Hand-portable HPLC with broadband spectral detection enables analysis of complex polycyclic aromatic hydrocarbon mixtures

The study addresses the challenge of field-deployable analysis of polycyclic aromatic hydrocarbons (PAHs), a class of persistent organic pollutants associated with carcinogenic, mutagenic, and teratogenic effects. PAHs are ubiquitous in the environment, originating from natural and anthropogenic sources, and human exposure occurs via air, water, and food. Regulatory agencies strictly limit PAH levels, but current monitoring often relies on laboratory-based HPLC or GC-MS after off-site sample collection. Existing portable systems typically use single-wavelength UV detectors and identify analytes primarily by retention time, which is unreliable in complex matrices where co-elution and matrix effects can shift retention behavior. Mass spectrometric hyphenation and extensive sample preparation used in labs are impractical in the field. The research question is whether a hand-portable HPLC equipped with broadband UV–vis spectral detection can separate, detect, and robustly identify complex PAH mixtures in the field, independent of retention time, including resolving co-eluting and hidden peaks. The purpose is to develop and validate a portable platform that provides time-resolved full-spectrum data and chemometric methods to enable spectral fingerprinting of PAHs and to compare its performance to a conventional lab HPLC.

Established PAH analyses employ HPLC and GC-MS, each with advantages and limitations: GC-MS identifies volatile, thermally stable PAHs via mass but struggles with less volatile species; LC can handle isomers not easily quantified by GC-MS but is prone to matrix interferences. Prototype portable GC, LC, and MS systems exist, including portable GC-MS with ion traps and portable LC using LED-based or single-wavelength UV detection, but these often sacrifice analytical performance or portability. Portable LC is promising due to the lack of vacuum/gas requirements, yet current field approaches typically depend on retention time for identification, which fails in complex mixtures due to matrix effects and co-elution. In laboratory settings, matrix effects are mitigated by sample prep and LC-MS hyphenation. Prior work on fast LC (e.g., superficially porous particles, gradient separations) improves throughput but increases complexity. Chemometric approaches such as multivariate curve resolution and spectral deconvolution have been reported in conjunction with LC to address overlapping peaks. Public spectral databases and automated literature extraction tools provide a foundation for building spectral fingerprint libraries for identification.

Instrument development: A stand-alone, hand-portable HPLC (anywhereHPLC) was built integrating pumping, injection, detection, electronics, and computing in a 255 × 250 × 126 mm case, 4.2 kg with 150 mL mobile phase. A constant-pressure pump uses pre-compressed nitrogen (up to ~300 bar) with micro-valves; the pump consumes effectively zero electrical power during operation, extending battery life (>24 h on a 10 Ah battery). The mobile phase reservoir supports ~19.1 h continuous operation under conditions used. Detection: A pulsed Xenon (PX) UV–vis light source (1–220 Hz, 45 µJ/pulse) fiber-coupled to a 10 mm pathlength, 2 µL z-type flow cell, with a miniaturized spectrometer (3648-pixel array) yields full-spectrum absorbance from 180–890 nm at ~0.2 nm resolution (instrument response window 180–890 nm). Spectral emission profile was measured onboard. Source stability was assessed on mains and battery at 1 Hz, showing negligible drift (average 0.01 ± 0.03% per 10^3 s across wavelengths). Photodegradation testing exposed a 24-PAH mixture held in the flow cell to PX at 1, 2, 100, and 220 Hz (0.2–45 µJ s⁻¹) for 30 min with no significant change in integrated absorbance; typical operation has ~1 s residency at 150 µL min⁻¹. Injector: Automated six-port, two-position valve with 5 µL PEEK loop and 60 nL wetted swept volume; switching <1 s; 340 bar rating. Injector reproducibility was evaluated using a 4-component certified test mix (48270-U) with n=3 runs. Retention time RSDs: 0.55–0.93%; peak area RSDs: 1.42–2.20%, comparable to an Agilent 1260/90. Chromatography: Mobile phase 70:30 (v/v) acetonitrile:water. Columns: Zorbax Eclipse PAH (2.1 × 100 mm, 3.5 µm) optimized for PAHs and Poroshell 120 EC-C18 (2.1 × 50 mm, 2.7 µm) for faster, lower-resolution separations. Injections: Zorbax runs 5 µL at 50 ng/µL per PAH; Poroshell runs 5 µL at 5 ng/µL per PAH. Comparative runs were performed on both the portable system and Agilent 1260/90 using the same columns. Single-wavelength chromatograms at 230 ± 2 nm and full-spectrum heatmaps (180–890 nm) were recorded. Retention time reproducibility: Poroshell, portable 0.09–0.57% RSD; lab 0.07–0.74% RSD. Zorbax, portable 2.15–4.54% RSD; lab 0.98–2.84% RSD. Data acquisition and processing: Custom Python/MATLAB code on an onboard single-board computer recorded spectra at up to 220 Hz. Spectra were smoothed (Savitzky–Golay, polynomial order 5, window ±4.5 nm), digitization noise reduced, then first-derivative transformed to remove baseline offsets. Processed spectra spanned 177–730 nm (major absorption ~200–450 nm). Feature detection and spectral deconvolution: Multiwavelength chromatograms were used to resolve overlapping/hidden peaks. Gaussian Mixture Models (GMMs) were fitted across wavelengths to estimate species contributions and retention times but were complemented due to overfitting risks. Principal components analysis (PCA) and K-means clustering (elbow method) estimated the number of species in congested regions. Multivariate Curve Resolution with Alternating Regression (MCR-AR) was applied to deconvolve spectra of co-eluting/hidden peaks, initialized using GMM-derived time points (maximizing SNR and minimizing interference) and fixed retention times. Spectral fingerprinting and classification: A Linear Discriminant Analysis (LDA) classifier was trained on reference UV–vis spectra sourced from the literature (e.g., NIST WebBook) and expanded with spectra measured from purchased standards to cover species lacking online spectra. Preprocessing mirrored that used for measurement data. For non-congested peaks, spectra at maximum absorbance were used; for overlapping/hidden peaks, MCR-AR deconvolved spectra were used. A probability-based LDA output was supplemented by a tiebreak using r^2 similarity to reference spectra. The final training set comprised 88 spectra (literature + new acquisitions). Matched spectra and classification outputs were archived (GitHub: https://github.com/AnalyticalSystemsResearch/). Limits of detection (LOD): LODs were determined per PAH, yielding 2.5–55.8 ng/mL. Discussion includes potential sensitivity improvements (flow cell pathlength/waveguides, detector arrays optimized for sensitivity, lower-noise light sources) and sample preconcentration. Field testing: Water samples were collected in the field at seven sites: Wales (Nant Bwrefwr, Brecon Beacons); London (River Pinn; Grand Union Canal, Park Royal; Millwall Dock/Thames; rainwater); Cyprus (Larnaca Salt Lake). Samples were spiked with the 24-PAH mix (5 ng/µL per PAH), filtered (0.22 µm), and analyzed on-site without further preparation. Recovery rates vs HPLC-grade water and triplicate variation were computed. Basic water quality (alkalinity, pH, hardness, copper) was assessed with test strips to contextualize matrix effects.

- The portable HPLC achieved full-spectrum UV–vis detection from 180 to 890 nm and delivered 3D time-resolved spectral data in a rucksack-portable form factor operating >24 h on battery. - Injector performance: Retention time RSDs 0.55–0.93%; area RSDs 1.42–2.20% for a certified test mix; comparable to a commercial Agilent 1260/90. - Chromatographic reproducibility: Retention time RSDs comparable between portable and lab systems (Poroshell portable 0.09–0.57% vs lab 0.07–0.74%; Zorbax portable 2.15–4.54% vs lab 0.98–2.84%). Methods developed on lab systems transferred readily to the portable instrument. - Spectral detection enabled identification in congested regions and revealed less congested wavelength windows beyond 230 nm, exploiting PAH spectral diversity (active 230–300 nm, tails beyond 400 nm). - Hidden/co-eluting peaks were detected and deconvolved using PCA/K-means, GMMs, and MCR-AR. Example: A three-peak region (5.0–7.0 min) with isomers (MW 228 Da) was resolved; deconvolution improved classification of chrysene when its spectrum was otherwise convolved. Another example: Co-eluting benzo[b]fluoranthene and benzo[e]pyrene (Rt 8.69 ± 0.07 and 8.74 ± 0.08 min) were separated spectrally; r^2 similarity improved from ~0.82–0.83 to ~0.89 after deconvolution; Fréchet distances decreased. - Classification: Using LDA with literature plus newly acquired reference spectra (n=88), the system achieved 100% identification of all 24 PAHs tested, including complete coverage of the US EPA 16 and 16 PAHs from the EU “15+1” list. Initially, 18/21 were 100% with literature-only references; expansion to include measured references yielded 24/24 at test conditions. - Limits of detection ranged from 2.5 to 55.8 ng/mL per PAH. - Source stability: PX light drift averaged 0.01 ± 0.03% per 10^3 s with no significant photodegradation across 1–220 Hz irradiation tests. - Field validation: In on-site analyses of spiked environmental waters (7 sites across Wales, London, Cyprus), recovery ranged 82.16–170.36% relative to HPLC-grade water; most triplicate variations ≤5%. Despite variable matrices (salinity, turbidity, organic matter), spectral classification success was 100% for detected PAHs, and retention-time-independent identification mitigated matrix effects.

The results demonstrate that adding broadband spectral detection to a portable HPLC provides an orthogonal dimension for analyte identification that mitigates the inherent limitations of retention-time-based identification in complex field matrices. Spectral fingerprinting coupled with unsupervised chemometric analyses (PCA/K-means, GMM, MCR-AR) enables detection and classification of overlapping and hidden peaks, addressing a principal challenge in PAH separations. Performance comparisons with a laboratory HPLC show similar retention time stability and chromatographic capability, suggesting straightforward method transfer between lab and field instruments. Although single-wavelength portable detectors are common, they lack the discriminating power needed for complex mixtures; the full-spectrum approach overcomes this without requiring MS hyphenation or elaborate sample preparation, both impractical in the field. The classifier’s independence from retention time enhances robustness to matrix-induced shifts, as evidenced by 100% identification rates in diverse environmental waters. While LODs (2.5–55.8 ng/mL) are higher than some lab-based EPA HPLC methods, proposed hardware and preconcentration improvements can narrow this gap. The platform and methodology generalize beyond PAHs to other analyte classes where UV–vis spectral diversity can be leveraged, and the ability to build and update spectral libraries from literature and newly acquired references supports scalable, field-ready fingerprinting.

This work introduces the first fully stand-alone portable HPLC with broadband UV–vis spectral detection (180–890 nm) and demonstrates its ability to separate, detect, deconvolve, and identify complex PAH mixtures in the field. The system compares favorably to a commercial lab HPLC in retention time stability and overall chromatographic performance. By exploiting full-spectrum data and chemometric deconvolution, the method identifies co-eluting and hidden species and achieves 100% identification coverage of 24 tested PAHs, including the US EPA 16. Field testing on diverse water samples confirmed robust, retention-time-independent identification despite matrix effects. The study provides an extensible pipeline for real-time, onboard, unsupervised feature detection and spectral classification using libraries compiled from literature and in-house references. Future work should focus on improving sensitivity (longer effective pathlengths via optofluidic waveguides, lower-noise light sources, more sensitive detector arrays) and integrating field-amenable preconcentration to reduce LODs, as well as expanding spectral libraries and potentially incorporating retention-time priors when appropriate.

- Sensitivity: Achieved LODs (2.5–55.8 ng/mL) are higher than some EPA-reported HPLC method detection limits (0.018–2.3 ng/mL). Improvements in flow cell pathlength (with temporal resolution trade-offs), optofluidic waveguides, more sensitive detector arrays, and lower-noise light sources are needed to enhance sensitivity. - Chromatographic resolution: Not all features are baseline-resolved, especially on the faster Poroshell method; gradient elution could improve separation but would add system complexity and reduce portability. - Dependence on spectral references: Initial classifier performance was limited by sparse literature spectra for some PAHs; expanding and curating high-quality reference libraries is required for broad applicability. - Matrix effects on quantification: While identification is robust, spike recoveries varied (82–170%) in field samples, indicating matrix effects can affect quantitative accuracy without pre-treatment or matrix-matched calibration. - Hardware trade-offs: Extending optical pathlength or modifying detector components may impact temporal resolution or dynamic range; the current spectrometer prioritizes dynamic range over sensitivity. - No MS hyphenation: The platform relies on UV–vis spectral fingerprinting without mass information; although effective here, some analyte classes may benefit from mass-based confirmation not currently portable in this system.