Rapid analysis of multiple components in tobacco using the Antaris II FT-NIR Analyzer

Importance of the topic

Reliable, rapid and multiplexed chemical analysis of tobacco is critical for quality control across raw materials, processing intermediates and finished cigarette products. Traditional wet-chemistry methods are time-consuming, reagent‑intensive and require skilled personnel, creating bottlenecks for high-throughput QC. Near‑infrared Fourier transform spectroscopy (FT‑NIR) offers a nondestructive, low‑cost alternative capable of delivering multi‑component results in under a minute, enabling faster process decisions, reduced waste and lower operational costs.

Objectives and study overview

This application study evaluated the Thermo Scientific Antaris II FT‑NIR analyzer to develop and validate calibration models for 16 tobacco components. The study used several hundred naturally occurring tobacco samples (calibration sets per analyte ranged from ~469 to 785 samples) representing multiple tobacco types and production regions. Each model was tested with roughly 39–56 independent validation samples to assess predictive performance. The overall aim was to determine whether FT‑NIR can replace or complement traditional wet chemical methods for routine tobacco analysis in industrial practice.

Used instrumentation

• Thermo Scientific Antaris II FT‑NIR analyzer with integrating sphere solid sampling module.
• Spinning sample cup for diffuse reflectance measurements.
• Spectral acquisition: 10,000–3,800 cm⁻¹ at 8 cm⁻¹ resolution, 70 scans per sample (≈1 minute acquisition).
• TQ Analyst chemometric software for model development (Partial Least Squares).
• Reference analyses performed by national institutes and factory analytical centers using standard wet chemical methods.

Methodology

Tobacco leaves were milled to a homogeneous powder and scanned in reflectance mode. Calibration development prioritized representing the natural variability of tobacco by including different leaf types, origins and processing states. Spectral pretreatments included Multiplicative Scatter Correction (MSC) to counter scattering, first derivative and Norris smoothing to remove baseline offsets and enhance relevant spectral features. Partial Least Squares (PLS) regression was used for all calibrations to handle broad, overlapping NIR bands. Outlier identification employed statistical tests (e.g., Dixon or Chauvenet) before final model fitting. Model quality was quantified using correlation coefficient, RMSEC, RMSECV and RMSEP; RMSEP on independent validation sets served as the primary indicator of predictive performance. Spectral ranges and PLS factor numbers were optimized per analyte to avoid under‑ or overfitting.

Main results and discussion

Robust calibrations were obtained for 16 analytes with generally high correlation coefficients (≈0.91–0.99) and low RMSEP values for key quality parameters. Representative RMSEP values (validation sets) included: nicotine 0.170 (%), total sugars 1.17 (%), reductive sugars 0.92 (%), total nitrogen 0.0882 (%), potassium 0.186 (%), chlorine 0.0529 (%), total volatile acids 0.00530 (%), total volatile bases 0.0205 (%), sulfate 0.159 (%), starch 0.56 (%), cellulose 0.00855 (%), polyphenols 2.7 (%), total petroleum ether extracts 0.00420 (%), petroleum ether extracts (neutral) 0.00361 (%), ash 0.945 (%), and water‑soluble ash bases 0.226 (%).

Performance varied by analyte: constituents with reproducible reference data and strong NIR signatures (e.g., nicotine, total nitrogen, volatile acids/bases, petroleum ether extracts) achieved the best predictive accuracy. Components with more complex reference chemistry or weaker spectral response (e.g., polyphenols, starch, ash) displayed higher errors, likely reflecting both spectral overlap and variability in reference methods. Large calibration sets (hundreds of standards) and careful pretreatment were essential to capture matrix variability and produce transferable models.

Practical benefits and applications

• Speed: multi‑component results in under one minute per sample versus hours or days for manual wet chemistry.
• Minimal sample preparation and nondestructive measurement reduce operator time and measurement variability.
• Lower per‑sample cost and less chemical waste compared with reagent‑based assays.
• Ability to monitor incoming materials, blending, curing and other process steps with rapid feedback for quality assurance and process control.
• Fourier transform technology and standardized software facilitate model transfer between instruments and sites, supporting wider deployment.

Future trends and potential uses

Expected developments that will broaden applicability include: expansion of spectral libraries and calibration sets to extend analyte coverage and improve robustness across geographic and varietal variability; integration with process analytical technology (PAT) for at‑line or online monitoring; use of advanced machine learning methods beyond PLS to extract weak or nonlinear relationships; cloud‑based model management and calibration transfer tools; miniaturized or portable NIR instruments for field checks; and automated sample handling to increase throughput and traceability. Together, these trends support broader adoption of FT‑NIR across analytics, R&D and manufacturing environments in the tobacco sector and related agricultural industries.

Conclusion

The study demonstrates that the Antaris II FT‑NIR analyzer can reliably quantify multiple critical tobacco components with speed and economy superior to traditional wet chemical workflows. For many analytes the predictive errors are sufficiently low for routine QC use; for others, further refinement of reference methods and expanded calibration sets can improve performance. Overall, FT‑NIR represents a practical, cost‑effective tool for rapid multi‑component tobacco analysis and can materially improve laboratory throughput and process control.

References

• McClure WF. Spectral Characteristics of Tobacco in the Near Infrared Region from 0.6 to 2.6 Microns. Tobacco Science. 1968;12:232–235.