Mild-Condition Subcritical Water Extraction of Kavalactones from Mulched American-Grown Kava

The problem is not yield, it is physical state

Kava prepared the traditional way is a suspension, not a solution. Most of the kavalactone load a drinker receives rides on fine plant solids that pass the strainer rather than dissolving in the water, and on these same three American cultivars that share has been measured directly: in short warm-water preparations the genuinely dissolved fraction ran between 5.4% and 31.9% of delivered kavalactone, and no water pH tested stopped particle-associated delivery from dominating. For a shell of kava that is not a defect — the suspension is drunk intact, solids and all. For anything that has to be filtered, dried and sold as a concentrate it is fatal, because fine filtration removes the particles and the kavalactones they carry go with them.

This study asked whether a mild sealed-vessel heat step can move kavalactones off the solids and into a phase that survives filtration, and how far the result sits from the published subcritical-water reference for kava.

Why hot water under pressure behaves like a different solvent

Water held above its atmospheric boiling point in a sealed vessel stays liquid under its own vapour pressure, and as temperature climbs its dielectric constant falls — from about 80 at ambient conditions toward about 2 near the critical point. That progressive fall is what lets hot water dissolve moderately lipophilic compounds that cold water will not touch. At 120.6 °C (249 °F) the dielectric constant is about 65: a real reduction, achieved without a drop of organic solvent.

The pressure that condition requires is only the pressure that keeps the water liquid — roughly 2.0 bar absolute, about 15 psig. That number is the whole point of the design, and it is returned to at the end.

The reference, and one thing it is easy to misread

The foundational subcritical-water study on kava is Kubátová, Miller and Hawthorne (2001), who ran kava root from 100 to 175 °C in a dynamic flow-through cell against an exhaustive 18-hour acetone sonication taken as 100%. Two features of it matter here. Water alone, with no post-cell dichloromethane, extracted as well as water plus co-solvent — so the gains at temperature come from the dielectric shift in the water itself, not from solvent chemistry. And their tabulated concentrations sit on a single mg/g axis against that exhaustive acetone reference of 125 ± 6 mg/g.

The easy misreading is this. Their abstract and kinetic figures describe complete extraction of the ground fraction in 2 h at 100 °C, yet the tabulated total for that same condition is 90 ± 6 mg/g against 125 ± 6 — 72%, not 100%. The two statements do not conflict: the kinetic figures are normalised to sonication of the residue left after water extraction, so they report exhaustion of the water-extractable pool, while the tabulated concentrations are referenced to exhaustive acetone on fresh material. This study uses the exhaustive-acetone axis throughout, because that is the axis the marc-difference estimator used here also sits on.

On that axis the temperature effect is steep across exactly the range of interest: 51 mg/g at 100 °C, 93 at 125 °C, 111 at 150 °C, 121 at 175 °C — 41%, 74%, 89% and 97% of the exhaustive reference. The operating temperature used here, 120.6 °C, sits just below their 125 °C point, so roughly 75% of exhaustive is the working ceiling this condition could be expected to approach. That figure is an interpolation from their published series, and it applies to sieved material under continuous flow.

How it was run

Root of three American-grown cultivars — Kali-ISA, Kali-Rogu and Kali-Hiwa, all from the Kali Kava farm in Hendry County, Florida — was mulched in a commercial food processor immediately before use, giving a mixed particle size that spans rather than matches the sieved 250–500 µm and shredded 2–4 mm fractions of the reference. Each charge was pre-soaked in reverse-osmosis water at 38 °C for 2 h, then sealed in a bench-scale Parr vessel and held for 2 h at 120.6 °C under autogenous pressure. Water was the only substance the root met at any stage: no acid, base, salt, solvent, surfactant, enzyme or processing aid, at the soak, the hold or the filtration.

The pre-soak is a conditioning step, not an extraction step. Kava root is roughly 43% starch by dry weight and the kavalactones sit in resin bodies embedded in that starchy matrix; dry granules impede aqueous diffusion. Soaking at 38 °C — well below starch gelatinisation and below any extraction-relevant threshold for the kavalactones — lets the 2-hour hold act on a wetted matrix from the start rather than spending part of its duration on imbibition. No unhydrated control was run, so the step's own contribution is not isolated here.

Each run was a single wash. The spent marc was separated, dried and set aside without re-extraction. Kali-ISA was run twice under identical conditions; Kali-Rogu and Kali-Hiwa once each. Eleven samples — three Raw, four marc, four extract — went to Flora Research Laboratories for HPLC-UV. Efficiency is the fraction of the root's kavalactones removed from the solids, computed as the difference between the Raw and marc assays on a moisture-corrected dry-weight basis.

What one wash removed

A single 2-hour wash at 120.6 °C took 64.3% of the kavalactones out of Kali-Rogu, 49.0% and 47.8% out of the two Kali-ISA runs, and 41.0% out of Kali-Hiwa. The spread across the four runs is 23.3 percentage points. The two runs on the same material, under identical conditions, differ by 1.2 points.

That ratio is the central quantitative observation of the study: the cultivar spread is roughly twenty times the run-to-run difference. It is why the cultivar effect is reported as a real feature of the dataset rather than as possible measurement noise.

What stayed behind

The complement is substantial in every run. Kali-Rogu left 35.7% of its starting load in the spent marc, the two Kali-ISA runs 51.0% and 52.2%, and Kali-Hiwa 59.0%. That material was not destroyed and did not go into the extract — it sat in a solid residue that was separated and set aside. Every efficiency here is therefore a per-cycle figure for one wash of a static batch, and the cumulative recovery available from the same charge under sequential washes is not measured.

Extract potency, which is not the same quantity

The dehydrated products ranged more than fourfold in potency: 2.44% total kavalactones for Kali-Hiwa, 5.30% and 5.94% for the two Kali-ISA extracts, and 11.2% for Kali-Rogu. Kali-Rogu is the one case where the dried extract assayed higher than its own starting root (11.2% against 10.5%), meaning the recovered solids were enriched in kavalactones rather than diluted by co-extracted material.

Potency and efficiency are not interchangeable. Efficiency depends only on what left the root; potency also depends on how much non-kavalactone solid came across with it. The two Kali-ISA extracts differ by 0.64 points of potency — about 11% of their mean — against 1.2 points of efficiency on the same pair of runs, so potency is the looser of the two measurements.

Where that lands against the benchmark

Because efficiency here is a fraction of the root's total kavalactones, it sits on the same percentage-of-recoverable axis as the reference values, all of which are read against an exhaustive 18-hour acetone sonication of 125 ± 6 mg/g. Against the conventional atmospheric-pressure methods the mild condition is clearly ahead: water Soxhlet for 6 h returned 48 ± 6 mg/g (38% of exhaustive) and boiling water for 2 h returned 57 ± 16 mg/g (46%), so all four runs beat Soxhlet and three of four beat boiling, with Kali-Hiwa at 41.0% falling 4.6 points short of it. Kali-Rogu at 64.3% clears both by a wide margin and lands between the 60-minute and 120-minute points of the 100 °C subcritical series (81 ± 5 mg/g, 65%, and 90 ± 6 mg/g, 72%). The high-temperature optimum, 175 °C for 20 minutes, returned 104 ± 10 mg/g — 83%.

Against the interpolated ceiling for this temperature — about 75% of exhaustive — the four runs captured 55% (Kali-Hiwa), 64% and 66% (the Kali-ISA pair) and 86% (Kali-Rogu) of what the reference series suggests is available at 120.6 °C. A single static wash on food-processor mulch at about 2.0 bar therefore reached between roughly half and seven-eighths of the temperature-matched ceiling established under continuous flow at about 60 bar on sieved material.

Three things differ from the reference at once — pressure is roughly thirtyfold lower, the matrix is coarser and unsieved, and the extraction is a single static wash rather than a flow-through that continuously displaces the partition equilibrium. This study cannot apportion the gap among them. The comparison is a placement on a shared axis, not a controlled contrast.

The one thing the gap tells you for certain

Of those three departures, wash number is the one for which the data give a positive indication rather than only a plausible mechanism. The marc retained 36–59% of the starting load in recoverable form, so additional washes have material available to take. It is also the cleanest of the three to isolate, because it requires no change to the equipment or the matrix. Sequential re-extraction of the marc under the same mild condition is underway, and the single-wash figures reported here should be read as a per-cycle floor rather than the ceiling of the method.

One nuance will need watching in that work. Because the marc preferentially holds the more lipophilic yangonin, further mild-condition washes may keep recovering the water-compatible kavalactones more readily than the lipophilic residue — so cumulative recovery could rise while the compositional gap between successive washes widens. Whether multi-wash recovery converges on the reference or plateaus below it because of a compound-selectivity limit is an empirical question this dataset cannot answer.

The duplicate, and what one duplicate is worth

The two Kali-ISA runs agree to 1.2 percentage points on a mean of 48.4% — about 2.5% in relative terms. That is worth stating plainly, because the protocol has several manual stages (food-processor mulching, slurry transfer, filtration, tray dehydration), any of which could plausibly have introduced several points of scatter. On this pair, they did not.

The weight to put on that is limited in two specific ways. One duplicate gives one degree of freedom: it establishes that a 1.2-point agreement is attainable, not that it is typical, and no confidence interval can be built from it. And repeatability shown on Kali-ISA does not transfer to the other two cultivars, whose matrices differ enough to produce a 23-point efficiency spread and might equally differ in how consistently they run. Kali-Hiwa in particular — lowest efficiency, by far the lowest extract potency, most likely to be sensitive to handling losses — is unreplicated.

Cultivar tracked strongly, but cultivar is confounded with lot

Extraction followed cultivar hard, with a 23-point spread under a protocol identical in every run. Kali-Rogu led at 64.3% and produced the only extract that assayed above its own root, the signature of a matrix that releases kavalactones into the aqueous phase while contributing little co-solubilised ballast. That is consistent with its kavain-forward, lower-lipophilicity starting profile (chemotype 423561) being well matched to aqueous mobilisation at this temperature. Kali-Hiwa sits at the other end; Kali-ISA between them.

The association is real in the dataset but cannot be attributed to cultivar identity. Each cultivar is represented by one lot from one harvest, so cultivar is fully confounded with lot, harvest age, post-harvest handling and the particular mulch produced that day. Separating a genotype-level effect from lot-level variation needs independent biological replicates across plants and harvests. The result is reported here as a lot-level difference under a fixed protocol.

The chemotype series is the strongest evidence in the study

Across all three cultivars and all four runs the extracts were enriched in the lower-log-P kavalactones — dihydrokavain (log P ≈ 2.1) and kavain (≈ 2.2) rose toward the top of the extract rankings, while the more lipophilic yangonin (≈ 3.1) and methysticin (≈ 3.0) fell back. Kali-Rogu shifted from a kavain-led root (423561) to a dihydrokavain-led extract (245361); Kali-Hiwa from 463251 to 425361; Kali-ISA, already dihydrokavain-led, held 254631 in both extracts.

The complementary signature is in the marc. Yangonin, the most lipophilic of the six, ranked first in both the Kali-Rogu marc (342516) and the Kali-Hiwa marc (342156) — an inversion of its mid-to-low position in the corresponding roots and extracts. The Kali-ISA marcs carried dihydrokavain- and dihydromethysticin-led residual profiles.

That the ranking moves at all is the informative part, because the comparable aqueous variable does not move it. Across nine short warm-water preparations of these same three cultivars spanning initial water pH 5.0 to 8.0, every beverage returned its own root's chemotype with no rank position displaced. Water chemistry over that range left the proportional profile intact; raising the temperature to 120.6 °C rearranged it in three of the four runs here.

The reference corroborates the mechanism independently at the compound level: in the 100 °C rate experiment, 90% of most kavalactones were recovered by 90 minutes but only 40% of desmethoxyyangonin and 60% of yangonin — the two most lipophilic of the set — and the Soxhlet and boiling-water extracts were likewise depleted mainly in those two. Lipophilic lag under water-only extraction at mild temperature is a reproducible property of the system, not an artefact of these cultivars or this apparatus. It carries a practical consequence: the mild condition fractionates as well as extracts, and bringing the lipophilic fraction across at comparable efficiency would take a higher temperature.

What changed physically

Holding the slurry at 120.6 °C moved 41–64% of the load off the plant solids and into a phase that survived fine filtration, in all four runs, yielding a clarified, dehydratable extract instead of a heterogeneous suspension. That is the qualitative change the step produces, and it is the property that matters for concentrate manufacture.

This was observed rather than instrumented. No particle-size, turbidity or colloid analysis was run on the filtrate, so the filterable phase is not demonstrated here to be a true solution as opposed to a colloidal or micellar dispersion. What is established is that the material passes fine filtration and dries to a homogeneous solid.

Water only, and where that leaves the product

The exclusive use of water separates this protocol from the commercial extraction landscape. Commercial kava formulations have historically been ethanol, methanol or acetone extracts standardised to a target kavalactone content, and the toxicology record on which regulatory caution rests is weighted toward those organic-solvent extracts and toward high-dose extract studies. FDA's 2020 scientific review noted that the WHO case analysis found a higher rate of hepatotoxicity for organic extracts than for other products, that the documented European transplant cases involved ethanolic and acetonic extracts, and that the rodent carcinogenicity findings derive from a concentrated organic extract. The Hawaiʻi Department of Health, reviewing the same question, noted that organic extraction recovers two to ten times the total kavalactones obtained by aqueous extraction. This protocol introduces no organic solvent at any stage, and its lipophilic-depleted product is by construction further from the organic-solvent constituent profile than a higher-temperature or solvent extract would be.

The regulatory position of the product is more qualified than its composition alone suggests, and the distinctions are worth stating precisely. FDA's 2020 determination is limited by its own terms to kava used as an ingredient added to conventional foods, and in a 2024 letter to Congress the agency stated that kava steeped in water as a single-ingredient beverage would be regulated as a conventional food. Hawaiʻi determined noble ʻawa root mixed with water or coconut water to be generally recognised as safe under the pre-1958 common-use exception; Michigan reached the same practical conclusion for noble kava infused in water, by a different route.

Three features of this particular product sit outside what those instruments expressly address, and they are flagged rather than claimed. Codex CXS 336R-2020 lists the dried forms as intact lateral roots, peeled rhizomes, peeled chips or powder, expressly excludes extraction residues, and does not apply to the final beverage as such — an instant-style dehydrated extract is not among the listed forms. Standard 2.6.3 of the Australia New Zealand Food Standards Code permits a beverage obtained by aqueous suspension of kava root using cold water only; a 120.6 °C pressurised extract is not a cold-water suspension and would not qualify. On one further FSANZ requirement the protocol does conform — permitted kava must contain no substance used as a food additive or processing aid, and nothing beyond water was introduced at any stage. Temperature, not composition, is what places the product outside that standard. The U.S. determinations describe steeping and aqueous extraction without specifying a temperature limit, so they are silent rather than adverse — but silence is not recognition.

A standard autoclave already holds these conditions

The practical case for the mild condition is equipment. At 120.6 °C and roughly 2.0 bar absolute, the required parameters fall inside the routine operating envelope of a standard vertical steam autoclave, which by design holds 121 °C at about 15 psig. The runs reported here used a bench-scale Parr vessel, but the same conditions can in principle be reproduced at production scale on a modified vertical autoclave — equipment that is widely available, well understood by operators and already engineered for this exact temperature–pressure regime. Approaching the 175 °C optimum instead requires purpose-built high-pressure systems operating on the order of 300 psi and above, with the capital cost, engineering complexity and operational hazard that follow. The scale-up claim is an engineering inference from the operating envelope: no autoclave-scale run was performed in this study.

The results quantify the price of that accessibility. The method delivered 41–64% per wash over a 2-hour hold, against 83% of the exhaustive reference in 20 minutes at the high-temperature optimum. It trades yield per cycle and throughput for low capital cost, operational safety and food-grade equipment — a trade that suits small and mid-scale solvent-less production, with the higher-temperature regime reserved for applications that need near-quantitative single-pass recovery.

Limitations

Replication is limited to a single duplicate pair. Kali-Rogu and Kali-Hiwa are single determinations with no variance estimate of their own, and every potency and chemotype value in the study is a single determination. The 23-point cultivar spread is a point-estimate difference whose significance cannot be assessed. Triplicate replication on every cultivar is the first requirement of any follow-on work.

Cultivar is confounded with lot, so differences described as cultivar-level are properly lot-level.

The benchmark comparison is not a controlled contrast: three variables differ simultaneously and no run under the reference conditions was performed here. A within-study high-pressure arm and a within-study particle-size series are both needed to apportion the gap.

Particle size was not measured. The mulch is described qualitatively, so the matrix cannot be reproduced quantitatively from this description.

The efficiency estimator is an upper bound on recovered yield. It measures kavalactone removed from the solids, so thermal degradation, adsorption to vessel and filter surfaces and handling losses all score as extracted. Gravimetric mass closure was not established, and thermal stability of the six kavalactones under a 2-hour hold at 120.6 °C was not assessed.

Matrix-specific analytical recovery was not established: the HPLC-UV method is validated on dried root powder, and spike-and-recovery data for the dehydrated extract and the spent marc were not generated.

The filterable phase was not characterised, so the shift from particle-bound to filterable material is an operational observation rather than a demonstration of true dissolution.

Scope is narrow in both directions. Three cultivars from one grower cannot represent the diversity of Piper methysticum, and a single temperature, hold time, pre-soak temperature and water chemistry cannot characterise the response surface. No sensory, stability or biological endpoint was evaluated.

What happens next

Planned work varies temperature (100, 120.6, 140 and 175 °C), hold time, pre-soak temperature and particle size, adds sequential-wash cycles, and includes an arm omitting the pre-soak entirely so the hydration step's contribution can be isolated from that of the subcritical hold. Sequential-wash runs on the same cultivars are already underway. Until those are complete, the figures here should be read as replicable targets for an accessible, solvent-less route rather than as a characterisation of the method's ceiling.

Disclosures and conflicts of interest

Both authors are commercially engaged in the kava trade and are affiliated with Root of Happiness Kava, a California kava business that would stand to benefit from a commercially viable solvent-less extraction route, and the cultivars tested are American-grown material from Kali Kava. This research received no external funding; all materials, analytical services and laboratory costs were borne by the authors. The HPLC analyses were performed by Flora Research Laboratories, an independent laboratory with no interest in the outcome, on samples submitted without any of the interpretations offered here. The limitations above are stated in full, no run was excluded from the dataset, and the benchmark comparison is presented as a placement rather than as a controlled contrast.