty days after transplanting. At the lowest and most realistic level of exposure, 0.7 mg/L, lettuce contained 11 mg/kg lithium, radish bulbs contained 11 mg/kg, radish leaves contained 17 mg/kg, and watercress contained 37 mg/kg. At 6.8 mg/L in the solution all plants contained several hundred mg/kg, and at 13.6 mg/L, radish leaves and watercress contained over 1000 mg/kg.
  • Hawrylak-Nowak, Kalinowska, and Szymańska (2012) grew corn and sunflower plants in glass jars containing 0 (control), 5, 25, or 50 mg/L lithium in a nutrient solution. After 14 days, they harvested the shoots, and found that lithium accumulated in the shoots in a dose-dependent manner. Even in the control condition, where no lithium was added to the solution, sunflower shoots contained 0.9 mg/kg and corn shoots contained 4.11 mg/kg lithium. At 5 mg/L solution, sunflower contained 422.5 mg/kg and corn contained 72.9 mg/kg; at 25 mg/L solution, sunflower contained 432.0 mg/kg and corn contained 438.0 mg/kg; at 50 mg/L solution, sunflower contained 3,292.0 mg/kg and corn contained 695.0 mg/kg. These levels are unrealistically high, but the example is still illustrative.
  • Kalinowska, Hawrylak-Nowak, and Szymańska (2013) grew lettuce hydroponically in solution containing 0, 2.5, 20, 50 or 100 mg/L lithium. Lithium concentrations above 2.5 mg/L progressively fucked the plants up more and more, but there was clear accumulation of lithium in the lettuce. There was some concentration in the leaves in a solution of 2.5 mg/L (though they don’t give the numbers), and when the lettuce was grown in a 20 mg/L solution, there was around 1000 mg/kg in the leaves.
  • Antonkiewicz et al. (2017) is an unusual paper on corn being grown hydroponically in solutions containing various amounts of lithium. They find that corn is quite resistant to lithium in its water — it actually grows better when exposed to some lithium, and only shows a decline at concentrations around 64 mg/L. (“The concentration in solution ranging from 1 to 64 [mg/L] had a stimulating effect, whereas a depression in yielding occurred only at the concentrations of 128 and 256 [mg/L].”) But the plant also concentrates lithium — even when only exposed to 1 mg/L in its solution, the plant ends up with an average of about 11 mg/kg in dry material.
  • Robinson et al. (2018) observed significant concentration in the leaves of several species as part of a controlled experiment. They planted beetroot, lettuce, black mustard, perennial ryegrass, and sunflower in controlled environments with different levels of lithium exposures. “When Li was added to soil in the pot experiment,” they report, “there was significant plant uptake … with Li concentrations in the leaves of all plant species exceeding 1000 mg/kg (dry weight) at Ca(NO3)2-extractable concentrations of just 5 mg/kg Li in soil, representing a bioaccumulation coefficient of >20.” For sunflowers in particular, “the highest Li concentrations occurred in the bottom leaves of the plant, with the shoots, roots and flowers having lower concentrations.”

Again, these are unrealistic for the amount of lithium you might find in your food, but they’re clear support for the idea that plants consistently accumulate lithium relative to the conditions they’re grown in. It doesn’t make sense that we see water having the highest concentration in the TDS data.

This is your sunflower leaf on 50 mg/L lithium

So for all these reasons, we’re pretty sure that the TDS numbers are wrong and that the lithium-specific literature is right. Specialty research that looks for lithium in particular is more reliable in our opinion than sources that happen to look at lithium as one contaminant along with a dozen others. 

But even so, you’d have to be terminally incurious to look at this and not wonder what was going on. Why do these five papers have measurements that don’t match the rest of the literature? 

What’s Going on in the TDS

Since these papers disagree with every other source, and they all share the same Total Diet Study approach, it seems like there must be something wrong with that approach. 

Sometimes this kind of mistake can come from problems with the equipment, dropping a decimal, or misreading units, like mistaking mg/kg for µg/kg.

But we have a hard time imagining that all of these different teams with (as far as we can tell?) no overlap in authors would be making exactly the same error of using the wrong units or moving a decimal place. It’s possible they all use the same slightly-misleading software or something; we have seen a few other papers that report lithium in one set of units, and every other element they test for in different units. But again, it would be weird for every single TDS study to screw this up in exactly the same way. 

So we went back and took a closer look at their methods. What we noticed is that every one of these TDS studies used the same analysis technique — inductively coupled plasma mass spectrometry, or ICP-MS. 

So we wonder if there might be an issue with ICP-MS. 

Let’s take a closer look at those TDS methods: 

The 1999 TDS paper from the United Kingdom:  

Samples of each food group … were homogenized and digested (0.5 g) in inert plastic pressure vessels with nitric acid (5 ml) using microwave heating (CEM MDS 2000 microwave digestion system). All elements except mercury, selenium and arsenic were analysed by Inductively Coupled Plasma-Mass Spectrometry (ICP-MS) (Perkin Elmer Elan 6000).

The 2005 first TDS paper from France:

The elementary analyses (about 18 000 results in all) were carried out by the Environmental Inorganic Contaminants and Mineral Unit of the AFSSA-LERQAP, which is the national reference laboratory. All the 998 individual food composite samples were homogenized and digested (about 0.6 g taken from each sample) in the quartz vessels with suprapure nitric acid (3 ml) using Multiwave closed microwave system (Anton-Paar, Courtaboeuf, France). The total content of all selected essential and non essential trace elements in the foods was determined by Inductively Coupled Plasma-Mass Spectrometry (ICP-MS) (VG PlasmaQuad ExCell-Thermo Electron, Coutaboeuf, France), a very powerful technique for quantitative multi-elemental analysis.

France again in 2012:

The National Reference Laboratory (NRL) for heavy metals was chosen to analyse 28 trace elements, and among them nine essential elements, Li, Cr, Mn, Co, Ni, Cu, Zn, Se and Mo, by inductively coupled plasma-mass spectrometry (ICP-MS) after microwave-assisted digestion. 

Sample digestion was carried out using the Multiwave 3000 microwave digestion system (Anton-Paar, Courtaboeuf, France), equipped with a rotor for 8 type X sample vessels (80-mL quartz tubes, operating pressure 80 bar). Before use, quartz vessels were decontaminated in a bath of 10% HNO3 (67% v/v), then rinsed with ultra-pure water, and dried in an oven at 40 °C. Dietary samples of 0.2–0.6 g were weighed precisely in quartz digestion vessels and wet-oxidised with 3 mL of ultra-pure water and 3 mL of ultra-pure HNO3 (67% v/v) in a microwave digestion system. One randomly-selected vessel was filled with reagents only and taken through the entire procedure as a blank. The digestion program had been optimised previously (Noël, Leblanc, & Guérin, 2003). After cooling at room temperature, sample solutions were quantitatively transferred into 50-mL polyethylene flasks. One hundred microlitres of internal standard solution (1 mg L−1) were added, to obtain a final concentration of 2 μg L−1, and then the digested samples were made up with ultrapure water to the final volume before analysis by ICP-MS.

ICP-MS measurements were performed using a VG PlasmaQuad ExCell (Thermo, Courtaboeuf, France). The sample solutions were pumped by a peristaltic pump from tubes arranged on a CETAC ASX 500 Model 510 autosampler (CETAC, Omaha, NE). 

The 2019 New Zealand TDS paper doesn’t give much detail at all. They just say: 

Samples were analysed for 10 toxic elements by ICP-MS at Hill Laboratories, Hamilton, New Zealand. 

Ok then.

Finally, the 2020 TDS paper from Italy

We measured content of fifteen trace elements (antimony, barium, beryllium, boron, cobalt, lithium, molybdenum, nickel, silver, strontium, tellurium, thallium, titanium, uranium, and vanadium) in 908 food and beverage samples through inductively coupled plasma mass spectrometry.

Using a clean stainless-steel knife, we cut solid foods by collecting samples from six different points in the plate. Then, we homogenized the samples using a food blender equipped with a stainless-steel blade and we placed a portion of 0.5 g in quartz containers previously washed with MilliQ water (MilliQPlus, Millipore, MA, USA) and HNO3. We liquid-ashed the samples with 10 ml solution (5 ml HNO3 + 5 ml·H2O) in a microwave digestion system (Discover SP-D, CEM Corporation, NC, USA) and we finally stored them in plastic tubes, and diluted to 50 ml with deionized water before analysis. Using an inductively coupled plasma mass spectrometer (Agilent 7500ce, Agilent Technologies, CA, USA), we performed trace element determination.

So, all of these papers use the same analysis technique, ICP-MS. We don’t know the exact technique used by the team in New Zealand, but all the other teams used microwave digestion with nitric acid (HNO3). Three of them (the French and Italian TDS studies) used quartz vessels.

The fact that all these studies use similar analysis techniques makes it much more plausible that something about this technique is screwing up something about the lithium detection.

This also seems likely because most other papers, the ones that find more than 1 mg/kg lithium in food, don’t use ICP-MS. Here’s a small selection.

The most recent paper finding more than 1 mg/kg lithium in plant matter seems to have used inductively coupled plasma optical emission spectrometry (ICP-OES), a related but distinct technique. This is Robinson et al. (2018), which found that plants can contain “several hundred mg/kg Li” in leaves. Here’s their procedure: 

Plant samples were washed in deionized water and dried at 60 °C until a constant weight was obtained. Subsequently, they were milled using a Cyclotech type 1093 cyclone grinder with an aluminium rotor. Plant material (0.5 g) was digested in 5 ml HNO3. The digests were diluted with Milli Q (Barnstead, EASYpure RF, 18.3 MΩ-cm) to a volume of 25 ml and filtered with a Whatman 52 filter paper (pore size 7 μm). … Pseudo-total element concentrations (henceforth referred to as “total”) were determined in the acid digests using ICP-OES (Varian 720 ES).

Ammari et al. (2011), looked at lithium in solids (plant leaves, including spinach, lettuce, etc.) and found concentrations in leaves from 2 to 27 mg/kg DM. They used this procedure: 

Collected leaves were gently washed in distilled water, air-dried, and then oven-dried to a constant weight at *70°C. Dry leaves were finely ground in a Moulinex Mill (Moulinex, Paris, France) to pass through a 40-mesh sieve. As Li is known to be present in cell vacuoles in inorganic soluble form, Li was determined in filtrates of oven-dry ground leaf samples (5 g) suspended in 50 ml of deionized water for 1 h. This procedure was used in the current study because not all the lithium present in natural unprocessed foods is taken up by the human body (pers. comm. with nutritionists; Dr. Denice Moffat, USA). Lithium extracted with deionized water represents the soluble fraction that is directly taken up by the gastrointestinal tract and considered the most bio-available. … The concentration of Li in leaf samples was measured with a flame photometer.

Anke’s 2005 paper doesn’t give a ton of detail, but seems to have used atomic absorption spectroscopy (AAS) for lithium, and reports numbers up to 7.5 mg/kg in foods. 

Magalhães et al. (1990) found up to 1,216 mg/kg in (hydroponically-grown, experimental) watercress and say: 

Thirty days after transplanting, the plants were harvested, shoots and roots separately, and their fresh weight determined. They were oven-dried at 700C for 72 hours, weighted, ground in a Wiley mill and analyzed for N, P, K, Ca, Mg, Fe and Li contents after digestion in H2SO4 and H202. N was determined by Nesslerization, P by an ammonium molybdate-amino naphthol sulfonic acid reduction method (Murphy & Riley 1962), K and Li by flame emission and Ca, Mg and Fe by atomic absorption (Sarruge & Haag 1974).

Drinkall et al. (1969), one of our oldest sources, found up to 148 mg/kg in pipe tobacco and used “the atomic absorption technique”. Specifically they say: 

Methods for determination of lithium in foodstuffs have in the past been limited almost entirely to the use of the spectrograph and the flame photometer. In the present investigation, however, it was decided to apply the technique of atomic absorption for this purpose. The chief reason for this choice was the lack of occurrence of spectral interference occasioned by elements other than lithium, Indeed, the only elements which were thought likely to prove troublesome were calcium and strontium. Even these, however, were found not to interfere. The instrument used throughout this work was the Unicam SP90 Atomic Absorption Spectrophotometer, a propane-air flame being employed.

So this diverse set of methods all found levels of lithium above 1 mg/kg, while the “ICP-MS with microwave digestion in nitric acid (usually in quartz vessels)” technique seems to reliably find way less than 1 mg/kg. This is starting to look like it’s an issue with the analysis.

If this is the case, then if we can find other papers that use ICP-MS with microwave digestion in nitric acid, they should also show low levels of lithium, well below 1 mg/kg. 

That’s exactly what we’ve found. Take a look at Saribal (2019). This paper used ICP-MS and looked at trace element concentrations in cow’s milk samples from supermarkets in Istanbul, Turkey. They found an average of 0.009 mg/L lithium in milk, way lower than the measurements for milk found in sources that don’t use ICP-MS. 

Saribal, like the TDS studies, used ICP-MS to look for lithium alongside a huge number of other elements — 19 in fact. The full list was: lithium, beryllium, chromium, manganese, cobalt, nickel, copper, arsenic, selenium, strontium, molybdenum, cadmium, antimony, barium, lead, bismuth, mercury, thallium, and uranium. Like the TDS studies, they did digestion in nitric acid: 

The quadrupole inductively coupled plasma mass spectrometer (ICP-MS) used in this work was Thermo Scientific X Series II (Thermo Fisher Scientific, Bremen, Germany).

One-milliliter portions of each milk samples were digested in 65% HNO3 and 2 mL 30% H2O2 (Merck, Poole, UK) on a heat block. The temperature was increased gradually, starting from 90 °C and increasing up to 180 °C. The mixture was cooled down and the contents were transferred to polypropyl- ene tubes with seal caps. Each digested sample was diluted to a final volume of 10 mL with double deionized water

Here’s another one. Kalisz et al. (2019) looked at “17 elements, including rare earth elements, in chilled and non-chilled cauliflower cultivars”. They used ICP-MS, they microwave digested with nitric acid, and they found lithium levels of less than 0.060 mg/kg. Here’s the method: 

We investigated the content of Ag, Al, Ba, Co, Li, Sn, Sr, Ti, Sb, and all rare-earth elements. … Curds were cut into pieces and dried at 70 °C in a dryer with forced air circulation. Then, the plant material was ground into a fine and non-fibrous powder using a Pulverisette 14 ball mill (Fritsch GmbH, Germany) with a 0.5-mm sieve. Next, 0.5 g samples were placed in to 55 ml TFM vessels and were mineralized in 10 ml 65% super pure HNO3 (Merck no. 100443.2500) in a Mars 5 Xpress (CEM, USA) microwave digestion system. The following mineralization procedure was applied: 15 min. time needed to achieve a temperature of 200 °C and 20 minutes maintaining this temperature. After cooling, the samples were quantitatively transferred to 25 ml graduated flasks with redistilled water. Contents of mentioned elements were determined using a high-dispersion inductively coupled plasma optical emission spectrometer (ICP-OES; Prodigy Teledyne Leeman Labs, USA).

There are a couple complications, but they’re worth looking at. Seidel et al. (2020) used ICP-MS and found reasonable-seeming numbers in a bunch of beverages. But, as far as we can tell, they didn’t digest the beverages at all. They just say:

Li concentrations in our 160 samples were determined via inductively coupled plasma mass spectrometry (ICP-MS) as summarized in Table 1.

Here’s Table 1 in case you’re curious: 

This seems like evidence that something about the digestion process might be to blame. 

There’s also Voica, Roba, and Iordache (2020), a Romanian paper which used ICP-MS and found up to 3.8 mg/kg in sheep’s milk and up to 4.2 mg/kg in pumpkins. This is pretty surprising — it’s the first ICP-MS paper we’ve seen that finds more than 1 mg/kg lithium in a sample of food. They even use microwave digestion with nitric acid! So at first glance, this looks like a contradiction — but when we looked closer, their method did differ in some interesting ways.

The lithium concentrations were determined by inductively coupled plasma – mass spectrometry (ICP-MS). 

Considering that samples have a very complex composition with large organic matter content, the total digestion of the matrix is mandatory to assure complete metal solubility. The studied samples were subjected to microwave assisted nitric acid digestion by using a closed iPrep vessel speed system MARS6 CEM One Touch. The digestion vessels were cleaned with 10 mL HNO3 using the microwave cleaning program and rinsed with deionized water. Approximately 0.3 g aliquots of the samples were weighed, followed by digestion in 10mL HNO3 60% at high pressure, temperature and in the presence of microwave irradiation. The vessel was closed tightly, placed on the rotor, and the digestion was carried out following the program presented in Table 1.

After complete digestion and cooling, the samples were filtered, transferred to 50 mL graduated polypropylene tubes and diluted to volume with deionized water.

A Perkin Elmer ELAN DRC-e instrument was used with a Meinhard nebulizer and a glass cyclonic spray chamber for pneumatic nebulization. The analysis was performed in the standard mode and using argon gas (purity ≥ 99.999%) for the plasma following the manufacturer’s recommendations.

The operating conditions were a nebulizer gas flow rate of 0.92 L/min; an auxiliary gas flow of 1.2 L/min; a plasma gas flow of 15 L/min; a lens voltage of 7.25 V; a radiofrequency power of 1100 W; a CeO/Ce ratio of 0.025; and a Ba++/Ba+ ratio of 0.020.

We don’t know exactly what the difference might be, but the fact that they mention that “considering that samples have a very complex composition with large organic matter content, the total digestion of the matrix is mandatory to assure complete metal solubility” suggests that they were aware of limitations of normal digestion methods that other teams may have been unaware of. And none of the other papers seem to have used pneumatic nebulization, so maybe that makes the difference and lets you squeeze all the lithium out of a pumpkin.  

yeah that’s one way to do it

Another difference we notice is that while Voica, Roba, and Iordache do use ICP-MS and the same kind of digestion as the TDS studies, they don’t test for anything else — they’re just measuring lithium. So maybe the thing that torpedoes the ICP-MS measurements is something about testing for lots of elements at the same time — a trait shared by all the TDS studies, Saribal (2019), and Kalisz et al. (2019), but not by Seidel et al. (2020) (the beverages paper) and not by Voica, Roba, and Iordache (2020).

A final (we promise) paper that helps triangulate this problem is Nabrzyski & Gajewska (2002), which looked at lithium in food samples from Gdańsk, Poland. They found an average of only 0.07 mg/kg in milk products and of only 0.11 mg/kg in smoked fish. This is not quite as low as the TDS studies but it’s much lower than everything else. And weirdly, they didn’t use ICP-MS, they used AAS. But they did digest their foods in nitric acid. Here’s the method: 

The representative samples were dry ashed in quartz crucibles and the ash was treated with suitable amounts of conc. HCl and a few drops of conc. HNO3. The obtained sample solution was then used for the determination of Sr, Li and Ca by the flame atomic absorption spectrometry (AAS) method. Ca and Li were determined using the air-acetylene flame and Sr with nitrous oxide-acetylene flame, according to the manufacturer’s recommendations.

So maybe this seems like more evidence that it’s something about the digestion process in particular, though this paper could also just be a weird outlier. It’s hard to tell without more tests.

Close Look at ICP-MS

We seem to have pretty clear evidence that ICP-MS, maybe especially in combination with microwave digestion / digestion with nitric acid, gives much lower numbers for lithium in food samples than every other analysis technique we’ve seen. 

So we wanted to know if there was any other reason to suspect that ICP-MS might give bad readings for lithium in particular. We did find a few things of interest.

If you check out the Wikipedia page for ICP-MS, lithium is mentioned as being just on the threshold of what the ICP-MS can detect. This makes sense because lithium is unusual, much smaller than all other other metals. See for example: “The ICP-MS allows determination of elements with atomic mass ranges 7 to 250 (Li to U)” and “electrostatic plates can be used in addition to the magnet to increase the speed, and this, combined with multiple collectors, can allow a scan of every element from Lithium 6 to Uranium Oxide 256 in less than a quarter of a second.” 

While ICP-MS is generally considered the gold standard for spectral analysis, like all methodologies, it has some limitations. Given that lithium is at the bottom of the range to begin with, it seems plausible to us that even small irregularities in the analysis might push it “off the end” of the range, disrupting detection. There’s more likely to be problems with lithium than with the other elements the TDS papers were analyzing.

We noticed that the 1999 UK TDS study had this to say about the upper limits of detection for ICP-MS: “The platinum group elements are notoriously difficult to analyse, as the concentrations, generally being close to the limits of detection, can be prone to some interferences in complex matrices when measured by ICP-MS.” 

Now lithium is on the low end of the range, not the high range. But since the UK TDS study authors were concerned that elements “close to the limits of detection, can be prone to some interferences in complex matrices when measured by ICP-MS”, it seems like interference might be an issue. This shows that “fall of the end of the range” is a real concern with ICP-MS analysis. So ICP-MS may be the gold standard for spectral analysis, but it falls short of being the platinum standard.

There’s also something interesting in Anke’s 2003 paper, where he says:

Lithium may be determined in foods and biological samples with the same techniques employed for sodium and potassium. However, the much lower levels of lithium compared with these other alkali metals, mean that techniques such as flame photometry often do not show adequate sensitivity. Flame (standard addition procedure) or electrothermal atomic absorption spectrophotometry are the most widely used techniques after wet or dry ashing of the sample. Corrections may have to be made for background/matrix interferences. Inductively coupled plasma atomic emission spectrometry is not very sensitive for this very low-atomic-weight element.

As usual with Anke this is very cryptic, and inductively coupled plasma atomic emission spectrometry (ICP-AES) isn’t the same technique as ICP-MS. But even so, Anke’s comment does suggest that there might be some limitations on ICP methods when measuring lithium, that they might not be very sensitive.

We also found an article by environmental testing firm WETLAB which describes several problems you can run into doing lithium analysis, including that “[w]hen Li is in a matrix with a large number of heavier elements, it tends to be pushed around and selectively excluded due to its low mass. This provides challenges when using Mass Spectrometry.” They also indicate that “ICP-MS can be an excellent option for some clients, but some of the limitations for lithium analysis are that lithium is very light and can be excluded by heavier atoms, and analysis is typically limited to <0.2% dissolved solids, which means that it is not great for brines.” We’re not looking at brines, but this may also hold true for digested food samples. WETLAB indicates their preferred methodology is ICP-OES.

Conclusion

Maybe nobody knows what’s going on here! It’s looking more and more like this is just a question that’s sitting out on the limits of human knowledge. It’s a corner case — to know why some papers find high levels and other papers find really low levels, you might have to jointly be an expert on ICP-MS, lithium analysis, and chemical analysis in food. Manfred Anke is the only guy we’ve ever heard of who seemed like he might be all three, and he’s been dead for more than ten years. So maybe there’s no one alive who knows the answer. But that’s why we do science, right?

In any case, we’re very glad to know about this complexity early on in the process of planning our own survey, since we had also been planning to use ICP-MS! We had assumed that ICP-MS was the best technique and that it would certainly give us the most accurate numbers. But measurement is rarely that simple — we should have been more careful, and now we will be.

How do we figure out what’s going on here, and what technique we should use? We could go back and pore over the literature in even more detail. But that would take a long time, and would probably be inconclusive. Much better is to simply test a bunch of foods using different techniques, pit ICP-MS against techniques like AAS and flame photometry, and see if we can figure out what’s going on. So that’s what we’re gonna do.

A Series of Unfortunate Omelettes: Lithium in Food Review & Survey Proposal

One thing that makes lithium a plausible explanation for the obesity epidemic is that clinical doses of lithium cause weight gain as a side-effect. A clinical dose of lithium is in the range of 1000 mg (“300 mg to 600 mg … 2 to 3 times a day”), and people pretty reliably gain weight on doses this high. In a 1976 review of case records, about 60% of people gained weight on clinical doses, with an average weight gain of about 10 kg.

But those are clinical doses, and it seems like the doses you’re getting from the environment are generally much smaller. There’s usually some lithium in modern drinking water, and there’s more lithium in drinking water now than there used to be. It seems to get into the water supply from things like drilled water wells, fracking, and fossil fuel prospecting, transport, and disposal. But even with all these sources of contamination, the dose you’re getting from your drinking water is relatively low, probably not much more than 0.2 mg per day. If you live right downstream from a coal plant, or you’re chugging liter bottles of mineral water on the regular, you could maybe get 5 or 10 mg/day. But no one is getting 1000 mg/day or even 300 mg/day from their drinking water. 

So what gives? 

Effects of Trace Doses

One possibility is that small amounts of lithium are enough to cause obesity, at least with daily exposure.

This is plausible for a few reasons. There’s lots of evidence (or at least, lots of papers) showing psychiatric effects at exposures of less than 1 mg (see for example meta-analysis, meta-analysis, meta-analysis, dystopian op-ed). If psychiatric effects kick in at less than 1 mg per day, then it seems possible that the weight gain effect would also kick in at less than 1 mg. 

There’s also the case study of the Pima in the 1970s. The Pima are a group of Native Americans who live in the American southwest, particularly around the Gila River Valley, and they’re notable for having high rates of obesity and diabetes much earlier than other groups. They had about 0.1 mg/L in their water by the 1970s (which was 50x the national median at the time), for a dose of only about 0.2-0.3 mg per day, and were already about 40% obese. All this makes the trace lithium hypothesis seem pretty reasonable.

Unfortunately, no one knows where the weight gain effects of lithium kick in. As far as we can tell, there’s no research on this question. It might cause weight gain at doses of 10 mg, or 1 mg, or 0.1 mg. Maybe 0.5 mg a week on average is enough to make some people really obese. We just don’t know.

Some people in the nootropics community take lithium, often in the form of lithium orotate (they use orotate rather than other compounds because it’s available over-the-counter), as part of their stacks. Based on community posts like this, this, and this, the general doses nootropics enthusiasts are taking are in the range of 1-15 mg per day. 

We haven’t done a systematic review of the subreddit (but maybe you should, that would be a good project for someone) but they seem to report no effects or mild positive effects at 1 or 2 mg lithium orotate and brain fog and fatigue at 5 mg lithium orotate and higher. Some of them report weight gain, even on doses this low. The fact that a couple extra mg might be enough to push you over the line suggests that the weight gain tipping point is somewhere under 10 mg, maybe a lot under. And for what it’s worth, all of this is consistent with the only randomized controlled trial examining the effects of trace amounts of lithium which found results at just 0.4 mg a day. 

Clinical and Subclinical Doses

Another possibility is that people really ARE getting unintended clinical doses of lithium. We see two reasons to think that this might be possible.

#1: Doses in the Mirror may be…

The first is that clinical doses are smaller than they appear. 

When a doctor prescribes you lithium, they’re always giving you a compound, usually lithium carbonate (Li2CO3). Lithium is one of the lightest elements, so by mass it will generally be a small fraction of any compound it is part of. A simple molecular-weight calculation shows us that lithium carbonate is only about 18.7% elemental lithium. So if you take 1000 mg a day of lithium carbonate, you’re only getting 187.8 mg/day of the active ingredient.

The little purple orbs are the pharmacologically active lithium ions, everything else is non-therapeutic carbonate

For bipolar and similar disorders, lithium carbonate has become such a medical standard that people usually just refer to the amount of the compound. It’s very unusual for an ion to be a medication, so this nuance is one that some doctors/nurses don’t notice. It’s pretty easy to miss. In fact, we missed it too until we saw this reddit comment from u/PatienceClarence/, which begins, “First off we need to differentiate between the doses of lithium orotate vs elemental lithium. For example, my dosage was 130 mg orotate which would give me 5 mg ‘pure’ lithium…” 

Elemental lithium is what we really care about, and when we look at numbers from the USGS or serum samples or whatever, they’re all talking about elemental lithium. When we say people get 0.1 mg/day from their water, or when we talk about getting 3 mg from your food, that’s milligrams of elemental lithium. When we say that your doctors might give you 600 mg per day, that’s milligrams lithium carbonate — and only 112.2 milligrams a day of elemental lithium. With this in mind, we see that the dose of elemental lithium is always much lower than the dose as prescribed. 

A high clinical dose is 600 mg lithium carbonate three times a day (for a total of 1800 mg lithium carbonate or about 336 mg elemental lithium), but many people get clinical doses that are much smaller than this. Low doses seem to be more like 450 mg lithium carbonate per day (about 84 mg/day elemental lithium) or even as little as 150 mg lithium carbonate per day (about 28 mg/day elemental lithium).

Once we take the fact that lithium is prescribed as a compound into account, we see that the clinical dosage is really closer to something like 300 mg/day for a high dose and 30 mg/day for a low dose. So at this point we just need to ask, is it possible that people might occasionally be getting 30 mg/day or more lithium in the course of their everyday lives? Unfortunately we think the answer is yes.

#2: Concentration in Food

The other reason to think that modern people might be getting clinical or subclinical doses on the regular is that there’s clear evidence that lithium concentrates in some foods. 

Again, consider the Pima. The researchers who tested their water in the 1970s also tested their crops. While most crops were low in lithium, they found that one crop, wolfberries, contained an incredible 1,120 mg/kg.

By our calculations, you could easily get 15 mg of lithium in a tablespoon of wolfberry jelly. If the Pima ate one tablespoon a day, they would be getting around 100 times more lithium from that tablespoon than they were getting from their drinking water.

The wolfberries in question (Lycium californium) are a close relative of goji berries (Lycium barbarum or Lycium chinense). The usual serving size of goji berries is 30 grams, which if you were eating goji berries like the ones the Pima were eating, would provide about 33.6 mg of lithium. This already puts you into clinical territory, a little more than someone taking a 150 mg tablet of lithium carbonate.

If you had a hankering and happened to eat three servings of goji berries in one day, you would get just over 100 mg of lithium from the berries alone. We don’t know how much people usually eat in one go, but it’s easy enough to buy a pound (about 450 g) of goji berries online. We don’t have any measurements of how much lithium are in the goji berries you would eat for a snack, but if they contained as much lithium as the wolfberries in the Gila River Valley, the whole 1 lb package would contain a little more than 500 mg of lithium.

So. Totally plausible that some plants concentrate 0.1 mg/L lithium in water into 1,120 mg/kg in the plant, because Sievers & Cannon have measurements of both. Totally plausible that you could get 10 or even 100 mg if you’re eating a crop like this. So now we want to know, are there other crops that concentrate lithium? And if so, what are they?

In this review, we take a look at the existing literature and try to figure out how much lithium there is in different foods. What crops does it concentrate in? Is there any evidence that foods are further contaminated in processing or transport? There isn’t actually all that much work on these questions, but we’ll take a look at what we can track down.

Let’s not bury the lede: we find evidence of subclinical levels of lithium in several different foods. But most of the sources that report these measurements are decades old, and none of them are doing anything like an exhaustive search. That’s why at the end of this piece, we’re going to talk a little bit about our next project, a survey of lithium concentrations in foods and beverages in the modern American food supply.

Because of this, our goal is not to make this post an exhaustive literature review; instead, our goal is to get a reasonable sense of how much lithium is in the food supply, and where it is. When we do our own survey of modern foods, what should we look at first? This review is a jumping off point for our upcoming empirical work.

Context for the Search

But first, a little additional context. 

There are a few official estimates of lithium consumption we should consider (since these are in food and water, all these numbers should be elemental lithium). This review paper from 2002 says that “the U.S. Environmental Protection Agency (EPA) in 1985 estimated the daily Li intake of a 70 kg adult to range from [0.650 to 3.100 mg].” The source they cite for this is “Saunders, DS: Letter: United States Environmental Protection Agency. Office of Pesticide Programs, 1985”, but we can’t find the original letter. As a result we don’t really know how accurate this estimate is, but it suggests people were getting about 1-3 mg per day in 1985.

These numbers are backed up by some German data which appear originally to be from a paper from 1991, which we will discuss more in a bit: 

In Germany, the individual lithium intake per day on the average of a week varies between [0.128 mg/day] and [1.802 mg/day] in women and [0.139] and [3.424 mg/day] in men. 

The paper also includes histograms of those distributions: 

Both of these say “mg/day” but we’re pretty sure that’s 1000x too high and they should say “µg/day”. If it were mg/day we think many of these people would be dead?

We want to call your attention to the shape of both of these distributions, because the shape is going to be important throughout this review. Both distributions are pretty clearly lognormal, meaning they peak early on but then have a super long tail off to the right. For example, most German men in this study were getting only about 0.2 to 0.4 mg of lithium per day, but twelve of them were getting more than 1 mg a day, and five of them were getting more than 2 mg a day. At least one person got more than 3 mg a day. And this paper is looking at a pretty small group of Germans. If they had taken a larger sample, we would probably see a couple people who were consuming even more. You see a similar pattern for women, just at slightly lower doses.

We expect pretty much every distribution we see around food and food exposure to be lognormal. The amount people consume per day should usually be lognormally distributed, like we see above. The distribution of lithium in any foods and crops will be lognormal. So will the distribution of lithium levels in water sources. For example, lithium levels in that big USGS dataset of groundwater samples we always talk about are distributed like this:

With scatterplot because those outliers are basically invisible on the histogram

Again we see a clear lognormal distribution. Most groundwater samples they looked at had less than 0.2 mg/L lithium. But five had more than 0.5 mg/L and two had more than 1 mg/L.

This is worth paying close attention to, because when a variable is lognormally distributed, means and medians will not be very representative. For example, in the groundwater distribution you see above, the median is .0055 mg/L and the mean is .0197 mg/L. 

These sound like really tiny amounts, and they are! But the mean and the median do not tell anywhere close to the full story. If we keep the long tail of the distribution in mind, we see that about 4% of samples contain more than 0.1 mg/L, about 1% of samples contain more than 0.2 mg/L, and of course the maximum is 1.7 mg/L. 

This means that about 4% of samples contain more than 20x the median, about 1% of samples contain more than 40x the median, and the maximum is more than 300x the median.

Put another way, about 4% of samples contain more than 5x the mean, about 1% of samples contain more than 10x the mean, and the maximum is more than 80x the mean.

We should expect similar distributions everywhere else, and we should expect means and medians to consistently be misleading in the same way. So if we find a crop with 1 mg/kg of lithium on average, that suggests that the maximum in that crop might be as high as 80 mg/kg! If this math is even remotely correct, you can see why crops that appear to have a low average level of lithium might still be worth empirically testing.

Another closely related point: that USGS paper only found those outliers because it’s a big survey, 4700 samples. Small samples will be even more misleading. Let’s imagine the USGS had taken a small number of samples instead. Here are some random sets of 6 observations from that dataset:

0.044, 0.007, 0.005, 0.036, 0.001, 0.002

0.002, 0.028, 0.005, 0.001, 0.009, 0.001

0.003, 0.006, 0.002, 0.001, 0.001, 0.006

We can see that small samples ain’t representative. If we looked at a sample of six US water sources and found that all of them contained less than 0.050 mg/L of lithium, we would miss that some US water sources out there contain more than 0.500 mg/L. In this situation, there’s no substitute for a large sample size (or, the antidote is to be a little paranoid about how long the tail is).

So if we looked at a sample of (for example) six lemons, and found that all of them contained less than 10 mg/kg of lithium, we might easily be missing that there are lemons out there that contain more than 100 mg/kg.

In any case, the obvious lognormal distribution fits really well with the kind of bolus-dose explanation we discussed with JP Callaghan, who said: 

My thought was that bolus-dosed lithium (in food or elsewhere) might serve the function of repeated overfeeding episodes, each one pushing the lipostat up some small amount, leading to overall slow weight gain. … I totally vibe with the prediction that intake would be lognormally distributed. … lognormally distributed doses of lithium with sufficient variability should create transient excursions of serum lithium into the therapeutic range.

In the discussion with JP Callaghan, we also said:

Because of the lognormal distribution, most samples of food … would have low levels of lithium — you would have to do a pretty exhaustive search to have a good chance of finding any of the spikes. So if something like this is what’s happening, it would make sense that no one has noticed. 

What we’re saying is that even if people aren’t getting that much lithium on average, if they sometimes get huge doses, that could be enough to drive their lipostat upward. If we take that model seriously, the average amount might not not be the real driver, and we should focus on whether there are huge lithium bombs out there, and how often you might encounter them. Or it could be even more complicated! Maybe some foods give you repeated moderate doses, and others give you rare megadoses. 

Two final notes before we start the review: 

First, if two sources disagree — one says strawberries are really high in lithium and the other says that strawberries are really low in lithium, or something — we should keep in mind that disagreement might mean something like “the strawberries were grown in different conditions (i.e. one batch was grown in high-lithium soil and the other batch wasn’t)” or even “apparently identical varieties of strawberries concentrate lithium differently”. There isn’t a simple answer to simple-sounding questions like “how much lithium is in a strawberry” because reality is complicated and words make it easy to hide that complexity without thinking about it.

Second, we want to remind you that whatever dose causes obesity, lithium is also a powerful sedative with well-known psychiatric effects. If you’re getting doses up near the clinical range, it’s gonna zonk you out and probably stress your kidneys. 

Ok. What crops concentrate lithium?

Lithium Concentration

Unfortunately we couldn’t find several of the important primary sources, so in a number of places, we’ve had to rely on review papers and secondary sources. We’re not going to complain “we couldn’t find the primary source” every time, but if you’re ever like “why are they citing a review paper instead of the original paper?” this is probably why.

We should warn you that these sources can be a little sloppy. Important tables are labeled unclearly. Units are often given incorrectly, like those histograms above that say mg/day when they should almost certainly say µg/day. When you double-check their citations, the numbers don’t always match up. For example, one of the review papers said that a food contained 55 mg/kg of lithium. But when we double-checked, their source for that claim said just 0.55 mg/kg in that food. So we wish we were working with all the primary sources but we just ain’t. Take all these numbers with a grain of salt.

Particularly important modern reviews include Lithium toxicity in plants: Reasons, mechanisms and remediation possibilities by Shahzad et al. (2016), Regional differences in plant levels and investigations on the phytotoxicity of lithium by Franzaring et al. (2016), and Lithium as an emerging environmental contaminant: Mobility in the soil-plant system by Robinson et al. (2018). Check those out if you finish this blog post and you want to know more.

It’s worth noting just how concerned some of these literature reviews sound. Shahzad et al. (2016) say in their abstract, “The contamination of soil by Li is becoming a serious problem, which might be a threat for crop production in the near future. … lack of considerable information about the tolerance mechanisms of plants further intensifies the situation. Therefore, future research should emphasize in finding prominent and approachable solutions to minimize the entry of Li from its sources (especially from Li batteries) into the soil and food chain.”

Older reviews include The lithium contents of some consumable items by Hullin, Kapel, and Drinkall — a 1969 paper which includes a surprisingly lengthy review of even older sources, citing papers as far back as 1917. Sadly we weren’t able to track down most of these older sources, and the ones we could track down were pretty vague. Papers from the 1930s just do not give all that much detail. Still, very cool to have anything this old. 

There’s also Shacklette, Erdman, Harms, and Papp (1978), Trace elements in plant foodstuffs, a chapter from (as far as we can tell) a volume called “Toxicity of Heavy Metals in the Environment”, which is part of a series of reference works and textbooks called “HAZARDOUS AND TOXIC SUBSTANCES”. It was sent to us by a very cool reader who refused to accept credit for tracking it down. If you want to see this one, email us.

A bunch of the best and most recent information comes from a German fella named Manfred Anke, who published a bunch of papers on lithium in food in Germany in the 1990s and 2000s. He did a ton of measurements, so you will keep seeing his name throughout. Unfortunately the papers we found from Anke mostly reference measurements from earlier work he did, which we can’t find. Sadly he is dead so we cannot ask him for more detail.

From Anke, in case anyone can track them down, we’d especially like to see a couple papers from the 1990s. Here they are exactly as he cites them:  

Anke’s numbers are very helpful, but we think they are a slight underestimation of what is in our food today. We’re pretty sure lithium levels in modern water are higher than levels in the early 1990s, and we’re pretty sure lithium levels are higher in US water than in water in Germany. In a 2005 paper, Anke says: “In Germany, the lithium content of drinking water varies between 4 and 60 µg/L (average : 10 µg/L).” Drinking water in the modern US varies between undetectable and 1700 µg/L (1.7 mg/L), and even though that 1700 is an outlier, about 8% of US groundwater samples contain more than 60 µg/L, the maximum Anke gives for Germany. The mean for US groundwater is 19.7 µg/L, compared to the 10 µg/L Anke reports.

So the smart money is that Anke’s measurements are probably all lower than the levels in modern food, certainly lower than the levels in food in the US.

Here’s another thing of interest: in one paper Anke estimates that in 1988 Germany, the average daily lithium intake for women was 0.373 mg, and the average daily lithium intake for men was 0.432 mg (or something like that; it REALLY looks like he messed up labeling these columns, luckily the numbers are all pretty similar). By 1992, he estimates that the average daily lithium intake for women was 0.713 mg, and the average daily lithium intake for men was 1.069 mg. He even explicitly comments, saying, “the lithium intake of both sexes doubled after the reunification of Germany and worldwide trade.”

That last bit about trade suggests he is maybe blaming imported foods with higher lithium levels, but it’s not really clear. He does seem to think that many foreigners get more lithium than Germans do, saying, “worldwide, a lithium intake for adults between [0.660 and 3.420 mg/day] is calculated.”

Anyways, on to actual measurements.

Beverages

Beverages are probably not giving you big doses of lithium, with a few exceptions.

Most drinking water doesn’t contain much lithium, rarely poking above 0.1 mg/L. Some beverages contain more, but not a lot more. The big exception, no surprise, is mineral water.

As usual, Anke and co have a lot to say. The Anke paper from 2003 says, “cola and beer deliver considerable amounts of lithium for humans, and this must be taken into consideration when calculating the lithium balance of humans.” The Anke paper from 2005 says that “amounts of [0.002 to 5.240 mg/L] were found in mineral water. Like tea and coffee, beer, wine and juices can also contribute to the lithium supply.” But the same paper reports a range of just 0.018 – 0.329 mg/L in “beverages”. Not clear where any of these numbers come from, or why they mention beer in particular — the citation appears to be the 1995 Anke paper we can’t find. 

In fact, Anke seems to disagree with himself. The 2005 paper mentions tea and coffee contributing to lithium exposure. But the 2003 paper says, “The total amount in tea and coffee, not their water-soluble fraction in the beverage, was registered. Their low lithium content indicates that insignificant amounts of lithium enter the diet via these beverages.”

This 2020 paper, also from Germany, finds a weak relationship for beer and wine and a strong relationship for tea with plasma concentrations for lithium. We think there are a lot of problems with this method (the serum samples are probably taken fasted, and lithium moves through the body pretty quickly) but it’s interesting.

Franzaring et al. (2016), one of those review papers, has a big figure summarizing a bunch of other sources, which has this to say about some beverages: 

For water, 1 ppm is approximately 1 mg/L

So obviously mineral water can contain a lot — if you drank enough, you could probably get a small clinical dose from mineral water alone. On the other hand, who’s drinking a liter of mineral water? Germans, apparently.

We think their sources for wine are Classification of wines according to type and region based on their composition from 1987 and Classification of German White Wines with Certified Brand of Origin by Multielement Quantitation and Pattern Recognition Techniques from 2004. The 1987 paper reports average levels of lithium in Riesling and Müller-Thurgau wines in the range of about 0.010 mg/L, and a maximum of only 0.022 mg/L. The 2004 paper looks at several German white wines, and reports a maximum of 0.150 mg/L. This is pretty unsystematic but does seem to indicate an increase. 

This paper from 2000 similarly finds averages of 0.035 and 0.019 mg/L in red wines from northern Spain. This 1994 paper and this 1997 paper both report similar values. We also found this 1988 paper looking at French red wines which suggests a range from 2.61 to 17.44 mg/L lithium. Possibly this was intended to be in µg/L instead of in mg/L? “All results are in milligrams per liter except Li, which is in micrograms per liter” is a disclaimer we’ve seen in more than one of these wine papers.

So it might be good to check, but overall we don’t think you’ll see much more than 0.150 mg/L in your wine, and most of you are hopefully drinking less than a full liter at a time.

She’s just so happy!

The most recent and most comprehensive source for beverages, however, is a 2020 paper called Lithium Content of 160 Beverages and Its Impact on Lithium Status in Drosophila melanogaster. Forget the Drosophila, let’s talk about all those beverages. This is yet another German paper, and they analyzed “160 different beverages comprising wine and beer, soft and energy drinks and tea and coffee infusions … by inductively coupled plasma mass spectrometry (ICP-MS).” And unlike other sources, they give all the numbers — If you want to know how much lithium they found in Hirschbraeu/Adlerkoenig, “Urtyp, hell” or the cola known as “Schwipp Schwapp”, you can look that up. 

They find that, aside from mineral water, most beverages in Germany contain very little lithium. Concentration in wine, beer, soft drinks, and energy drinks was all around 0.010 mg/L, and levels in tea and coffee barely ever broke 0.001 mg/L.

The big outlier is the energy drink “Acai 28 Black, energy”, which contained 0.105 mg/L. This is not a ton in the grand scheme of things — it’s less than some sources of American drinking water — but it’s a lot compared to the other beverages in this list. They mention, “it has been previously reported that Acai pulp contains substantial concentrations of other trace elements, including iron, zinc, copper and manganese. In addition to acai extract, Acai 28 black contains lemon juice concentrate, guarana and herb extracts, which possibly supply Li to this energy drink.”

BEWARE

We want to note that beverages in America may contain more lithium, just because American drinking water contains more lithium than German drinking water does. But it’s doubtful that people are getting much exposure from beverages beyond what they get from the water it’s made with. 

Basic Foods

We also have a few leads on what might be considered “basic” or “component” foods.

Anke mentions sugars a bit, though doesn’t go into much detail, saying, “honey and sugar are also extremely poor in lithium…. The addition of sugar apparently leads to a further reduction of the lithium content in bread, cake, and pastries.“ At one point he lists the range of “Sugar, honey” as being 0.199 – 0.527 mg/kg, with a mean of 0.363 mg/kg. That’s pretty low.

We also have a little data from the savory side. This paper from 1969 looked at levels in various table salts, finding (in mg/kg):

On the one hand, those are relatively high levels of lithium. On the other hand, who’s eating a kilogram of salt? Even if table salt contains 3 mg/kg, you’re just never gonna get even close to getting 1 mg from your salt.

Plant-Based Foods

It’s clear that plants can concentrate lithium, and some plants concentrate lithium more than others. It’s also clear that some plants concentrate lithium to an incredible degree. This last point is something that is emphasized by many of the reviews, with Shahzad et al. (2016) for example saying, “different plant species can absorb considerable concentration [sic] of Li.” 

Plant foods have always contained some lithium. The best estimate we have for preindustrial foods is probably this paper that looked at foods in the Chocó rain forest around 1970, and found (in dry material): 3 mg/kg in breadfruit; 1.5 mg/kg in cacao, 0.4 mg/kg in coconut, 0.25 mg/kg in taro, 0.4 mg/kg in yam, 0.6 mg/kg in cassava, 0.5 mg/kg in plantain fruits, 0.1 mg/kg in banana, 0.3 mg/kg in rice, 0.01 mg/kg in avocado, 0.5 mg/kg in dry beans, and 0.05 mg/kg in corn grains. Not nothing, but pretty low doses overall.

There are a few other old sources we can look at. Shacklette, Erdman, Harms, and Papp (1978) report a paper by Borovik-Romanova from 1965, in which she “reported the Li concentration in many plants from the Soviet Union to range from 0.15 to 5 [mg/kg] in dry material; she reported Li in food plants as follows ([mg/kg] in dry material): tomato, 0.4; rye, 0.17; oats, 0.55; wheat, 0.85; and rice, 9.8.” That’s a lot in rice, but we don’t know if that’s reliable, and we haven’t seen any other measurements of the levels in rice. We weren’t able to track the Borovik-Romanova paper down, unfortunately.

From here, we can try to narrow things down based on the better and more modern measurements we have access to.

Cereals

We haven’t seen very much about levels in cereals / grains / grass crops, but what we have seen suggests very low levels of accumulation.

Hullin, Kapel, and Drinkall (1969) mention an earlier review which found that the Gramineae (grasses) were especially “poor in lithium”, giving a range of 0.47-1.07 mg/kg. 

Borovik-Romanova reported, in mg/kg, “rye, 0.17; oats, 0.55; wheat, 0.85; and rice, 9.8” in 1965 in the USSR. Most of these concentrations are very low. Again, rice is abnormally high, but this measurement isn’t at all corroborated. And since we haven’t been able to find this primary source, there’s a good chance it should read 0.98 instead.

Anke, Arnhold, Schäfer, & Müller (2005) report levels from 0.538 to 1.391 mg/kg in “cereal products”, and in a 2003 paper, say “the different kinds of cereals grains are extremely lithium-poor as seeds.” Anke reports slightly lower levels in derived products like “bread, cake”. 

There’s also this unusual paper on corn being grown hydroponically in solutions containing various amounts of lithium. They find that corn is quite resistant to lithium in its water, actually growing better when exposed to some lithium, and only seeing a decline at concentrations around 64 mg/L. (“the concentration in solution ranging from 1 to 64 [mg/L] had a stimulating effect, whereas a depression in yielding occurred only at the concentrations of 128 and 256 [mg/L].”) But the plant also concentrates lithium — even when only exposed to 1 mg/L in its solution, the plant ends up with an average of about 11 mg/kg in dry material. Unfortunately they don’t seem to have measured how much ends up in the corn kernels, or maybe they didn’t let the corn develop that far. Seems like an oversight. (Compare also this similar paper from 2012.)

Someone should definitely double-check those numbers on rice to be safe, and corn is maybe a wildcard, but for now we’re not very worried about cereal crops.

Leafy Vegetables

A number of sources say that lithium tends to accumulate in leaves, suggesting lithium levels might be especially high in leafy foods. While most of us are in no danger of eating kilograms of cabbage, it’s worth looking out for. 

In particular, Robinson et al. (2018) observed significant concentration in the leaves of several species as part of a controlled experiment. They planted beetroot, lettuce, black mustard, perennial ryegrass, and sunflower in controlled environments with different levels of lithium exposures. “When Li was added to soil in the pot experiment,” they report, “there was significant plant uptake … with Li concentrations in the leaves of all plant species exceeding 1000 mg/kg (dry weight) at Ca(NO3)2-extractable concentrations of just 5 mg/kg Li in soil, representing a bioaccumulation coefficient of >20.” For sunflowers in particular, “the highest Li concentrations occurred in the bottom leaves of the plant, with the shoots, roots and flowers having lower concentrations.”

Obviously this is reason for concern, but these are plants grown in a lab, not grown under normal conditions. We want to check this against actual measurements in the food supply. 

Hullin, Kapel, and Drinkall (1969) report that an earlier source, Bertrand (1943), “found that the green parts of lettuce contained 7.9 [mg/kg] of lithium.” They wanted to follow up on this surprisingly high concentration, so they tested some lettuce themselves, finding: 

This pretty clearly contradicts the earlier 7.9 mg/kg, though the fact that lettuce can contain up to 2 mg/kg is still a little surprising. This could be the result of lettuce being grown in different conditions, the lognormal distribution, etc., but even so it’s reassuring to see that not all lettuce in 1969 contained several mg per kg.

In this study from 1990, the researchers went and purchased radish, lettuce and watercress at the market in Brazil, and found relatively high levels in all of them:

Let’s also look at this modern table that reviews a couple more recent sources, from Shahzad et al.:

FW = Fresh Weight and DM = Dry Matter, we think? 

None of these are astronomical, but it’s definitely surprising that spinach contains more than 4 mg/kg and celery and chard both contain more than 6 mg/kg, at least in these measurements.

So not to sound too contrarian but, maybe too many leafy greens are bad for your health. 

Fruits & Non-Leafy Veggies

Anke, Arnhold, Schäfer, & Müller (2005) say that “fruits and vegetables supply 1.0 to 7.0 mg Li/kg,” and report levels from 0.383 to 6.707 mg/kg in fruits. 

This is a wide range, and a pretty high ceiling. But as usual, Anke is much vaguer than we might hope. He gives some weird hints, but no specific measurements. In the 2003 paper, Anke says, “as a rule, fruits contain less lithium than vegetative parts of plants (vegetables). Lemons and apples contained significantly more lithium, with about 1.4 mg/kg dry matter, than peas and beans.”

More specific numbers have been hard to come by. We’ve found a pretty random assortment, like how Shahzad et al. report that “in a hydroponic experiment, Li concentration in nutrient solution to 12 [mg/L], increased cucumber fruit yield, fruit sugar, and ascorbic acid levels, but Li did not accumulate in the fruit (Rusin, 1979).” It’s interesting that cucumbers survive just fine in water containing up to 12 mg/L, and that suggests that lithium shouldn’t accumulate in cucumbers under any realistic water levels. But cucumbers are not a huge portion of the food supply.

What we do see all the time is sources commenting on how citrus plants are very sensitive to lithium. Anke says, “citrus trees are the most susceptible to injury by an excess of lithium, which is reported to be toxic at a concentration of 140–220 p.p.m. in the leaves.” Robinson et al. (2018) say, “citing numerous sources, Gough et al. (1979) reported a wide variation in plant tolerance to Li; citrus was found to be particularly sensitive, whilst cotton was more tolerant.” Shahzad et al. say, “Bradford (1963) found reduced and stunted growth of citrus in southern California, U.S.A., with the use of highly Li-contaminated water for irrigation. …  Threshold concentrations of Li in plants are highly variable, and moderate to severe toxic effects at 4–40 mg Li kg−1 was observed in citrus leaves (Kabata-Pendias and Pendias, 1992).” This Australian Water Quality Guidelines for Fresh and Marine Waters document says, “except for citrus trees, most crops can tolerate up to 5 mg/L in nutrient solution (NAS/NAE 1973). Citrus trees begin to show slight toxicity at concentrations of 0.06–0.1 mg/L in water (Bradford 1963). Lithium concentrations of 0.1–0.25 mg/L in irrigation water produced severe toxicity symptoms in grapefruit … (Hilgeman et al. 1970)”.

All tantalizing, but we can’t get access to any of those primary sources. For all we know this is a myth that’s been passed around the agricultural research departments since the 1960s.

The citrus is tantalizing, get it? 

Even if citrus trees really are extra-sensitive to lithium, it’s not clear what that means for their fruits. Maybe it means that citrus fruits are super-low in lithium, since the tree just dies if it’s exposed to even a small amount. Or maybe it means that citrus fruits are super-high in lithium — maybe citrus trees absorb lithium really quickly and that’s why lithium kills them at relatively low levels.

So it’s interesting but at this point, the jury is out on citrus.

Nightshades

Multiple sources mention that the Solanaceae family, better known as nightshades, are serious concentrators of lithium. Hullin, Kapel, and Drinkall mention that even in the 1950s, plant scientists were aware that nightshades are often high in lithium. Anke, Schäfer, & Arnhold (2003) mention, “Solanaceae are known to have the highest tolerance to lithium. Some members of this family accumulate more than 1000 p.p.m. lithium.” Shacklette, Erdman, Harms, and Papp (1978) even mention a “stimulating effect of Li as a fertilizer for certain species, especially those in the Solanaceae family.”

Shahzad et al. (2016) say, “Schrauzer (2002) and Kabata-Pendias and Mukherjee (2007) noted that plants of Asteraceae and Solanaceae families showed tolerance against Li toxicity and exhibited normal plant growth,” and, “some plants of the Solanaceae family, when grown in an acidic climatic zone accumulate more than 1000 mg/kg Li.” We weren’t able to track down most of their sources for these claims, but we did find Schrauzer (2002). He mentions that Cirsium arvense (creeping thistle) and Solanum dulcamara (called things like fellenwort, felonwood, poisonberry, poisonflower, scarlet berry, and snakeberry; probably no one is eating these!) are notorious concentrators of lithium, and he repeats the claim that some Solanaceae accumulate more than 1000 mg/kg lithium, but it’s not clear what his source for this was.

Hullin, Kapel, and Drinkall mention in particular one source from 1952 that found a range of 1.8-7.96 [mg/kg] in members of the Solanaceae. 7.9 mg/kg in some nightshades is enough to be concerned, but they don’t say which species this measurement comes from. 

The finger seems to be pointing squarely at the Solanaceae — but which Solanaceae? This family is huge. If you know anything about plants, you probably know that potatoes and tomatoes are both nightshades, but you may not know that nightshades also include eggplants, the Capsicum (including e.g. chili peppers and bell peppers), tomatillos, some gooseberries, the goji berry, and even tobacco. 

We’ve already seen how wolfberries / goji berries can accumulate crazy amounts under the right circumstances, which does make this Solanaceae thing seem even more plausible. 

Anke, Schäfer, & Arnhold (2003) mention potatoes in particular in one section on vegetable foods, saying: “All vegetables and potatoes contain > 1.0 mg lithium kg−1 dry matter.” There isn’t much detail, but the paper does say, “peeling potatoes decreases their lithium content, as potato peel stores more lithium than the inner part of the potato that is commonly eaten.”

That same paper that tries to link diet to serum lithium levels does claim to find that a diet higher in potatoes leads to more serum lithium, but we still think this paper is not very good. If you look at table 4, you see that there’s not actually a clear association between potatoes and serum levels. Table 5 says that potatoes come out in a regression model, but it’s a bit of an odd model and they don’t give enough detail for us to really evaluate it. And again, these serum concentrations were taken fasted, so they didn’t measure the right thing.

It’s much better to just measure the lithium in potatoes directly. Anke seems to have done this in the 1990s, but he’s not giving any details. We’ll have to go back all the way to 1969, when Hullin, Kapel, and Drinkall included three varieties of potatoes in their study (numbers in mg/kg):

These potatoes, at least, are pretty low in lithium. The authors do specifically say these were peeled potatoes, which may be important in the light of Anke’s comment about the peels. These numbers are pretty old, and modern potatoes probably are exposed to more lithium. But even so, these potatoes do not seem to be mega-concentrators, and Hullin, Kapel, and Drinkall did find some serious concentrators even back in 1969. 

This is especially interesting to us because it provides a little support for the idea that the potato diet might cause weight loss by reducing your lithium intake and forcing out the lithium already in your system with a high dose of potassium, or something. At the very least, it looks like you’d get less lithium in your diet if you lived on only potatoes than if you somehow survived on only lettuce (DO NOT TRY THE LETTUCE DIET).

Apparently the nightshade family’s tendency to accumulate lithium does not include the potatoes (unless the peeling made a huge difference?). This suggests that the high levels might have come from some OTHER nightshade. Obviously we have already seen huge concentrations in the goji berry (or at least, a close relative). But what about other nightshades, like tomatoes, eggplant, or bell peppers? 

Hullin, Kapel, and Drinkall do frustratingly say, “[The lithium content] of the tomato will be reported elsewhere.” But they don’t discuss it beyond that, at least not in this paper. We’ll have to look to other sources.

Shacklette et al. report: “Borovik-Romanova reported the Li concentration in [dry material] … tomato, 0.4 [mg/kg].” This is not much, though these numbers are from 1965, and from the USSR.

A stark contrast can be found in one of Anke’s papers, where they state, “Fruits and vegetables supply 1.0 to 7.0 mg Li/kg food DM. Tomatoes are especially rich in Li (7.0 mg Li/kg DM).” 

This is a lot for a vegetable fruit! It occurs to us that tomatoes are pretty easy to grow hydroponically, and you could just dose distilled water with a known amount of lithium. If any of you are hydroponic gardeners and want to try this experimentally, let us know! 

But tomatoes are obviously beaten out by wolfberries/goji berries, and they also can’t compare to this dark horse nightshade: tobacco.

SURPRISE

That’s right — Hullin, Kapel, and Drinkall (1969) also measured lithium levels in tobacco. They seem to have done this not because it’s another nightshade, but because previous research from the 1940s and 1950s had found that lithium concentrations in tobacco were “extraordinarily high”. For their own part, Hullin and co. found (mg/kg in ash): 

This is a really interesting finding, and in a crop we didn’t expect people to examine, since tobacco isn’t food.

At the same time, measuring ash is kind of cheating. Everything organic will be burned away in the cigarette or pipe, so the level of any salt or mineral will appear higher than it was in the original substance. As a result, we don’t really know the concentration in the raw tobacco. This is also the lithium that’s left over in the remnants of tobacco after it’s been smoked, so these measurements are really the amount that was left unconsumed, which makes it difficult to know how much might have been inhaled. Even so, the authors think that “the inhalation of ash during smoking could provide a further source of this metal”. 

This is also interesting in combination with the fact that people with psychiatric disorders often seem to self-medicate with tobacco. Traditionally schizophrenics are the ones drawn to being heavy smokers, but smoking is disproportionately common in bipolar patients as well. Researchers have generally tried to explain this in terms of nicotine, which we think of as being the active ingredient in tobacco, but given these lithium levels, maybe psychiatric patients smoke so much because they’re self-medicating with the lithium? Or maybe lithium exposure through the lungs causes schizophrenia and bipolar disorder? (For comparison, see Scott Alexander discussing a similar idea.)  

We didn’t find measurements for any other nightshades, but we hope to learn more in our own survey.

Animal-Based Foods

Pretty much everything we see suggests that animal products contain more lithium on average than plant-based foods. This makes a lot of general sense because of biomagnification. It also makes particular sense because many food animals consume huge quantities of plant stalks and leaves, and as we’ve just seen, stalks and leaves tend to accumulate more lithium than other parts of the plants.

toxic waste make bear sad

But the bad news is that, like pretty much everything else, levels in animal products are poorly-documented and we have to rely heavily on Manfred Anke again. He’s a good guy, we just wish — well we wish we had access to his older papers.

It’s like he’s toying with us!!!

Meat

Meat seems to contain a consistently high level of lithium. Apparently based on measurements he took in the 1990s, Anke calculates that meat products contain an average of about 3.2 mg/kg, and he gives a range of 2.4 to 3.8 mg/kg. 

In Anke, Arnhold, Schäfer, & Müller (2005) he elaborates just a little, saying, “Poultry, beef, pork and mutton contain lithium concentrations increasing in that order.”

In place of more detailed measurements, Anke, Schäfer, & Arnhold (2003) give us this somewhat difficult paragraph: 

On average, eggs, meat, sausage, and fish deliver significantly more lithium per kg of dry matter than most cereal foodstuffs. Eggs, liver, and kidneys of cattle had a mean lithium content of 5 mg/kg. Beef and mutton contain more lithium than poultry meat. Green fodder and silage consumed by cattle and sheep are much richer in lithium than the cereals largely fed to poultry. Sausage and fish contain similar amounts of lithium to meat. 

Beyond this, we haven’t found much detail to report. And even Anke can’t keep himself from mentioning how meat plays second fiddle to something else:

… Poultry, beef, pork and mutton contain lithium concentrations increasing in that order. Most lithium is delivered to humans by eggs and milk (> 7000 µg/kg DM). 

This is backed up by Hullin, Kapel, and Drinkall (1969), who said: 

Among foods of animal origin, those which have been found to contain lithium include eggs (Press, 1941) and milk (Wright & Papish, 1929; Drea, 1934).

So let’s leave meat behind for now and look at the real heavy-hitters.

Dairy

The earliest report we could find for milk was this 1929 Science publication mentioned by Hullin, Kapel, and Drinkall. But papers this old are pretty terse. It’s only about three-quarters of a page, and the only information they give about lithium is that it is included in the “elements not previously identified but now found to be present” in milk. 

Anke can do one better, and estimates an average for “Milk, dairy products” of 3.6 mg/kg with a range of 1.1 to 7.5 mg/kg. This suggests that the concentration in dairy products is pretty high across the board, but also that there’s considerable variation.

Anke explains this in a couple ways. First of all, he says that there were, “significant differences between the lithium content of milk”, and he suggests that milk sometimes contained 10 mg/kg in dry matter. This seems to contradict the range he gives above, but whatever. 

He also points out that other dairy products contain less lithium. For example, he says that butter is “lithium-poor”, containing only about 1.2 mg/kg dry matter, which seems to be the bottom of the range for dairy. “In contrast to milk,” he says, “curd cheese and other cheeses only retain 20–55% of lithium in the original material available for human nutrition. The main fraction of lithium certainly leaves cheese and curd cheese via the whey.”

This is encouraging because we love cheese and we are glad to know it is not responsible for poisoning our brains — at least, not primarily. It’s also interesting because 20-55% is a pretty big range; we’d love to know if some cheeses concentrate more than others, or if this is just an indication of the wide variance he mentioned earlier in milk. Not that we really need it, but if you have access to the strategic cheese reserve, we’d love to test historical samples to see if lithium levels have been increasing. 

What he suggests about whey is also pretty intriguing. Whey is the main byproduct of turning milk into cheese, so if cheese is lower in lithium than milk is, then whey must be higher. Does this mean whey protein is super high in lithium?

Whey protein display in The Hague, flanked by boars

Eggs

The oldest paper we could find on lithium in eggs is a Nature publication from 1941 called “Spectrochemical Analysis of Eggs”, and it is half a page of exactly that and nothing else. They do mention lithium in the eggs, but unfortunately the level of detail they give is just: “Potassium and lithium were also present [in the eggs] in fair quantity.”

Anke gives his estimate as always, but this time, it’s a little different: 

Anke gives an average (we think; he doesn’t label this column anywhere) of 7.3 mg/kg in eggs. This is a lot, more than any other food category he considers. And instead of giving a range, like he does for every other food category, he gives the standard deviation, which is 6.5 mg/kg.

This is some crazy variation. Does that mean some eggs in his sample contained more than 13.8 mg/kg lithium? That’s only one standard deviation above the average, two standard deviations would be 20.3 mg/kg. A large egg is about 50 g, so at two standard deviations above average, you could be getting 1 mg per egg. 

That does seem to be what he’s suggesting. But if we assume the distribution of lithium in eggs is normal, we get negative values quickly, and an egg can’t contain a negative amount of lithium.

Because lithium concentrations can’t be negative, and because of the distributions we’ve seen in all the previous examples, we assume the distribution of lithium in eggs must be lognormal instead.

A lognormal distribution with parameters [1.7, .76] has a mean and sd of very close to 7.3 and 6.5, so this is a reasonable guess about the underlying distribution of eggs in Germany in 1991.

Examination of the lognormal distribution with these parameters suggests that the distribution of lithium in eggs (at least in Germany in 1991) looks something like this: The modal egg in this distribution contains about 3 mg/kg lithium. But about 21% of the eggs in this distribution contain more than 10 mg/kg lithium. About 4% contain more than 20 mg/kg. About 1% contain more than 30 mg/kg. About 0.4% contain more than 40 mg/kg. And two out of every thousand contain 50 mg/kg lithium or more. 

That’s a lot of lithium for just one egg. What about the lithium in a three-egg omelette? 

ACHTUNG

To answer this Omelettenproblem, we started by taking samples of three eggs from a lognormal distribution with parameters [1.7, .76]. That gives us the concentration in mg/kg for each egg in the omelette.

Again, a large egg is about 50 grams. In reality a large egg is slightly more, but we’ll use 50 g because some restaurants might use medium eggs, and because it’s a nice round number. 

So we multiply each egg’s mg/kg value by .05 (because 50 g out of 1000 g for a kilogram) to get the lithium it contains in mg, and we add the lithium from all three eggs in that sample together for the total amount in the omelette.

We did this 100,000 times, ending up with a sample of 100,000 hypothetical omelettes, and the estimated lithium dose in each. Here’s the distribution of lithium in these three-egg omelettes in mg as a histogram: 

And here it is as a scatterplot in the style of The Economist

As you can see, most omelettes contained less than 3 mg lithium. In fact, most contained between 0.4 and 1.6 mg.

This doesn’t sound like a lot, but we think it’s pretty crazy. A small clinical dose is something like 30 mg, and it’s nuts to see that you can get easily like 1/10 that dose from a single omelette. Remember that in 1985, the EPA estimated that the daily lithium intake of a 70 kg US adult ranged from 0.650 to 3.1 mg — but by 1991 Germany, you can get that whole dose in a single sitting, from a single dish! 

Even Anke estimated that his German participants were getting no more than 3 mg a day from their food. But this model suggests that you can show up at a cafe and say “Kellner, bringen Sie mir bitte ein Omelette” and easily get that 3 mg estimate blown out of the water before lunchtime.

Even this ignores the long tail of the data. The omelettes start to peter out at around 5 mg, but the highest dose we see in this set of 100,000 hypothetical breakfasts was 11.1 mg of lithium in a single omelette.

The population of Germany in 1990 was just under 80 million people. Let’s say that only 1 out of every 100 people orders a three-egg omelette on a given day. This means that every day in early 1990s Germany, about 800,000 people were rolling the dice on an omelette. Let’s further assume that the distribution of omelettes we generated above is correct. If all these things are true, around 8 unlucky people every day in 1990s Germany were getting smacked with 1/3 a clinical dose of lithium out of nowhere. It’s hard to imagine they wouldn’t feel that. 

Processed Food

One thing we didn’t see much of in this literature review was measurements of the lithium in processed food.

We’re very interested in seeing if processing increases lithium. But no one seems to have measured the lithium in a hamburger, let alone a twinkie. 

There are a few interesting things worth mentioning, however — all from Anke, Schäfer, & Arnhold (2003), of course.

Mostly Anke and co find that processed foods are not extreme outliers. “Ready-to-serve soups with meat and eggs were [rich] in lithium,” they say, “whereas various puddings, macaroni, and vermicelli usually contained < 1 mg lithium/kg dry matter. Bread, cake, and pastries are usually poor sources of lithium. On average, they contained less lithium than wheat flour. The addition of sugar apparently leads to a further reduction of the lithium content in bread, cake, and pastries.”

Even in tasty treats, they don’t find much. We don’t know how processed German chocolate was at the time, but they say, “the lithium content of chocolates, chocolate candies, and sweets amounted to about 0.5 mg/kg dry matter. Cocoa is somewhat richer in lithium. The addition of sugar in chocolates reduces their lithium content.”

The only thing that maybe jumps out as evidence of contamination from processing is what they say about mustard. “Owing to the small amounts used in their application,” they begin, “spices do not contribute much lithium to the diet. It is surprising that mustard is relatively lithium-rich, with 3.4 mg/kg dry matter, whereas mustard seed contains extremely little lithium.” Mustard is generally a mixture of mustard seed, water, vinegar, and not much else. We saw in the section on beverages that wine doesn’t contain much lithium, so vinegar probably doesn’t either. Maybe the lithium exposure comes from processing?

Misc

We notice that for many categories of food, we seem to have simply no information. How much lithium is in tree nuts? Peanuts? Melons? Onions? Various kinds of legumes? How much is in major crops like soy? This is part of why we need to do our own survey, to fill these gaps and run a more systematic search.

It’s interesting, though not surprising, to see such a clear divide between plant and animal foods. In fact, we wonder if this can explain why vegetarian diets seem to lead to a little weight loss and vegan diets seem to lead to a little more, and also why neither of them work great.

Meat seems to contain a lot of lithium, but honestly not that much more than things like tomatoes and goji berries. Vegetarians will consume less lithium when they stop eating meat, but if they compensate for not eating meat by eating more fruit, they might actually be worse off. If they compensate by eating more eggs, or picking up whey protein, they’re definitely worse off! 

Vegans have it a little better — just by being vegan, they’ll be cutting out the three most reliable sources of lithium in the general diet. As long as they don’t increase their consumption of goji berries to compensate, their total exposure should go down. Hey, it makes more sense than “not eating dairy products gives you psychic powers because otherwise 90% of your brain is filled with curds and whey.”

But even so, a vegan can get as much lithium as a meat-eater if they consume tons of nightshades, so even a vegan diet is not a sure ticket to lithium removal. Not to mention that we have basically no information on plant-based protein sources (legumes, nuts) so we don’t know how much lithium vegans might get from that part of their diet.

In Conclusion

There’s certainly lithium in our food, sometimes quite a bit of lithium. It seems like most people get at least 1 mg a day from their food, and on many days, there’s a good chance you’ll get more.

That said, most of the studies we’ve looked at are pretty old, and none of them are very systematic. Sources often disagree; sample sizes are small; many common foods haven’t been tested at all. The overall quality is not great. We don’t think any of this data is good enough to draw strong conclusions from. Personally we’re avoiding whey protein and goji berries for right now, but it’s hard to get a sense of what might be a good idea beyond that. So as the next step in this project, we’re gonna do our own survey of the food supply.

The basic plan is pretty simple. We’re going to go out and collect a bunch of foods and beverages from American grocery stores. As best as we can, we will try to get a broad and representative sample of the sorts of foods most people eat on a regular basis, but we’ll also pay extra-close attention to foods that we suspect might contain a lot of lithium. Samples will be artificially digested (if necessary) and their lithium concentration will be measured by ICP-MS. All results will be shared here on the blog.

Luckily, we have already secured funding for the first round of samples, so the survey will proceed apace. If you want to offer additional support, please feel free to contact us — with more funding, we could do a bigger survey and maybe even do it faster. We could also get a greenhouse and run some hydroponic studies maybe.

If you’re interested in getting involved in other ways, here are a few things that would be really helpful:

1. If you would be willing to go out and buy an egg or whatever and mail it in to be tested, so we could get measurements from all over the country / the world, please fill out this form.

2. If you work at the FDA or a major food testing lab or Hood Milk or something, or if you’re a grad student with access to the equipment to test your breakfast for lithium and an inclination to pitch in, contact phil@whylome.org to discuss how you might be able to contribute to this project.

Philosophical Transactions: Lithium in Scottish Drinking Water with Al Hatfield

Previous Philosophical Transactions:

Al Hatfield is a wannabe rationalist (his words) from the UK who sent us some data about water sources in Scotland. We had an interesting exchange with him about these data and, with Al’s permission, wanted to share it with all of you! Here it is:


Hi,

I know you’re not that keen on correlations and I actually stopped working on this a few months ago when you mentioned that in the last A Chemical Hunger post, but after reading your post today I wanted to share it anyway, just in case it does help you at all. 

It’s a while since I read all of A Chemical Hunger but I think this data about Scottish water may support a few things you said:

– The amount of Lithium in Scottish water is in the top 4 correlations I found with obesity (out of about 40 substances measured in the water)

– I recall you predicted the top correlation would be about 0.5, the data I have implies it’s 0.55, so about right.

– I recall you said more than one substance in the water may contribute to obesity, my data suggested 4 substances/factors had correlations of more than 0.46 with obesity levels and 6 were more than 0.41.

Method

– Scottish Water test and record how much of up to 43 substances is in each reservoir/water source in Scotland https://www.scottishwater.co.uk/your-home/your-water/water-quality/water-quality

– their data is in pdf format but I converted it to Excel

– Scottish Water don’t publish Lithium levels online but I did a Freedom of Information request and they emailed it to me and I added it to the spreadsheet.

– I used the website to get the water quality data for a reservoir for every city/big town in Scotland and lined it up in the spreadsheet.

– I used Scottish Health Survey – Local Area Level data to find out what percentage of people are obese in each area of Scotland and then matched it as well as I could to a reservoir/water source.

– I then used the Data Analytics add-on in Excel to work out the correlations between the substances in the water and obesity.

Correlations with obesity (also in attachment)

Conductivity 0.55

Chloride 0.52

Boron 0.47

Lithium 0.47

Total Trihalomethanes 0.42

Sodium 0.42

Sulphate 0.38

Fluoride 0.37

Colony Counts After 3 Days At 22øc 0.34

Antimony 0.33

Gross Beta Activity 0.33

Total organic carbon 0.31

Gross Alpha Activity 0.30

Cyanide 0.26

Iron 0.26

Residual Disinfectant – Free 0.23

Arsenic 0.23

Pesticides – Total Substances 0.23

Coliform Bacteria (Total coliforms) 0.23

Copper 0.19

PAH – Sum Of 4 Substances 0.19

Nitrite 0.17

Colony Counts After 48 Hours At 37øc 0.16

Nickel 0.13

Nitrite/Nitrat e formula 0.13

Nitrate 0.12

Cadmium 0.11

Turbidity 0.08

Bromate 0.08

Colour 0.06

Lead -0.10

Manganese -0.12

Hydrogen ion (pH) -0.12

Aluminium -0.15

Chromium -0.15

Ammonium (total) -0.22

2_4-Db -0.25

Residual Disinfectant – Total -0.36

2_4-D -0.42

Dicamba -0.42

MCPB -0.42

MCPP(Mecoprop) -0.42

Scottish Water definition of Conductivity

Conductivity is proportional to the dissolved solids content of the water and is often used as an indication of the presence of dissolved minerals, such as calcium, magnesium and sodium.

Anyway, not sure if that’s any help to you at all but I enjoy your blog and thought I would send it in. Let me know if you have any questions.

Thanks 

Al


Hi Al,

Wow, thanks for this! We’ll take a look and do a little more analysis if that’s all right, and get back to you shortly. 

Do you know the units for the different measurements here, especially for the lithium? We’d be interested in seeing the original PDFs as well if that’s not too much hassle.

Thanks! 

SMTM


Hi,

You’re welcome! That’s great if you can analyse it as I am very much an amateur. 

The units for the Lithium measurements are µgLi/l. I’ve attached the Lithium levels Scottish Water sent me. I think they cover every water source they test in Scotland (though my analysis only covered about 15 water sources).

Sorry I don’t have access to the original pdfs as they’re on my other computer and I’m away at the moment. But I have downloaded a couple of pdfs online. Unfortunately the online versions have been updated since I did my analysis in late November, but hopefully you can get the idea from them and see what measurements Scottish Water use.

Let me know if you’d like anything else.

Thanks,

Al


Hey Al,

So we’ve taken a closer look at the data and while everything is encouraging, we don’t feel that we’re able to draw any strong conclusions.

We also get a correlation of 0.47 between obesity and lithium levels in the water. The problem is, this relationship isn’t significant, p = 0.078. Basically this means that the data are consistent with a correlation anywhere between -0.06 and 0.79, and since that includes zero (no relationship), we say that it’s not significant.

This still looks relatively good for the hypothesis — most of the confidence interval is positive, and these data are in theory consistent with a correlation as high as 0.79. But on the whole it’s weak evidence, and doesn’t meet the accepted standards.

The main reason this isn’t significant is that there are only 15 towns in the dataset. As far as sample sizes go, this is very small. That’s just not much information to work with, which is why the correlation isn’t significant. For similar reasons, we haven’t done any more complicated analyses, because we won’t be able to find much with such a small sample to work with. 

Another problem is that correlation is designed to work with bivariate normal distributions — two variables, both of them approximately normally distributed, like so: 

Usually this doesn’t matter a ton. Even if you’re looking at a correlation where the two variables aren’t really normally distributed, it’s usually ok. And sometimes you can use transformations to make the data more normal before doing your analysis. But in this case, the distribution doesn’t look like a bivariate normal at all:  

Only four towns in the dataset have seriously elevated lithium levels, and those are the four fattest towns in the dataset. So this is definitely consistent with the hypothesis.

But the distribution is very strange and very extreme. In our opinion, you can’t really interpret a correlation you get from data that looks like this, because while you can calculate a correlation coefficient, correlation was never intended to describe data that are distributed like this.

On the other hand, we asked a friend about this and he said that he thinks a correlation is fine as long as the residuals are normal (we won’t get into that here), and they pretty much are normal, so maybe a correlation is fine in this case? 

A possible way around this problem is nonparametric correlation tests, which don’t assume a bivariate normal distribution in the first place. Theoretically these should be kosher to use in this scenario because none of their assumptions are violated, though we admit we don’t use nonparametric methods very often. 

Anyways, both of the nonparametric correlation tests we tried were statistically significant — Kendall rank correlation was significant (tau = 0.53, p = .015), and so was the Spearman rank correlation (rho = 0.64, p = .011). Per these tests, obesity and lithium levels are positively correlated in this dataset. The friend we talked to said that in his opinion, nonparametric tests are the more conservative option, so the fact that these are significant does seem suggestive. 

We’re still hesitant to draw any strong conclusions here. Even if the correlations are significant, we’re working with only 15 observations. The lithium levels only go up to 7 ppb in these data, which is still pretty low, at least compared to lithium levels in many other areas. So overall, our conclusion is that this is certainly in line with the lithium hypothesis, but not terribly strong evidence either way.

A larger dataset of more than 15 towns would give us a bit more flexibility in terms of analysis. But we’re not sure it would be worth your time to put it together. It would be interesting if the correlation were still significant with 30 or 40 towns, and we could account for some of the other variables like Boron and Chloride. But, as we’ve mentioned before, in this case there are several reasons that a correlation might appear to be much smaller than it actually is. And in general, we think it can sometimes be misleading to use correlation outside the limited set of problems it was designed for (for example, in homeostatic systems).

That said, if you do decide to expand the dataset to more towns, we’d be happy to do more analysis. And above all else, thank you for sharing this with us!

SMTM

[Addendum: In case anyone is interested in the distribution in the full lithium dataset, here’s a quick plot of lithium levels by Scottish Unitary Authority: 

]


Thanks so much for looking at it. Sounds like I need to brush up on my statistics! Depending how bored I get I may extend it to 40 towns some time, but for now I’ll stick with experimenting with a water filter.

All the best,

Al

Philosophical Transactions: JP Callaghan on Lithium Pharmacokinetics

In the beginning, scientific articles were just letters. Scholars wrote to each other about whatever they were working on, celebrating their discoveries or arguing over minutiae, and ended up with great stacks of the things. People started bringing interesting letters to meetings of the Royal Society to read aloud, then scientists started addressing their letters to the Royal Society directly, and eventually Henry Oldenburg started pulling some of these letters together and printing them as the Philosophical Transactions of the Royal Society, the first scientific journal.

In continuance of this hallowed tradition, in this blog post we are publishing some philosophical transactions of our own: correspondence with JP Callaghan, an MD/PhD student at a large Northeast research university going into anesthesia. He has expertise in protein statistical mechanics and kinetic modeling, so he reached out to us with several ideas and enlightened criticisms.

With JP Callaghan’s help we have lightly edited the correspondence for clarity, turning the multi-threaded format of the email exchange into something more linear. We found the conversation very informative, and we hope you do as well! So without further ado: 


JP Callaghan:  Hi guys, great work on A Chemical Hunger

I’m sure someone already suggested this but the Fulbright program executes the “move abroad” experiment every year. In fact, they do the reverse experiment as well, paying foreigners to move to the US. The Phillipines Fulbright program seems especially active.

(The Peace Corps is already doing this experiment as well, but that’s probably probably more confounded since people are often living in pretty rustic locations.)

You could pretty easily imagine paying these folks a little extra money to send you their weight once a month or whatever.

SLIME MOLD TIME MOLD:  Thank you! Yeah, we’ve been trying to figure out the best way to pursue this one, using existing data if possible. Fulbright is a good idea, especially US <–––> Philippines, and especially because we suspect young people will show weight changes faster. We’ve also thought about trying to collect a sample of expats, possibly on reddit, since there are a lot of anecdotes of weight loss in those communities.

The tricky thing is finding someone who has an in with one of these groups. We probably can’t just cold call Fulbright and ask how much all their scholars weigh, though we’ll start asking around. 

JPC: Unfortunately my connection with the Fulbright was brief, superficial, and many years ago. I can ask around at my university, though. I’m not filled with unmitigated optimism, but the worst they can do is say no/ignore me.

Also, I wanted to mention that lithium level measurements are extremely common measurements in clinical practice. It’s used to monitor therapeutic lithium (for e.g. bipolar folks). (Although I will concede usually they are measuring .5 – 1.5 mmol/L which would be way higher than serum levels due to contamination.) Also, it’s interesting that the early pharmacokinetic studies also measured urine lithium (see e.g. Barbara Ehrlich’s seminal 1980 paper) so there’s precedent for that as well. I’m led to understand from my lab medicine colleagues that it’s a relatively straightforward (aka cheap) electrochemical assay, at least in common clinical practice.

SMTM:  We’ve looked into measurement a bit. We’re concerned that serum levels aren’t worth measuring, since lithium seems to accumulate in the brain and we suspect that would be the mechanism (a commenter suggested it might also be accumulation in bone). But if we were to do clinical measurements, we’d probably measure lithium in urine or maybe even in saliva, since there’s evidence they’re good proxies for one another and for the levels in serum, and they’re easier to collect. Urine might be especially important if lithium clearance rate ends up being a piece of the puzzle, which it seems like it might. 

JPC: It is definitely true that lithium accumulates inside cells (definitely rat neurons and human RBCs, probably human neurons, but maybe not human muscle; see e.g. that Ehrlich paper I mentioned). The thing is, lithium kinetics seem to be pretty fast. Since it’s an ion, it doesn’t partition into fat the way other long-lasting medications and toxins do, and so it’s eliminated fairly quickly by the kidneys. (THC is a classic example of a hydrophobic “contaminant”; this same physical chemistry explains why a long-time pothead will test positive for THC for months, but you can stop using cocaine and, 72 hours later, screen negative.)

It might be worth your time to look at some of the lithium washout experiments that have been done over the years (e.g. Hunter, 1988 where they see lithium levels rapidly decline after stopping lithium therapy that had been going on for a month).

I suppose, though, that I’m not aware of any data that specifically excludes the possibility that there is a very slow “third compartment” where lithium can deposit (such as, as your commenter suggested, bone; although I don’t know much about whether or not lithium can incorporate into the hydroxyapatite matrix in bone. It’s mostly calcium phosphate and I’m not sure if lithium could “find a place” in that crystalline matrix).

Anyway, though, my understanding is that lithium kinetics in the brain are relatively fast. (For instance, see Ebadi, et al where they measure [Li] in rat brains over time.) So even if you have a highly accumulated slow bone compartment, the levels of lithium you’d get in the brain would still be super low, because it equilibrates with the blood quickly and therefore is subject to rapid elimination by the kidneys.

However, I don’t think you need to posit accumulation for your hypothesis. If you’re exposed to constant, low levels of lithium, you reach an equilibrium. There’s some super low serum concentration, some rather-higher intracellular concentration, and it’s all held in steady state by the constant intake via the GI tract (say, in the water) and constant elimination by the kidneys. Perhaps this is what you’re getting at when you say the rate of elimination might be very important?

Instead, consider some interesting pharmacodynamics: low-level (or maybe widely fluctuating, since lithium is also quickly cleared?) exposure to lithium messes with the lipostat. This process is probably really slow, maybe because weight change is slow or maybe because of some kind of brain adaptation process or whatever. We have good reason to suspect low-level lithium has neurological effects already anyway through some of the population-level suicide data I’m sure you’re aware of.

Urine and serum levels of lithium are only good proxies for one another at steady state. I really strongly suggest you guys look at that Ehrlich paper. She measures serum, intra-RBC, and urine [Li] after a dose of lithium carbonate (the most common delayed-release preparation of pharmaceutical lithium).

Another good one is Gaillot et al which demonstrates how important the form of lithium (lithium carbonate vs LiCl) is to the kinetics. (As an aside, this might be a reason for lithium grease to be so bad; lithium grease is apparently some kind of weird soap complex with fatty acids, maybe it gets trapped in the GI tract or something.)

SMTM: The rat studies are interesting but don’t rats seem like a bad comparison for determining something like rate of clearance? Besides just not being human, their metabolisms are something like 6-8x faster than ours and their lifespans are about 20 times shorter. Also human brains are huge. What do you think?

JPC: Certainly I agree that rats are not people and are bad models in many ways. I think that renal function is the key parameter you’d want to compare. The most basic measure of kidney function is the GFR (glomerular filtration rate), which basically measures how much fluid gets pushed through the “kidney filter” per unit time. Unfortunately in people we measure it in volume/time/body surface area and in rats volume/time/mass which makes a comparison less obvious than I was hoping. To be honest, I am not sure how well rat kidney function and human kidney function is comparable. (Definitely more comparable than live and dead human kidney function, though 😉.)

What do you mean by ”their metabolisms are something like 6-8x faster than ours”? Like, calories/mass/time? Usually when I think about “metabolic rate” I am thinking of energy usage. When we think about drug elimination, the main things that matter are 1) liver function (for drugs that are hepatically metabolized) 2) various tissue enzyme function (e.g. plasma esterases for something like esmolol) and 3) renal function. I don’t generally think about basal metabolic rate as being a pertinent factor, really, except perhaps in cases where it’s a proxy for hepatic metabolism.

Lithium is eliminated (“cleared”) almost exclusively by the kidney and it undergoes no metabolic transformations, so I wouldn’t worry about anything but kidney function for its clearance.

You’re right, though, the 20x lifespan difference could be an issue. If we are worried about accumulation on the timescale of years, then obviously a shorter rat life is a problem. But (if I read your blog posts right) rats as experimental animals are also getting fatter so presumably the effect extends to them on the timescale of their life? (Did you have data in rats? I don’t remember.)

Indeed, if it’s actually just that there a constant low-level “infusion” of lithium via tapwater, grease exposure at work, etc giving rise to a low steady-state lithium (rather than actual bioaccumulation) this would explain why the effect does extend to these short-lived experimental animals.

SMTM: You make good points about laboratory animals. There are data on rats and they do seem to be getting heavier. Let’s stick a pin in this one for a now, you may find this next bit is relevant to the same questions:

In your opinion, are the studies you cite consistent or inconsistent with the findings of Amdisen et al. 1974 and Shoepfer et al. 2021? Also potentially relevant is Amidsen 1977. We describe their findings near the end of this section — basically they seem to suggest that Li accumulates preferentially in the bones, thyroid, and parts of the brain. The total sample size is small but it seems suggestive. We agree accumulation may not be essential to the theory but doesn’t this look like evidence of accumulation? We’ve attached copies of Amdisen et al. 1974 and Amdisen 1977 as PDFs in case you want to take a closer look. [SMTM’s Note: If anyone else wants to see these papers, you can email us.]

Especially interesting that Ebadi et al. say, “it has been shown that sodium intake exerts a significant influence on the renal elimination of lithium (Schou, 1958b)”, somewhat in line with our speculation here. We’ll have to look into that. 

Brains

JPC: Thanks for the papers. As you predicted, I’m finding them super interesting.

Shoepfer et al, 2021 is a lovely, very interesting paper (complete with some adorable Deutsch-English). I was aware of it but had not taken the time to read it yet.

By my read, it is primarily seeking to establish this new, nuclear fission based approach to measuring lithium in pathology tissue. After spending some time with it, I don’t really know how to interpret their findings. The main reason I am not sure what to do with this paper is that the results are in dead peoples’ brains. Indeed, they specifically note in their ‘limitations’ section: “The lithium distribution patterns so far obtained with the NIK method, thus in no way contradicting given literature references, are based on post mortem tissue.” The reason this is pertinent is that there is a lot of active transport of other monovalent cations (K, Na) and so I would worry that this is true for lithium as well and (obviously) this is almost certainly disrupted in dead people.

The second thing is that the tissue was fixed in (presumably) formalin and stained with hematoxylin and eosin before measuring lithium, which then comes out in units of mass/mass. Obviously in living tissue there’s lots of water and whatnot, and the mass-density of water and formalin is going to be pretty different.

So, as the authors say, I would say it’s neither consistent nor inconsistent with other data.

SMTM: It’s true that all the brain samples we have in humans are in dead brain tissue, but this seems like an insurmountable issue, right? Looking at dead tissue is the only way to get even a rough estimate of how much lithium is in the brain, since as far as we know there’s no way to test the levels in a living human brain, or if there is, no one has taken those measurements and it’s outside our current budget. 

In any case, the most relevant findings from these studies, at least in our opinion, are 1) that lithium definitely reaches brain tissue and sticks around for a while, and 2) regardless of absolute levels, there seems to be relatively more lithium in parts of the brain that regulate appetite and weight gain. These conclusions seem likely to hold even given all the reasonable concerns about dead tissue. What do you think?  

JPC: I agree. In my mind, the main question is whether or not lithium persists in the brain after cessation of lithium therapy. Put more rigorously, what is the rate of exchange between the “brain compartment” and (probably) the “serum compartment.” (I guess it could also be eliminated by CSF too maybe? Or “glymphatics”? idk I guess nobody really understands the brain.)

The main issue I have is this: if you’re exposed, say, to 20 ppb lithium and your serum has 20 ppb lithium and so does the cytoplasm in your neurons, this is actually the null hypothesis (that lithium is an inert substance that just flows down its concentration gradient). It’s obviously false (we know lithium concentrates in RBCs of healthy subjects, for instance), but this paper doesn’t help me decide if lithium 1) passively diffuses throughout the body 2) is actively concentrated in neurons, or even 3) is actively cleared from cells, simply because I don’t really know what to do with the number.

The second issue is the preparation. Maybe formalin fixation washes lithium away, or when it fixes cell membranes maybe the lithium is allowed to diffuse out. Maybe it poorly penetrates myelin sheaths, and has a tendency to concentrate the lithium inside cells by making the extracellular environment more hydrophobic (nature abhors an unsolvated ion).

Another reason I am so skeptical of the “slow lithium kinetics” hypothesis is just the physical chemistry of lithium. It’s a tiny, charged particle. Keeping these sorts of ions from moving around and distributing evenly is actually really hard in most cases. There are a few cases of ionic solids in the human body (various types of kidney stones, bones, bile stones] but for the most part these involve much less soluble ions than lithium and everything is dissolved and flows around at its whim except where it’s actively pumped.

SMTM: This is a good point, and in addition, the fact that tourists and expats seem to lose weight quickly does seem to be a point in favor of fast lithium over slow lithium. If those anecdotes bear out in some kind of more systematic study, “slow lithium kinetics” starts looking really unlikely. Another possibility, though, is that young people are the only ones who lose weight quickly on foreign trips, and there’s something like a “weight gain in the brain, reservoir in the bone” system where people remain dosed for a long time once enough has built up in their bones (or some other reservoir).

JPC: Very possible. Also young people generally have better renal function. There are tons of people walking around with their kidneys at like 50% or worse who don’t even know it.

A third and distant issue what I mentioned about the active transport of Na and K that happens in neurons (IIRC something like 1/3 of your calories are spent doing this) ceasing when you’re dead. This is also a fairly big deal, though, since there are various cation leak channels in cell membranes (for electrical excitability reasons, I think; ask an electrical engineer or a different kind of biophysicist) through which Li might also escape. (Since, after all, a reasonable hypothesis for the mechanism of action is that Li uses Na channels.)

Between these three difficulties, I do actually see this as borderline insurmountable for ascertaining how much lithium is in an alive brain based on these data. Basically, it comes down to “I don’t know how much lithium I should expect there to be in these experiments.”

However, “relatively more lithium in parts of the brain that regulate appetite and weight gain” is a good point. I think that this is something you actually can reasonably say: it seems like there is more lithium in these areas than other areas. The within-experiment comparisons definitely seem more sound. It would also be consistent with the onset of hunger/appetite symptoms below traditionally-accepted therapeutic ranges.

I do also want to clarify what I mean by “no accumulation.” There is of course a sort of accumulation for all things at all times. You take a dose of some enteral medication, it leaches into your bloodstream from your gut, accumulating first in the serum. It then is distributed throughout the body and accumulates in other compartments (brain, liver, kidney, bone, whatever). Assuming linear pharmacokinetics, there’s some rate that the drug goes in to and out of each of these compartments. 

If you keep taking the drug and the influx rate (from the serum into a compartment) is higher than the efflux rate (back to the serum from the compartment), the steady state in the compartment will be higher than the serum at steady state. In some sense, this could be called “accumulation.” But in another sense, if both these rates are fast, your accumulation is transient and quickly relaxes to zero if you clear the serum compartment of drug (which we know happens in normal individuals in the case of lithium). Although the concentration in the third compartment is indeed higher than in the serum, if you stop taking the drug, it will wash out (first from the serum then, more slowly, from the accumulating compartment).

SMTM: Thanks, this clarification is helpful. To make sure we understand, “accumulation” to you means that a contaminant goes to a part of the body, stays there, and basically never leaves. But you’re open to “a sort of accumulation” where 50 units go into the brain every day and only 10 units are cleared, leading to a more-or-less perpetual increase in the levels. Is that right? 

JPC: Yes. I would frame this in terms of rates, though. So 5 x brain concentration units go to the brain and 1 x brain concentration units go out of the brain per unit time, such that you get a steady state concentration difference between the serum in the brain of in_rate / out_rate (in this case).

You guys seem mathy so I’ll add: for an arbitrary number of compartments this is just a first-order ODE. You can represent this situation as rate matrix K where element i, j represents the rate (1/time) that material flows from compartment i to j (or maybe j to i, I can never remember). Anyway this usually just boils down to something looking like an eigenvector problem to get the stationary distribution of things. (Obviously things get more complicated when you have pulsatile influx.)

The key question, though, is what effect does this high concentration in the accumulating compartment have on the actual physiology? If we have slowly-resolving, high concentration in the brain, then I think we could call this clinical (ie neuropharmacologically significant) accumulation. However, I think the case in the brain is that you have higher-than-serum concentrations, but that these concentrations quickly resolve after cessation of lithium therapy. My reasoning for this is that lithium pharmacokinetics are classically well-modeled with two- and three-compartment models, which mostly have pretty fast kinetics (rate parameters with half lives in the hours range).

SMTM: This is interesting because our sense is sort of the opposite! Specifically, our understanding is that most people who go off clinical doses of lithium do not lose much weight and tend to keep most of the weight they gained as a side effect (correct us if we’re wrong, we haven’t seen great documentation of this). 

This seems at least suggestive that relatively high levels of lithium persist in the brain for a long time. On the other hand, clinical doses are really, really huge compared to trace doses, so maybe there is just so much in the brain compartment that it sometimes takes decades to clear. Ok we may not actually disagree, but it seemed like an interesting minor point of departure that might be worth considering.

JPC: I don’t know about this! I agree that slower (months to years) kinetics of lithium in the brain could explain this. An alternative (relatively parsimonious) explanation would be that, as Guyenet proposes, there simply is no mechanism for shedding excess adiposity. So if you gain weight as the result of any circumstance, if it stays on long enough for the lipostat to habituate to it, you just have a new, higher adiposity setpoint and have great difficulty eliminating that weight. That is, not being able to get the weight off after lithium-related weight gain might just be normal physiology.

The idea that clinical doses are just huge is sort of interesting. Normally, we think of the movement of ions in these kinetics models as having first-order kinetics (i.e. flux is proportional to concentration), but if you have truly shitboats of lithium in the brain, you could imagine that efflux might saturate (i.e. there are only so many transporters for the lithium to get out, since I imagine the cell membrane itself is impenetrable to Li+). This could be interesting. Not sure how you’d investigate it though. Probably patch-clamp type studies in ex vivo neurons? These are unfortunately expensive and extremely technical.

Amidsen

JPC: I see Amdisen et al. 1974 describes a fatal dose of lithium, which is very different pharmacokinetically from therapeutic doses. Above about 2.0 mmol/L (~2x therapeutic levels), lithium kinetics become nonlinear—that is, the pharmacokinetics are no longer fixed and the drug begins to influence its own clearance. In the case of lithium, high doses of lithium reduce clearance, leading to a vicious cycle of toxicity. This is a big deal clinically, often leading to the need for emergent hemodialysis.

So this is consistent with the papers I mentioned earlier (Ehrlich et al, Galliot et al) in the sense that cannot really conflict because they are reporting on two very different pharmacokinetic regimes.

You can’t directly compare the lithium kinetics in this patient to those in healthy people. You can see in figure 1 that the patient’s “urea” (I assume what we’d call BUN today?) explodes, which is a result of renal failure. It sounds like the patient wasn’t making any urine, i.e. has zero lithium clearance.

Figure 1 from Amdisen et al. 1974

SMTM: True, it’s hard to tell. But FWIW lithium also seems to be cleared through other sources like sweat, so even renal failure doesn’t mean zero lithium clearance, just severely reduced. (Though not sure the percent. 50% through urine? 80%? 99%?)

JPC: Yes this is true, of course. My intuition would be that it’s closer to 99% or even like 99.9%. The kidney’s “function” (I guess you have to be a bit careful not to anthropomorphize/be teleological about the kidney here, but you know what I mean) is to eliminate stuff from the blood via urine, which it does very well, whereas sweat and other excreta have other functions.

Let’s assume for a second that lithium and sodium are the same and that the body doesn’t distinguish (obviously false; all models are wrong but some are useful) and let’s do some math.

In the ICU we routinely track “ins and outs” very carefully. Generally normal urine output is 0.5 – 1.5 mL/kg body weight/hr. In a 70 kg adult call it >800 mL/day. But because we also know how much fluid is going in, we know how much we lose to evaporation (sweat, spitting, coughing up gunk, etc), which we call “insensible losses.” This is usually 40-800 mL/day.

A normal sweat chloride (which we use to check for cystic fibrosis) is <29 mM. Because sweat doesn’t have a static charge, we know there’s some positive counterion. Let’s assume it’s all sodium. So call it 30 mM NaCl, and calculate 800 mL x 30 mM = 24 mmol NaCl and 40 mL x 30 mM = 1.2 mmol. These are collected using (I think) topical pilocarpine to stimulate sweat production, so this would be an upper bound probably. It’s pretty close to what they find here which is in athletes during training (full disclosure I didn’t read the whole thing), which seems like it would be similar to the pilocarpine case (i.e. unlikely to be sustained throughout the day).

We also measure 24-hour sodium elimination when investigating disorders of the kidney. A first-reasonabe-google-hit normal range is 40-220 mmol Na/24 hours. (Of course, this is usually done when fluid-restricting the patient, so this would be on the low end of normal. If you go to Shake Shack and eat a giant salty burger your urine urea and Na are going to skyrocket. If you’re in a desert, your urine will be WAY concentrated, but maybe lower volume. It’s hard to generalize so this is at best a Fermi estimation type of deal.)

Anyhow, we’re looking at somewhere between 2x and 250x more sodium eliminated in the urine. Again my guess is that we’d be closer to the 250x number and not the 2x number for some of the reasons I mention above. Also I worry you can’t just multiply insensible losses * sweat [Na] because as water evaporates it gets drawn out of the body as free water to re-hydrate the Na, or something.

In writing this up, I also found this paper which also does some interesting quantification of sweat electrolytes (again we get a mean sweat [Na] of 37 and [Cl] of 34), but in some of the later plots (Figure 2) we can see that [Na] and [Cl] go way low and that the average seems to be being pulled up by a long tail of high sweat electrolytes.

So not sure what to take away from that but I thought I’d share my work anyway. 🙂

Bone

JPC: In the case of bone, however, there might be something here! You could imagine the bone being a large but slowly-exchanging depot of lithium. I’d be interested to see if anyone has measured bone lithium levels in folks who were, say, on chronic therapeutic lithium. I’m not aware of anything like that.

SMTM: It seems to fit Amdisen et al. 1974. That case study is of a woman who was on clinical levels of lithium for three years, and had relatively high concentrations in her bones. Like you say, a fatal dose of lithium is very different pharmacokinetically from therapeutic doses, but the rate at which lithium deposits in bone is presumably (?) much slower than for other tissues, so this may be a reasonable estimate of how much had made it into her bones from three years of clinical treatment. Sample size of one, etc., but like you say there doesn’t seem to be any other data on lithium in bones. 

JPC: I think it’s hard to say for sure if high concentration in her bones is due to the chronic therapy or the overdose. However, they note higher (0.77 vs 0.59 mmol/kg) in dense bone (iliac crest) than in spongey bone (vertebral body; there’s a better name than spongey… maybe cumulus? I don’t remember.). That’s interesting because it suggests to me (assuming that the error in the measurement is << 0.77-0.59) there is more concentrating effect in mineralized bone than all the cellular components (osteoclasts, osteoblasts, hematopoietic cells etc). 

Anyway it’s suggestive that maybe there is deposition in bone. I wouldn’t hang my hat on it, but it is definitely consistent with it. I also agree that bone mineralization/incorporation seems like it ought to be on a longer timescale than cellular transport, so that is consistent as well. Obviously n=1, etc etc, but it’s kind of cute.

SMTM: Maybe we should see if we could do a study, there must be someone out there with a… skeleton bank? What do you call that? 

JPC: A cadaver lab? I think most medical schools have them (ours does). In an academic medical setting, I would just get an IRB to collect bone samples from all the cadavers or maybe everyone who gets an autopsy that’s sufficiently extensive to make it easy to collect some bone. This would be a convenience sample, of course, but it would be interesting. Correlate age, zip code, renal function if known?

Because the patient is dead, there’s no risk of harm, and because they’re already doing the autopsy/dissection/whatever it should be relatively straightforward to collect in most cases (I mean, they remove organs and stuff to weigh and examine them so grabbing a bit of bone is easy). Unfortunately all these people got sick and died so you have a little bit of a problem there. For example, if someone had cancer and was cachectic, what can you learn from that? Idk.

In vivo bone biopsies are also a relatively common procedure done by interventional radiology under CT guidance (it’s SUPER COOL). You also have the problem that people are getting their biopsies for a reason, and usually the reason boils down to “we think that this bone looks weird,” so your samples would be almost by definition abnormal.

SMTM: Great! Maybe we can find someone with a cadaver lab and see if we can make it happen. This is a very cool idea.

Control Systems

SMTM: Earlier you mentioned the idea that the body’s set point can only be raised, but it seems really unlikely to us that there’s no mechanism for shedding excess adiposity. 

JPC: Hmm. You guys are definitely better read on this subject than I am, but do I fear I have oversimplified the Guyenet hypothesis somewhat. My recollection is that it is more that there’s no driving force for the lipostat setpoint to return to a healthy level if it has habituated to a higher level of adiposity.

I like the analogy to iron. (I don’t think that Guyenet makes this connection, but I read The Hungry Brain years ago so I’m not sure.) It turns out that the body has no way of directly eliminating iron, so when iron levels get high, the body just turns off the “get more iron” system. Eventually, iron slowly makes its way out of the body because bleeding, entropy, etc etc and the iron-absorption system clicks back on. (This is relevant because patients who receive frequent transfusions, such as those with sickle cell, get iron overload due to their inability to eliminate the extra iron.)

I guess, by analogy, it would be that the mechanism for shedding adiposity would be “turn off the big hunger cues.” It’s not no mechanism, it’s just a crappy, passive, poorly-optimized mechanism. (Presumably because, like how nobody got transfusions prior to the 20th century, there was never an unending excess of trivially-accessible and highly palatable food in our evolutionary history.)

SMTM: Well, overfeeding studies raise people’s weights temporarily but they quickly go back to where they were before. Anecdotally, a lot of people who visit lean countries lose decent amounts of weight in just a few weeks. And occasionally people drop a couple hundred pounds for no apparent reason (if the contamination hypothesis is correct, this probably happens in rare cases where a person serendipitously eliminates most of their contamination load all at once). And people do have outlets like fidgeting that seem to be a mechanism beyond just “turn off the big hunger cues.” All this seems to suggest that weight is controlled in both directions.

JPC: Proponents of the above hypothesis would explain this by saying that the lipostat doesn’t have time to habituate to the new setpoint during the timescale of an overfeeding study, and so they lose the weight by having their “acute hunger cues” turned off. Whereas as weight creeps up year after year, the lipostat slowly follows the weight up. You do bring up a good point about fidgeting, though.

My thought was that bolus-dosed lithium (in food or elsewhere) might serve the function of repeated overfeeding episodes, each one pushing the lipostat up some small amount, leading to overall slow weight gain. 

I think combining the idea that the brain concentrates lithium with an “up only” lipostat might give you this effect? If we say 1) lithium probably concentrates first in areas controlling hunger and thirst, leading to an effect on this at lower-than-theraputic serum concentrations, you might see weeks of weight-gain effect from a bolus 2) that we know that weight gain can occur on this timescale and then not revert (see the observation, which I read about in Guyenet, that most weight is gained between thanksgiving and NYE). What do you think?

SMTM: To get a little more into the weeds on this (because you may find it interesting), William Powers says in some of his writing (can’t recall where) that control systems built using neurons will have separate systems for “push up” and “push down” control. If he’s right, then there are separate “up lipostats” and “down lipostats”, and presumably they function or fail largely separately. This suggests that a contaminant that breaks one probably doesn’t break the other, and also suggests that the obesity epidemic would probably be the result of two or more contaminants.

JPC: Yes! Super interesting. There are lots of places in the brain where this kind of push-pull system is used. I remember very clearly a neuroscience professor saying, while aggressively waving his hands, that “engineers love this kind of thing and that’s probably why the brain does it too.” I wonder if he was thinking of Powers’ work when he said that.

SMTM: Let’s say that contaminant A raises the set point of the “down lipostat”, and contaminant B raises the set point of the “up lipostat”. Someone exposed to just A doesn’t necessarily get fatter, but they can drift up to the new set point if they overeat. At the same time, with exercise and calorie restriction, there’s nothing keeping them from pushing their weight down again. 

Someone exposed to both A and B does necessarily get fatter, because they are being pushed up, and they have to fight the up lipostat to lose any weight, which is close to impossible. (This might explain why calorie restriction seems to work as a diet for some people but doesn’t work generally.) 

Someone exposed to just B, or who has a paradoxical reaction to A, sees their up and down lipostats get in a fight, which looks like cycles of binging and purging and intense stress. This might possibly present as bulimia.

There isn’t enough evidence to tell to this level of detail, but a plausible read based on this theoretical perspective is that we might see something like, lithium raises the set point of the down lipostat and PFAS raise the set point of the up lipostat, and you only get really obese if you get exposed to high doses of both. 

JPC: Very interesting! It’s definitely appealing on a theoretical level. (See: your recent post on beauty in science.) I just don’t know anything about the state of the evidence in the systems neuroscience of obesity to say if it’s consistent or inconsistent with the data. (Same is of course true of the lipostat-creep hypothesis above.)

I’m not sure about why you think the two systems would function separately? Certainly, for us to see a change, there would have to be a failure of one or the other population preferentially but I’m not sure why this would be less common than one effect or the other. They’d be likely anatomical neighbors, and perhaps even developmentally related. I guess it would all depend on the actual physiology. I’m thinking, for instance, of how the eye creates center-surround receptive fields using the same photoreceptors in combination with some (I think) inhibitory interneurons (neural NOT gates). The same photoreceptor, hooked up a different way, acts to activate or inhibit different retinal ganglion cells (the cells that make up the optic nerve… I think. It’s been a while.). Another example might be the basal ganglia, which (allegedly) functions to select between different actions, but mostly our drugs act to “do more actions” by being pro-dopaminergic (for instance to treat Parkinsons) or “do fewer actions” by being antidopaminergic (as in antipsychotics like haloperidol).

SMTM: Yeah good points and good question! We have reasons to believe that these systems (and other paired systems) do function more or less separately, but it might be too long to get into here. Long story short we think they are computationally separate but probably share a lot of underlying hardware. 

Dynamics

SMTM: What do you think of a model based on peak lithium exposure? Our concern is that most sources of exposure are going to be lognormally distributed. Most of the time you get small doses, but very rarely you get a really really large dose. Most food contains no lithium grease, but every so often some grease gets on your hamburger during transport and you eat a big glob of it by accident. 

Lognormal Distribution

Or even more concerning: you live downriver from a coal power plant, and you get your drinking water from the river. Most of the time the river contains only 10-20 ppb Li+, nothing all that impressive. But every few months they dump a new load of coal ash in the ash pond, which leaches lithium into the river, and for the next couple of days you’re drinking 10,000 ppb of lithium in every glass. This leads to a huge influx, and your compartments are filled with lithium. 

This will deplete over time as your drinking water goes back to 10 ppb, but if it happens frequently enough, influx will be net greater than efflux over the long term and the general lithium levels in your compartments will go up and up. But anyone who comes to town to test your drinking water or your serum will find that levels in both are pretty low, unless they happen to show up on one of the very rare peak exposure days. So unless you did exhaustive testing or happened to be there on the right day, everything would look normal.

JPC: I totally vibe with the prediction that intake would be lognormally distributed. From a classic pharmacokinetic perspective, I would expect lognormally-distributed lithium boluses to actually be buffered by the fact that renal clearance eliminates lithium in proportion to its serum concentration–that is, it gets faster as lithium concentrations go up.

But I’m a big believer that you should shut up and calculate so I coded up a three compartment model (gut -> serum <-> tissue), made up some parameters* that seemed reasonable and gave the qualitative behavior I expected). Then either gave the model either 300 mg lithium carbonate three times a day (a low-ish dose of the the preparation given clinically), or three-times-a-day doses drawn from a lognormal distribution with two parameter sets (µ=1.5 and σ=1.5 or σ=2.5; this corresponds to a median dose of about 4.4 mg lithium carbonate in both cases, since the long tail doesn’t influence the median very much).

* k_gut->serum = 0.01 per minute

* k_serum->brain = 0.01 per minute

* k_brain->serum = 0.0025 per minute

* k_serum->urine = 0.001 per minute

* V_d,serum = 16 L

In my opinion, this gives us the following hypothesis: lognormally distributed doses of lithium with sufficient variability should create transient excursions of serum lithium into the therapeutic range.

Because this model includes that slow third compartment, we can also ask what the amount of lithium in that compartment is:

My interpretation of this is that the third compartment smooths the very spiky nature of the serum levels and, in that third compartment, you get nearly therapeutic levels of lithium in the third compartment for whole weeks (days ~35-40) after these spikes, especially if you get two spikes back to back. (Which it seems to me would be likely if you have, like, a coal ash spill or it’s wolfberry season or whatever.)

There clearly are a ton of limitations here: the parameters are made up by me, real kinetics are more like two slow compartments (this has one), lithium carbonate is a delayed preparation that almost certainly has different kinetics from food-based lithium, and I have no idea how realistic my lognormal parameters are, to name a few. However, I think the general principle holds: the slow compartment “smooths” the spikes, and so doing seems to be able to sustain highish [Li] even when the kidney is clearing it by feasting when Li is plentiful and retaining it during famine periods.

I’m not sure if this supports your hypothesis or not (do you need sustained brain [Li] above some threshold to get weight gain? I don’t think anyone knows…) but I thought the kinetics were interesting and best discussed with actual numbers and pictures than words. What do you guys think? Is this what you expected?

SMTM: Yes! Obviously the specifics of the dynamics matter a lot, but this seems to be a pretty clear demonstration of what we expected — that it’s theoretically possible to get therapeutic levels in the second compartment (serum) and sometimes in the third compartment (brain?), even if the median dose is much much lower than a therapeutic dose. 

And because of the lognormal distribution, most samples of food or serum would have low levels of lithium — you would have to do a pretty exhaustive search to have a good chance of finding any of the spikes. So if something like this is what’s happening, it would make sense that no one has noticed. 

It would be interesting to make a version of this model that also includes low-level constant exposure from drinking water (closer to 0.1 mg per day) and looks at dynamics over multiple years, getting an impression of what lifetime accumulation might look like, but that sounds like a project for another time.

Thyroid

JPC: Another thought is that thyroid concentrations may also matter. If lithium induces a slightly hypothyroid effect, people will gain weight that way too, since common (even classic) symptoms of hypothyroidism are weight gain and decreased activity. (It also proposes an immediate hypothesis [look at T3 vs TSH] and intervention [give people just a whiff of levothyroxine and see if it helps].) There’s also some thought that lithium maybe impacts thirst (full disclosure have not read this article except the abstract)?

SMTM: Also a good note, and yes, we do see signs of thyroid concentration. Some sort of thyroid sample would also be less invasive than a brain sample, right? 

JPC: Yes. We routinely biopsy thyroid under ultrasound guidance for the evaluation of thyroid nodules (i.e. malignant vs benign). These biopsies might be a source of tissue you could test for lithium, but I’m not sure. The pathologists may need all the tissue they get for the diagnosis, they may not. Doing it on healthy people might be hard because it’s expensive (you need a well-trained operator) and more importantly it’s not a risk free procedure: the thyroid is highly vascular and if you goof you can hit a blood vessel and “brisk bleeding into the neck” is a pretty bad problem (if rare).

That said, it is definitely less invasive than a brain biopsy, and actually safer than the very low bar of “less invasive than a brain biopsy” implies.

Clinical

SMTM: Do you have clinical experience with lithium? 

JPC: Minimal but non-zero. I had a couple of patients on lithium during my psychiatry rotation and I think one case of lithium toxicity on my toxicology rotation. I do know a lot of doctors, though, so I could ask around if they’re simple questions.

SMTM: Great! So, trace doses might be the whole story, but we’re also concerned about possible lithium accumulation in food (like we saw in the wolfberries in the Gila River Valley). We wonder if people are getting subclinical or even clinical doses from their food. We do plan to test for lithium in food, but it also occurred to us that a sign of this might be cases of undiagnosed lithium toxicity. 

Let’s make up some rough numbers for example. Let’s say that a clinical dose is 600,000 µg and lithium toxicity happens at 800,000 µg. Let’s also say that corn is the only major crop that concentrates lithium, and that corn products can contain up to 200,000 µg, though most contain less. Most of the time you eat fewer than four of these products a day and get a subclinical dose of something like 50,000 – 300,000 µg. But one day you eat five corn products that all happen to be high in lithium, and you suddenly get 1,000,000 µg. You’ve just had an overdose. If common foods concentrate lithium to a high enough level, this should happen, at least on occasion. 

If someone presents at the ER with vomiting, dizziness, and confusion, how many docs are going to suspect lithium toxicity, especially if the person isn’t on prescription lithium for bipolar? Same for tremor, ataxia, nystagmus, etc. We assume (?) no one is routinely checking the lithium blood levels of these patients for lithium, that no one would think to order this blood test. Even if they did, there’s a pretty narrow time window for blood levels detecting this spike, as far as we understand. 

So our question is something like, if normal people are occasionally presenting with lithium toxicity, would the medical system even notice? Or would these cases be misdiagnosed as heavy metal exposure / dementia / ischemic stroke / etc.? If so, is there any way we can follow up with this? Ask some ER docs to start ordering lithium tests in any mystery cases they see? Curious to know what you think, if this seems at all plausible or useful.

JPC: I have a close friend who is an ED doc! She and I talked about it and here’s our vibe:

With a presentation as nonspecific as vomiting, dizziness, and confusion, my impression is that most ED docs would be unlikely to check a lithium level, especially if the patient is well enough to say convincingly “no I didn’t take any pills and no I don’t take lithium.” At some point, you might send off a lithium level as a hail-Mary, but there are so many things that cause this that a very plausible story would be: patient comes to ED with nausea/vomiting, dizziness, and altered mental status. The ED gives maybe fluids, checks some basic labs, does an initial workup, and doesn’t find anything. Admits the patient. The next day the admitting team does some more stuff, checks some other things, and comes up empty. The patient gets better after maybe 24-48h, nobody ever thinks to check a lithium level, and since the patient is feeling better they’re discharged without ever knowing why.

Another version would go: patient is super sick, maybe their vomiting and diarrhea get them super dehydrated and give them an AKI (basically temporary kidney failure). People think “wow maybe it’s really bad gastritis or some kind of primary GI problem or something?” The patient is admitted to the ICU with some kind of gross electrolyte imbalance because they’re in kidney failure and they pooped out all their potassium, someone decides they need hemodialysis, and this clears the lithium. Again the patient gets better, and everyone is none the wiser.

Tremor, ataxia, nystagmus, etc. are more focal signs and even if someone doesn’t have a history of lithium use, and in this case our impression is that people would be more likely to check a lithium level. We also think it wouldn’t always happen. Even in classic presentations of lithium toxicity, sometimes people miss the diagnosis. (Emergency medicine is hard; people aren’t like routers where they blink the link light red when the motherboard is fried or power light goes orange if the AC is under voltage. Things are often vague and complicated and mysterious.)

Something you’d have to explain is how this isn’t happening CONSTANTLY to people with really borderline kidney function. Perhaps one explanation might be that acute lithium intoxication (i.e. not against a background of existing lithium therapy) generally presents late with the neuro stuff (or so I hear).

We think that this is plausible if it is relatively uncommon or almost always pretty mild. If we were having an epidemic of this kind of thing (like on the scale of the obesity epidemic) I think it would be weird that nobody has noticed. Unless of course it’s a pretty mild, self-resolving thing. Then, who knows! AFAIK still nobody really knows why sideaches happen—figuring it out just isn’t a priority.

On occasion, the medical-scientific community also has big misses. There’s an old line that “half of what you learn in medical school is false, you just don’t know which half.” We were convinced until 1982 that ulcers were caused by lifestyle and “too much acid”; turns out that’s completely wrong and actually it’s bacteria. I saw a paper recently that argued that pretty much all MS might be due to EBV infection (no idea if it’s any good).

I think you could theoretically “add on” a lithium level to anybody that’s getting a head CT with the indication being “altered mental status.” “Add on” just means that the lab will just take the blood they already have from the patient and run additional testing, if they have enough in the right kind of tube. The logic is that patients with new-onset, dramatic, and unexplained mental status changes often get head CTs to rule out a bleed or other intracranial badness, so a head CT ordered this way could be a sign that the ordering doc may be feeling stumped.

If you wanted to get fancy, you could try to come up with a lab signature of “nausea/vomiting/diarrhea of unclear origin” (maybe certain labs being ordered that look like a fishing expedition) and add on a lithium there as well. 

SMTM: Good point, but, isn’t it possible that it IS happening constantly to people with really borderline kidney function? The symptoms of loss of kidney function have some overlap with the symptoms of lithium intoxication, maybe people with reduced kidney function really do have this happen to one degree or another whenever they draw the short straw on dietary lithium exposure for the day. Lots of people have mysterious ailments that lead to symptoms like nausea and dizziness, seemingly at random.

Or we could look at it from the other angle — lithium can cause kidney damage, kidney disease is (very roughly) correlated with obesity at the state level, and as far as we can tell, rates of kidney disease are going up, right? Is it possible that many cases interpreted as chronic kidney disease are “actually” chronic lithium intoxication?

JPC: I guess it’s definitely possible. The “canonical” explanation to this would be that diabetes (which is obviously linked to obesity) destroys your kidneys. But, if it’s all correlated together as a vicious cycle (lithium → obesity → CKD → lithium) that’s kind of appealing too. I bet a lot is known about the obesity-diabetes-kidney disease link though and my bet without looking into it would be that there’s some problem with that hypothesis.

My thought here was that if people with marginal/no kidney function are getting mild cases, I would expect people with normal kidney function to be basically immune. Or, if people with normal kidney function get mild cases, people with marginal kidneys should get raging cases. This is because serum levels of stuff are related to the inverse of clearance. The classic example is creatinine, which is filtered by the kidney and used as a (rough) proxy for renal function.

SMTM: This is super fascinating/helpful. For a long time now we’ve been looking for a “silver bullet” on the lithium hypothesis — something which, if the hypothesis is correct, should be possible and would bring us from “plausible” to “pretty likely” or even “that’s probably what’s going on”. For a long time we thought the only silver bullet would be actually curing obesity in a sample population by making sure they weren’t consuming any lithium, but that’s a pretty tall order for a variety of reasons, not least because (as we’ve been discussing) the kinetics remain unclear! But recently we’ve realized there might be other silver bullets. One would be finding high levels of lithium in food products, but there are a lot of different kinds of foods out there, and since the levels are probably lognormal distributed you might need an exhaustive search. 

But now we think that finding people admitted to the ER with vague symptoms and high serum lithium, despite not taking it clinically, could be a silver bullet too. Even a single case study would be pretty compelling, and we could use any cases we found to try to narrow down which foods we should look at more closely. Or if we can’t find any of these cases, a study of lithium levels in thyroid or in bone could potentially be another silver bullet, especially if levels were correlated with BMI or something. 

JPC: I’m always hesitant to describe any single experiment as a silver bullet, but I agree that even a single case report, under the right conditions, of high serum lithium in someone not taking lithium would be pretty suspicious. You’d have to rule out foul play and primary/secondary gain (i.e. lying) but it would definitely be interesting. As far as finding lithium in bone or thyroid (of someone not taking lithium), I’d want to see some kind of evidence that it’s doing something, but again it’d definitely be supportive.

SMTM: Absolutely. We also don’t really believe in definitive experiments. The goal at this stage is to look for places where there might be evidence that could promote this idea from “plausible” to “likely”.