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.

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.

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.  

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.