Analytics Methods

Spectroscopic Analysis of Hemp Flour

As vast fields of hemp are planted across Earth, we needn’t focus strictly on secondary metabolites like cannabinoids and terpenes, especially since food products like hemp hearts or hemp seed oil have offered healthy sources of omega fatty acids and protein for a long time. Hemp flour is produced by defatting the hemp seed cake left over after hemp seed oil production.

In 2019, researchers evaluated the use of field-portable near-infrared spectroscopy (NIRS) to detect cannabinoids in hemp flours. [1] NIRS measures signal resultant from stretching vibrations of hydrogen bonds (e.g., carbon-hydrogen, oxygen-hydrogen, nitrogen-hydrogen) when NIR light irradiates the sample. The light hits the molecule, the bonds shimmy and shake, and they do so at characteristic frequencies which the instrument detects. A benefit of spectroscopy is that samples require virtually no preparation, and such was the case here as the researchers evaluated the flours directly. It’s also fast, and here, the researchers advertise a 2.5 second analysis time via NIRS.

NIR spectra often look visually similar, and thus, using NIR spectra for quantitation typically requires sophisticated software to differentiate chemistries between samples. The spectra are typically coupled with data from a primary measuring technique like chromatography, the data used in tandem to create a calibration dataset. These researchers used gas chromatography-mass spectrometry (GC-MS) to establish an accurate characterization of commercial hemp flours that typically don’t have any cannabinoids. Then, known amounts (0.001 to 0.1% w/w) of cannabinoid standards were added to the flours using a rotary evaporator. The GC-MS data was combined with NIR spectra to generate a predictive model.

Although no distinct sample size (N) was listed, ten flours were evaluated for cannabidiol, tetrahydrocannabinol, and cannabigerol (CBD, THC, and CBG, respectively), and the study’s plots seem to convey that the five spectra measured for each sample were used individually and not averaged. Short of counting every dot in a plot, the N helps identify whether all data was used or whether some points were removed as outliers. When data is removed, it’s customary to include corresponding statistical explanations as to why that action was taken. Otherwise, one can design desirable looking models on paper, but their practicality may be seriously limited.

The analyst has multiple choices available for spectral data processing which, when done incorrectly, can alter the data. Such is the case for baseline correction. The authors baseline corrected their spectra, but do not specify how. Baseline correction can be very subjective, with different analysts producing different results. [2] So, many analysts choose to bypass this subjectivity by transforming the spectra into their first or second derivatives. These authors did both.

Although 25% of the spectra were used to validate the model (known chemistries are treated as unknowns, and the predicted results are compared to known values), it’s unclear how this set was created. Ideally, the validation set would be chosen such that the model is tested across its range, and there are algorithms available to do this.

When data models are created, there are plots and graphs (e.g. loading plots, regression coefficients, etc.) that illustrate the model’s power. Unfortunately, this study didn’t provide this information or the number of factors used to explain the data — a key thing to note — because more factors mean gorgeous appearing models. But that beauty is only skin deep, since more factors means the model will attempt to predict random noise within the data that cannot be reproduced. So, while the model might work well now, once there’s a different sample set, it’s predicative power likely will be reduced.

When choosing to write about this study, I hoped to convey the power of spectroscopy, something I’ve done throughout my career. But, as the mantra often reads, we need more data. These new hemp and cannabis analytical applications must provide much more validation data and demonstrate through that data that the models are practical for everyday use.

References

  1. Risoluti R, Gullifa G, Battistini A, Materazzi S. Monitoring of cannabinoids in hemp flours by MicroNIR/Chemometrics. Talanta. 2020;211:120672. [journal impact factor = 5.339; times cited = 0 (Semantic Scholar)]
  2. Lupoi J. et al. Assessment of thermal maturity trends in Devonian–Mississippian source rocks using Raman spectroscopy: Limitations of peak-fitting method. Frontiers in Energy Research. 2017;5(1). [journal impact factor = 2.746; times cited 21 (Semantic Scholar)]

Image by chrisbeez from Pixabay

About the author

Jason S. Lupoi, Ph.D.

Jason S. Lupoi, Ph.D.

Leave a Comment