Can Urine Fluorescence Identify THC Without GC–MS? Four Cannabinoids, One Small Validation Study

What question did the original researchers actually test?

The research question was analytical: could fluorescence excitation–emission matrices and a mathematical decomposition distinguish overlapping signals from four cannabinoid-related standards, predict their concentrations in known mixtures, and produce results resembling GC–MS in a few processed urine samples? This is different from a driving-impairment experiment, a population-based urine screen, or a study of the interval between cannabis administration and specimen collection. No participant in this paper performed a driving simulator, standardized psychomotor examination, or observed on-road task.

The authors first examined separate purchased reference standards, then constructed three-component and four-component solutions with assigned concentrations. Only after calibrating and testing this chemical model did they examine the donated urine material. The order matters: the model’s ability to recover known added concentrations under a controlled chemical design does not automatically establish how it will perform in heterogeneous samples from routine patients or forensic cases. In particular, unknown medications, fluorescence-active matrix constituents, specimen handling differences, and compounds omitted from the model could influence practical specificity.

For litigation, the proper description is an exploratory analytical-method paper with a limited real-urine demonstration. Treating it as a clinically validated confirmatory urine assay or a direct blood concentration study would mistake both the specimen and the endpoint.

Which four analytes were modeled—and which ones are not interchangeable?

The four names were Δ9-tetrahydrocannabinol (THC), cannabidiol (CBD), cannabinol (CBN), and THC–COOH. THC is the parent cannabinoid commonly discussed in pharmacological and driving cases. CBD and CBN are chemically distinct cannabinoids; neither a CBD nor a CBN fluorescence signal is a measurement of active THC in a person’s blood. THC–COOH in forensic usage refers to a carboxy metabolite of THC, ordinarily 11-nor-9-carboxy-THC, and is different from 11-hydroxy-THC and from the plant cannabinoid precursor tetrahydrocannabinolic acid (THCA).

This nomenclature deserves care because the original article also describes THC–COOH as a biosynthetic precursor, even while discussing its relation to THC metabolism. A laboratory relying on the approach would need the exact chemical identity, authentic standard, retention or spectral criteria, and reporting label resolved before attaching forensic meaning to a result. An abbreviation cannot independently prove the chemical identity of the material generating an optical component.

None of the analytical concentrations in this paper represents an actual measured THC concentration in living participants’ venous whole blood, separated serum, or plasma. A urine marker may document that a particular analyte is present under the method’s validation rules; it does not become an active blood drug measurement by being converted from ng/mL to another unit.

How many independent donors and prepared mixtures were studied?

The Methods identify urine donated by two nonconsumer volunteers and two frequent cannabis users. The urine from nonconsumers was fortified with unspecified amounts of the four standards to form reference urine samples. The two consumer specimens were separately tested after laboratory preparation. These were four urine donors, not hundreds of independently recruited test subjects. The original source does not provide a clinical cohort sufficient to estimate the method’s sensitivity for cannabis exposure across ages, doses, collection times, renal functions, and co-exposures.

For the artificial chemical mixtures, the authors describe two families: one containing CBD, CBN, and THC–COOH, and another adding THC. They report designing 25 calibration and 10 prediction samples for each mixture type, although the printed Table 2 shows only representative mixtures and directs the reader to accompanying information for other entries. Each calibration solution is a chemical preparation. It is not a new cannabis-using participant, an independent urine donor, or a validated blind proficiency challenge drawn from operational casework.

The article’s plotted urine comparisons also use a figure label of n = 5; that notation must not be mistaken for five independent urine donors, given the Methods’ express description of four donors in total. Replicate readings, technical measurements, prepared standards, and distinct people must be counted separately. The design does not report a large group of known THC-positive and known THC-negative real-world urine samples from which a diagnostic confusion matrix could be calculated.

What did the instrument measure, rather than chemically separate?

The optical method used a Hitachi F-4600 luminescence spectrometer with excitation wavelengths from 230 to 360 nm and emission wavelengths from 250 to 550 nm, sampled in 5-nm increments. An excitation–emission matrix records fluorescence intensity across combinations of illuminating and emitted wavelengths. It does not, by itself, separate components by chromatographic retention time or record the mass-to-charge transitions of each specific molecule.

That distinction explains the need for PARAFAC, a parallel-factor computational model that estimates components from a three-way fluorescence data array. Its input axes represent different samples, excitation wavelengths, and emission wavelengths; a chosen number of components is used to reconstruct overlapping patterns. The researchers evaluated candidate component numbers with core-consistency diagnostics, residuals, loadings, and split-half analysis. They selected a three-component model for the solution without THC and a four-component model for the solution containing all four standards.

The authors compared modeled excitation and emission patterns with signals from individual reference solutions. That is meaningful support for their interpretation within the tested mixtures. But an internally consistent decomposition is still conditional on how the model was calibrated and on whether unmodeled compounds are present. A fluorescence component assigned the name THC is not equivalent to full molecular identification in an unknown forensic urine sample merely because an algorithm produces a numerical score.

Why do THC and CBN produce a particularly important spectral-overlap problem?

The researchers reported a strong shared fluorescence maximum for THC and CBN near 255 nm excitation / 310 nm emission. A single intensity measurement at that wavelength pair therefore would not reliably distinguish one from the other. CBN also displayed a second feature near 310/360 nm; the authors used the larger multiwavelength signal pattern to help distinguish the otherwise overlapping components. CBD was characterized by a peak near 235/300 nm, and the material labeled THC–COOH near 265/285 nm.

This finding illustrates a general analytical principle: when compounds share a response, the analytical answer depends on the entire validated measurement and model, not one attractive peak. The source describes strong agreement between predicted and measured component profiles in known mixtures. It does not show that all possible environmental, medicinal, food-derived, or endogenous fluorescent compounds can be excluded from the same predicted components in unknown biological specimens.

The authors also observed that CBN fluorescence ceased to increase linearly at concentrations around or above 2.0 µg/mL, suggesting self-quenching. Therefore, even a compelling response at low concentration cannot automatically be extrapolated to arbitrarily high concentrations. Forensic quantitative methods require defined working ranges, interference studies, carryover checks, matrix-specific verification, and data acceptance rules rather than relying on the qualitative appearance of a fitted spectrum.

What did the single-standard calibration actually show?

The original Table 1 reports positive relationships between concentration and fluorescence intensity in methanol standards, with coefficients of determination from 0.994 to 0.999. The corresponding standard-solution detection and quantification estimates appear below. These numbers are not the same as the performance of the later multicomponent model in urine, and they should not be transferred to a routine forensic case as if all matrices behaved identically.

These estimates are based on standard-solution regressions, with detection and quantification limits calculated from the dispersion of the regression intercept relative to the slope. The authors report relationships at n = 5, p < 0.01 for several individual standards. A significant calibration regression means intensity changes with prepared concentration under those conditions. It does not establish that any unknown specimen is a true positive, that a user’s exposure is recent, or that the measurement can discriminate impaired from unimpaired people.

For clarity, the limits in this table are in ng/mL. The mixture-model limits below are published in µg/mL. Those units differ by a factor of one thousand. Interchanging them is not a harmless formatting error: it can reverse whether a reported number is above or below the study’s own stated quantification threshold.

How well did the three- and four-component models recover known mixtures?

The authors fitted calibration models to the prepared solutions, then assessed predictions of compounds in mixtures. Their Table 3 provides average recovery, root-mean-square error of prediction (RMSEP), and limit of quantification (LOQ). These metrics are fundamentally different. Recovery asks whether the estimated concentration is close to a known reference level. RMSEP summarizes prediction error across prepared samples. LOQ indicates the proposed lower concentration boundary for quantitative reporting under the method’s conditions; it is not the same thing as a concentration that happens to be printed on a plot.

The tabulated average recoveries range from 95% to 107%; those averages are useful descriptions of the prepared mixtures but cannot substitute for evidence of casework precision across low-concentration, adulterated, or unanticipated urine matrices. For the four-component model, the reported THC RMSEP is 0.070 µg/mL, or 70 ng/mL. That magnitude matters when judging the weight of small differences in individual predicted values.

The source abstract summarizes a narrower RMSEP span of approximately 0.03–0.07 µg/mL, whereas its numerical Table 3 also prints values as small as 0.003 µg/mL. The original numeric table—not a repaired or guessed replacement—provides the analyte-specific entries used here. Any formal method validation should reconcile the summary, component definitions, reporting limits, and raw standards before a laboratory adopts a result as quantitatively defensible.

Why are the reported LOQs more consequential than the attractive low-concentration plots?

A key interpretive issue is the difference between the excellent-looking detection figures for individual methanol standards and the considerably higher LOQs assigned to the PARAFAC mixed-component models. Table 1 lists individual-standard QLs of a few to a few dozen ng/mL. Table 3, however, prints mixture-model LOQs of 0.26–0.71 µg/mL: 260–710 ng/mL. The latter limits govern interpretation of this more complex quantification task unless independently validated, matrix-specific performance data justify another reporting range.

The article also depicts and discusses model-predicted values on the order of 60–160 ng/mL in its low-concentration and urine figures. Many such displayed values are below every LOQ printed for the mixture models. Without access to a reconciled reporting convention, the relevant supporting analysis, and laboratory validation records, those plotted low numbers should not be presented to a court as fully validated quantitative concentrations simply because they agree visually with a GC–MS symbol.

Being below a published quantification limit does not automatically establish that no analyte is present. Rather, it changes the permitted analytical claim: the method may show a response or estimated concentration, but defensible numerical quantification at that level has not been demonstrated by the tabulated LOQ alone. Conversely, a high correlation or low average recovery bias in spiked test mixtures does not cancel the LOQ problem. These are separate performance characteristics and must each be reported accurately.

Importantly, this observation identifies a published measurement/reporting issue; it does not prove the researchers’ raw data are erroneous or establish the correct alternative LOQ. Such a correction would require the full underlying calibration calculations, the supplementary tables, matrix-specific quality controls, and a documented laboratory review. The source’s numbers should not be silently changed to make the graphs and limits appear consistent.

What urine processing was required before the fluorescence measurement?

The paper describes several substantial laboratory steps, rather than passing an untreated urine container under a light source. Its protocol treated a 2-mL urine aliquot with sodium hydroxide, heated it for 20 minutes, cooled and acidified it toward pH 3.5, extracted it with n-heptane under ultrasound, centrifuged it, combined organic phases, and evaporated solvent at 55°C under reduced pressure. For the fluorescence route, the resulting residue was dissolved in methanol before analysis.

For the comparison GC–MS route, the solid extract was derivatized with N,O-bis(trimethylsilyl)trifluoroacetamide at 70°C for 40 minutes, then analyzed on a Shimadzu GC–MS platform. This is not a head-to-head comparison between a completely reagent-free urine dip test and a GC–MS instrument. Both workflows depended on substantial sample preparation, and the study itself employed organic extraction solvent and methanol.

Accordingly, the paper’s broad characterization of the fluorescence process as “solvent-less” should not be repeated without specifying what operation that phrase is intended to cover. The optical readout may avoid a chromatographic separation step, but the described sample preparation plainly did not avoid all organic solvents. For a chain-of-custody or admissibility review, the extraction protocol, sample losses, possible chemical changes during alkaline heating, and method-specific recovery are relevant—not merely the advertised speed of spectral acquisition.

What happened in the actual donated urine samples?

The researchers tested two urine samples from people described as frequent cannabis users and two reference samples derived from nonconsumer urine with added analytes. The study says that the methods could not detect THC in the two consumer urine samples, which it attributed to the small amount remaining after metabolism. That observation is a real negative outcome within the reported experiment; it is not a validated finding that all frequent users lack urinary parent THC or that either instrument has a zero false-negative rate.

The paper gives a concrete fortified-reference example: for THC, the PARAFAC estimate was 65 ± 4.4 ng/mL against a GC–MS value of 68 ng/mL. On its face that is close agreement for the example. Yet the same article’s four-component Table 3 assigns a 260 ng/mL LOQ for THC. The 65-ng/mL model value is lower than that reported model LOQ, so any claim that the example is quantitatively validated without qualification would exceed the displayed evidence.

The authors also describe CBD and CBN mean relative errors of approximately −2.3% to +2.9% between methods in the limited urine tests. Those are promising experimental comparisons; they are not a representative operating-characteristic analysis for thousands of unknown specimens. Crucially, the true number of independent human urine donors remains four. The same donor’s repeated readouts and fortified aliquots cannot enlarge that biological cohort.

In practical forensic language, the actual user urines produced no reported parent-THC detection; the intentionally fortified reference matrix illustrated numerical estimates. The result supports further analytical development. It does not demonstrate that fluorescence alone could replace a validated GC–MS or LC–MS/MS confirmatory method for all legally consequential urine cases.

Did the investigators report significant results, nonsignificant results, or both?

There are positive findings: individual reference-standard concentration and fluorescence were strongly linearly related, with selected reported correlations significant at p < 0.01; PARAFAC fitted three or four expected components; tabulated model recoveries clustered around 100%; and several fortified-reference estimates resembled GC–MS results. These findings concern analytical calibration, model fitting, and prepared-sample agreement under the reported conditions. They should not be translated into clinical diagnostic accuracy or individual forensic validity without separate performance testing.

There are also important negative or unestablished findings, although not all are formal nonsignificant hypothesis tests. The two actual cannabis-user urine specimens reportedly yielded no detected parent THC. CBN fluorescence lost simple linearity at higher prepared concentrations, demonstrating a model-domain boundary. The study did not report a statistically powered specificity comparison across diverse nonuser and user populations, a prospective false-positive/false-negative rate, an individual impairment prediction, or a blood–urine correlation. An absence of those experiments is not a formal p-value showing equivalence or irrelevance.

Even if a calibration regression is statistically significant, it can coexist with poor selectivity for unknown contaminants, high LOQ, or too few operational specimens to estimate performance precisely. The proper evidentiary weight follows the question answered by each test, not the mere existence of an impressive p-value.

Does a modeled urine THC–COOH signal identify parent THC in someone’s blood?

No. Urine, blood, serum, plasma, oral fluid, hair, and exhaled breath answer different biological questions. Urine reflects renal excretion and specimen dilution; a carboxy metabolite is chemically and pharmacokinetically different from parent THC. Separated serum and plasma omit cellular blood components; whole blood includes them. Concentration readings from one compartment cannot be converted into another by assuming numerical equality. The study did not measure corresponding blood specimens from its four urine donors at relevant times.

Moreover, a measurement of an indirectly reconstructed optical component is not a behavioral observation. Whether a particular person was intoxicated, could safely perform a task, was driving, experienced a collision, or caused an injury cannot be inferred solely from a urine fluorescence response or model-assigned concentration. Those are separate biological, performance, factual, and legal propositions requiring case-specific evidence.

For foundational guidance on preventing specimen-category errors, see the independently published same-site reference on biological matrices in forensic toxicology. That context clarifies why analytical matrix and reporting scope must travel with every result.

How would a forensic laboratory distinguish a promising screen from confirmation?

Screening methods can be valuable when their limitations are documented and presumptive positives are handled correctly. Confirmatory identification requires a sufficiently specific procedure and laboratory acceptance criteria appropriate for the legal or medical question at issue. EEM/PARAFAC is an attractive exploratory technology because it collects multidimensional optical information and uses component modeling; however, its signal assignment is not identical to molecular confirmation by an independently validated mass-spectrometric identification method.

For routine evidentiary deployment, a laboratory should demonstrate real-urine matrix performance across donors and concentrations, including specificity, common fluorescent interferents, any preparation-induced conversion, extraction recovery, stability, carryover, intra- and interday precision, true decision thresholds, documented QC acceptance, and independent blinded specimens. A method must also explain why the cited multicomponent LOQs do or do not govern the concentration range displayed in a report. Actual proficiency and casework review should not rely on prepared mixtures as though they were unknown human samples.

For an independent reference framework, review the same-site confirmatory drug testing guide. A confirmatory laboratory conclusion requires both chemical identity and quality evidence for the particular analyte and specimen; subsequent conclusions about dose, timing, behavior, and causation have their own additional evidentiary burdens.

What source records matter if an expert is asked to defend this result?

An expert assessing a real EEM/PARAFAC case should seek the raw excitation–emission matrices, instrument files, wavelength and slit settings, scattering corrections, reference-standard identities, calibration solution records, component-number selection diagnostics, model files and version, blank/interference tests, reported limit calculations, extraction batch records, and all QC results. When GC–MS is claimed as corroboration, the matched specimen identification, derivatization records, instrument method, chromatograms, calibrators, acceptance criteria, and quantitative traceability matter as well.

The original biological specimen’s collection and handling are equally important. Was urine collected fresh, preserved, stored, frozen, thawed, hydrolyzed, diluted, fortified, pooled, or transferred? Which parts of the procedure are simply laboratory preparation and which are patient sample history? Was the reported number a measured instrument signal, a PARAFAC component score, a calibration prediction, or a final accepted assay concentration? These are not interchangeable entries on an expert report.

The independently published same-site specimen collection and preservation guide provides the broader collection context. Its concepts apply here: without an identifiable and properly documented source specimen, neither a sophisticated optical model nor a chromatographic comparison can independently reconstruct the person’s exposure or condition.

What should the paper not be cited as proving?

It should not be cited as proof that a light-based urine device detects all recent cannabis use, that an arbitrary small peak confirms four cannabinoids, or that the investigators established clinical-grade diagnostic accuracy in a population. The sample was four urine donors, the model was tested heavily in prepared chemical mixtures, and the original model’s published quantification limits are materially higher than many illustrated low-concentration readings. The source does not report comprehensive operational specificity, sensitivity, or case-specific likelihood ratios.

It also should not be cited to equate urine THC–COOH with parent THC in venous whole blood, estimate hours since use, calculate an individual’s THC dose, support a DUI statutory threshold, classify a witness’s reliability, or establish that cannabis caused a crash or injury. No standardized driving, reaction-time, road test, accident reconstruction, or individualized clinical assessment was performed in this primary study. These limits preserve the value of a genuinely promising analytical experiment without overstating what it establishes.

The correct bottom line is narrower and more useful: multidimensional fluorescence and PARAFAC recovered four expected signals in controlled mixtures, and limited treated-urine examples showed some agreement with GC–MS; however, the method as reported is not a proven standalone confirmatory forensic urine assay or a measure of acute human impairment.

Frequently asked questions about urine fluorescence cannabinoid testing

1. Can PARAFAC detect THC when CBN has the same fluorescence peak?

It can mathematically help distinguish the modeled signals because the study used an entire excitation–emission matrix rather than the common 255/310-nm peak alone. CBN also showed a second feature near 310/360 nm. That supports component separation in prepared mixtures; it does not by itself prove chemically specific THC identification in every unknown urine sample.

2. Did the study test urine from hundreds of cannabis users?

No. The Methods describe two frequent cannabis users and two nonconsumers who supplied urine for fortified reference testing. Prepared calibration and prediction mixtures and repeated instrument readings do not add independent human donors. Population-level false-positive and false-negative rates were not established.

3. Is a 6-ng/mL standard-solution QL the same as the THC urine-model LOQ?

No. Six ng/mL is the THC quantification limit reported for an individual methanol standard in Table 1. The four-component PARAFAC model prints a THC LOQ of 0.26 µg/mL—260 ng/mL—in Table 3. The lower number cannot simply be substituted for the complex-model limit in a forensic urine report.

4. Why did the researchers not detect THC in the cannabis users’ urine?

They attributed the lack of parent-THC detection in the two consumer specimens to low concentrations after metabolism. The paper does not establish each person’s actual time since use or identify a validated individual detection window; the negative result should not be turned into a universal biological rule or claim that the person never used cannabis.

5. Did the method avoid organic solvents?

The recorded optical readout differs from GC–MS separation, but the Methods describe n-heptane extraction and methanol reconstitution after alkaline heating, acidification, and evaporation. Characterizing the complete urine procedure as wholly solvent-free would omit these documented steps.

6. Can a urine fluorescence result be converted into a whole-blood THC level or impairment finding?

No conversion or impairment model was tested. Urine is a different matrix, the optical signal can represent parent cannabinoids or metabolites, and neither dose nor behavior follows from an isolated concentration. A legal impairment or event-causation opinion requires other appropriate evidence, not a numerical conversion invented from this paper.

7. Does agreement with GC–MS make this a validated confirmatory assay?

No. Comparisons with GC–MS in a few prepared and donated urine examples provide encouraging preliminary agreement, but a routine confirmatory method also needs independently demonstrated identity criteria, sensitivity and specificity across genuine unknown specimens, appropriate quantification limits, interference and quality controls, and documented laboratory acceptance requirements. The reported Table 3 LOQs and low plotted urine values require particular review.