Edible Formulation and Dose Uniformity

Lipophilicity: why the active must be in a fat phase or an emulsion human data

Cannabinoids are extremely lipophilic and essentially insoluble in water. That single physical property determines everything about edible formulation. An active that will not dissolve in the aqueous phase of a food cannot be absorbed from the gut in any meaningful quantity, because absorption across the intestinal epithelium requires the molecule to be presented in a solubilised state — dissolved in dietary lipid and incorporated into mixed micelles by bile salts, or already dispersed as fine solubilised droplets by the formulation itself. A cannabinoid stirred as a powder into a water-based food is not a low-dose product, it is a product where much of the dose passes through. This is also why food effects on oral cannabinoid pharmacokinetics are large and well documented, with a high-fat meal raising exposure substantially over a fasted state, and why lipid excipients are not inert bulk. The formulator has two legitimate routes: carry the active in a genuine fat phase of sufficient quantity, or build an emulsion so the active is presented as a fine dispersed lipid phase regardless of what the consumer ate.

• Absorption requires solubilisation. Fat in the product or fat in the meal does that work; without either, absorption is poor and erratic.
• Route one, fat phase: chocolate, butter or oil-based confectionery where there is enough lipid to dissolve the dose.
• Route two, emulsion: beverages, gummies and low-fat formats where the lipid phase is engineered rather than incidental.
• The food effect is a source of dose variance the producer cannot control, which is part of the argument for emulsified formats.
• A gummy is a mostly aqueous gel. Getting an even dose into one is an emulsification problem, not a mixing problem.

Sources: Huestis MA 2007* · Zgair A 2016* · Millar SA 2018* · Barrus DG 2016*

Emulsification as formulation science in vitro

An oil-in-water emulsion is a kinetically stabilised dispersion of lipid droplets in an aqueous continuous phase, and every part of that phrase is a formulation lever. The emulsifier adsorbs at the oil-water interface, lowers interfacial tension so droplets can be broken up by the energy the process supplies, and then provides a steric or electrostatic barrier that stops them coalescing again. Energy input is what sets droplet size: gentle stirring gives a coarse emulsion with droplets of many micrometres, while high-pressure homogenisation, microfluidisation or high-intensity ultrasound produce droplets in the tens to low hundreds of nanometres. Droplet size then governs two things that matter commercially. Optically, fine emulsions are translucent rather than milky, which is why a clear cannabinoid beverage is a claim about droplet size. Pharmacokinetically, smaller droplets mean far greater interfacial area and faster lipolysis and solubilisation, which shortens time to onset and raises exposure — the mechanism described on the carriers page, where the honest limits of the commercial claims are also set out. Emulsion stability over shelf life, not at fill, is the specification that a producer should be testing: droplet growth by coalescence and Ostwald ripening, creaming, and the pH and ionic-strength sensitivity of protein- and hydrocolloid-stabilised systems all move the product over months.

• Coarse emulsion: droplets of micrometres, opaque, made with low-energy mixing and a food emulsifier like lecithin.
• Fine or nanoemulsion: droplets in the tens to low hundreds of nanometres, translucent, made with high-pressure or ultrasonic energy plus a high-power surfactant.
• Faster onset is a real and reproducible consequence of fine droplets; the size numbers printed on packaging usually are not verifiable.
• Instability modes to test for: creaming, coalescence, Ostwald ripening, and phase separation on freeze-thaw and thermal cycling.
• An emulsion that has crept in droplet size has also changed its absorption profile, so stability testing is dose testing.

Sources: McClements DJ 2012* · Barrus DG 2016* · Millar SA 2018* · Rowe RC 2020*

First-pass metabolism and 11-hydroxy-THC: the mechanism behind harder and later contested human data

The pharmacology that makes edibles behave differently from inhalation is not mysterious and it belongs in a formulation reference because it is a consequence of the route the formulator chose. An ingested cannabinoid is absorbed from the gut into the portal circulation and passes through the liver before it reaches systemic blood. Hepatic metabolism of Δ9-THC proceeds substantially by hydroxylation at the C-11 position to 11-hydroxy-Δ9-THC, which is itself psychoactive — characterised in human work from the early 1970s onward as at least as potent as the parent compound and crossing into the brain readily — before further oxidation to the inactive 11-nor-9-carboxy-THC that urine screening detects. Inhalation bypasses that pass: the drug goes from alveolus to arterial blood to brain, and the 11-hydroxy metabolite is formed in much smaller proportion relative to parent THC. The practical result is a route difference in three dimensions at once. Timing: absorption and hepatic transit take tens of minutes to hours, so onset is delayed and the peak is late, and a user who does not feel an effect at 45 minutes and takes a second unit is re-dosing before the first has peaked. Character: the metabolite profile the brain sees is different, with a much larger 11-hydroxy contribution, which is the mechanistic account of why an oral dose is described as stronger and more bodily at the same nominal milligrams. Variance: bioavailability is low and highly variable between people and between meals, so the same labelled dose produces a much wider spread of exposures than an inhaled one. The clinical epidemiology matches the pharmacology, with emergency presentations attributable to edibles over-represented relative to their share of sales.

• Ingested: gut, portal vein, liver, then systemic. Inhaled: alveolus, arterial blood, brain.
• 11-OH-THC is an active metabolite formed disproportionately on the oral route and is at least as potent as Δ9-THC in human studies.
• Delayed onset plus a late peak is the condition under which re-dosing errors happen; it is a formulation and labelling problem, not only a user-education one.
• Oral bioavailability is low and variable, commonly quoted in the region of 4 to 20 percent, with large food effects on top.
• Higher variance at the same label means a unit dose has to be conservative for the label to mean anything across a population.

Contested — caveat. The existence and activity of 11-OH-THC and the route difference in its formation are well established. The commonly repeated claim that the metabolite is a specific multiple more potent than Δ9-THC comes from small early human studies with different endpoints and should not be quoted as a fixed number. Oral bioavailability ranges vary widely across the literature depending on dose form, meal state and analytical method.

Sources: Lemberger L 1972* · Huestis MA 2007* · Wall ME 1983* · Monte AA 2019* · Millar SA 2018*

Dose uniformity per unit: the arithmetic and why the batch average is not the answer human data

Dose uniformity is the one safety property an edible producer fully controls, and the arithmetic is simple enough that getting it wrong is always a process failure rather than a knowledge failure. The nominal per-unit dose is total active in the batch divided by the number of units: 2000 mg of THC dosed into 400 gummies is 5 mg per gummy. That calculation is necessary and it is not the safety statement. The safety statement is about variance between units, because a consumer eats one unit and not the batch average. A batch that assays correctly at 5 mg per unit on average but ranges from 1 mg to 14 mg across the tray is a compliant average and a dangerous product, and it is dangerous in a specific way: the consumer who receives the 14 mg unit is also the consumer who most likely calibrated their expectations on an earlier 1 mg unit from the same box. The distribution, not the mean, is the specification. Two corollaries follow. First, potency losses have to be accounted for in the input calculation rather than discovered at assay, because heat, pH and process time destroy some of what went in. Second, the honest quantity to report internally is a mean with a spread, and a production process should be characterised by that spread before it ships.

• Nominal per-unit dose = total active in the batch ÷ number of units. Necessary, not sufficient.
• The specification that matters is the between-unit distribution: mean plus spread, not mean alone.
• Account for process loss on the input side. If the process destroys a known fraction, the batch input is sized for it and verified, not guessed.
• A hot spot is the failure this arithmetic exists to prevent: a local region of the matrix carrying a multiple of the nominal dose.
• A published audit of commercial edible products found that only a minority were accurately labelled for cannabinoid content, with both under- and over-labelling present. This is a documented industry-wide failure, not a hypothetical.

Sources: Vandrey R 2015* · Barrus DG 2016* · MELEK hemp-science shelf 2026*

Geometric dilution and full incorporation industry practice, not published data

The technique that produces uniformity is geometric dilution, and it is old pharmacy practice for exactly this problem. A small mass of potent active cannot be distributed evenly into a large mass of matrix in one step, because the mixing process can only reduce the scale of segregation so far per pass, and the difference in scale between a few grams of distillate and fifty kilograms of gummy slurry is too large. Geometric dilution solves it by doubling: the active is combined with a roughly equal mass of a compatible diluent and mixed to homogeneity, then that premix is combined with an equal mass of matrix and mixed, and so on until the batch is made. Each step is a mix over a manageable ratio, and the number of steps grows only as the logarithm of the dilution factor. For a cannabinoid the compatible diluent for the first steps is a lipid or the emulsion concentrate, because the active must be in solution before it is dispersed — attempting to disperse undissolved resin gives a suspension of concentrated droplets, which is a hot spot by construction. Two process facts complete the picture. Viscosity fights you: a viscous or partially set matrix does not mix in any useful sense, because flow is laminar and the fluid elements do not exchange, so a concentration gradient formed before the matrix stiffened is trapped permanently. And set time is a deadline: in a gelling system the window in which the batch can still be homogenised closes as the gel network forms, so the mix has to be complete before the set begins and depositing has to be fast enough that the last mould is filled from the same fluid as the first.

• Dissolve first, disperse second. An undissolved particle of active is a hot spot that mixing will not remove.
• Double the mass at each step rather than adding the active straight to the batch.
• Verify homogeneity of the premix before it goes into the batch. An inhomogeneous premix cannot be fixed downstream.
• Viscous and setting matrices trap gradients: laminar flow does not mix, and a gel does not flow at all.
• Depositing from a held reservoir has its own drift — the active can migrate or settle in the reservoir over a long run, which is why the sampling plan below spans the run.

Sources: MELEK hemp-science shelf 2026* · Barrus DG 2016* · Rowe RC 2020*

Verifying uniformity: the sampling plan industry practice, not published data

A producer does not know a batch is uniform because the process was designed to be uniform; they know because units from across the run were assayed individually and the spread was measured. The minimum honest plan is to pull units from the start, the middle and the end of a depositing run, assay them individually rather than compositing them, and record each result. Compositing defeats the purpose: grinding five gummies together and assaying the blend recovers the mean the producer already calculated and destroys the only information the exercise was for, which is the variance. Positional sampling matters because the two dominant failure mechanisms are both positional. Settling or creaming in the holding reservoir produces a drift from start to end of run, so a start-and-end comparison detects it directly. Incomplete incorporation produces scatter without a trend, which shows up as a wide spread among units taken from the same position. For moulded products, position within the mould tray is a third axis worth sampling if the depositing head traverses. The output of the exercise is a characterised process: a mean, a spread, and a known worst case, which is what allows a per-unit label to be defended.

• Assay individual units, never a composite, when the question is uniformity.
• Sample start, middle and end of the run to catch reservoir settling as a trend.
• Scatter without a trend points at incorporation; a trend points at the reservoir or the depositor.
• Record it as a process characterisation, so a later drift is detectable against a baseline rather than judged against nothing.
• A compliance or customer-facing result needs an accredited laboratory; in-house assay is for knowing where you are. See the analytical page on the equipment shelf.

Sources: Vandrey R 2015* · MELEK hemp-science shelf 2026*

Thermal and pH stability during production and on the shelf in vitro

The dose in the product at sale is not the dose that went in, and the chemistry of the gap is well characterised. Decarboxylation of the acid cannabinoids to their neutral forms is temperature- and time-dependent and continues wherever the product sees heat, which means an incompletely decarboxylated input keeps converting during cooking and in storage and the total-THC figure moves. Oxidative degradation of Δ9-THC to cannabinol continues in the finished product for the whole of its shelf life, driven by oxygen, light and temperature, and the classic storage work identified light as the single largest factor in cannabis preparation losses, with subsequent long-term studies confirming progressive loss under ambient conditions. Acidic conditions matter in gummy and beverage systems, because low pH accelerates cannabinoid degradation and isomerisation over time, and a citric-acid-sharpened confection is a mildly acidic reactor held at room temperature for a year. The practical consequences are a formulation input calculation that accounts for expected loss, a packaging specification that takes oxygen and light seriously rather than aesthetically, and a shelf life that is dated from manufacture and supported by real stability data rather than assumed.

• Decarboxylation is time-at-temperature. An input that is not fully decarboxylated will keep converting in the product.
• Δ9-THC to CBN oxidation proceeds on the shelf. Light is the dominant driver identified in the storage literature; oxygen and heat compound it.
• Low pH in gummies and beverages accelerates cannabinoid loss and isomerisation over shelf life.
• Packaging is a formulation component: opaque or UV-blocking material, low oxygen headspace or an oxygen barrier, and a seal that survives distribution.
• A rising CBN figure on a retained-sample assay is a readout of how the product has been stored, which the CBN page treats at length.

Sources: Fairbairn JW 1976* · Trofin IG 2012* · Zamengo L 2019* · Wang M 2016* · Huestis MA 2007*

Labelling arithmetic: per unit and per package human data

The labelling consequence of everything above is that a package total without a per-unit figure is unusable information. A consumer holding a bag labelled 100 mg THC cannot determine a dose from it, because the number they need is the milligrams in the piece they are about to eat, and dividing by the count on the bag assumes a uniformity they have no way to verify and that the published audit literature says is frequently absent. A defensible label carries both numbers and states the count: milligrams per unit, units per package, and milligrams per package, with the arithmetic between them consistent. Two further points belong on any serious label. The delayed-onset warning is not boilerplate; it is the direct consequence of the pharmacokinetics above and the specific mechanism by which the most common acute harm from this product category occurs, so it should state a time and not merely say effects may be delayed. And where the active is a cannabinoid other than Δ9-THC, or where the product contains a substantial acid-form contribution, the total-THC arithmetic has to be stated on the same basis the regulator uses, which is the subject of the total-THC-math page. Most jurisdictions additionally cap per-serving and per-package amounts, frequently at 5 or 10 mg per serving, and the cap is a property of the unit rather than of the average.

• Required for use: milligrams per unit. Everything else is context.
• State count and package total as well, and make the three numbers consistent.
• A package total alone is not a dose statement and should be treated as a labelling defect.
• The onset warning should carry a time figure, because the harm mechanism is re-dosing inside the onset window.
• Per-serving caps are common and jurisdiction-specific; check the operative rule rather than a neighbouring state.
• Score, wrap or portion the product so the unit the label describes is the unit a consumer can actually separate.

Sources: Vandrey R 2015* · Monte AA 2019* · Barrus DG 2016* · MELEK hemp-science shelf 2026*