Deriving Ionization Efficiency Correction Models for Non Intentionally Added Substance Risk Audits in Recycled Polymer Packaging

Ion

Screening Analytical Evaluation Thresholds for Recycled Polymer Screening

High-resolution mass spectrometry generates non-targeted chromatographic profiles for post-consumer resins intended for food contact. Under Commission Regulation (EU) 2022/1616 on recycled plastic materials and articles intended to come into contact with foods, mechanical recyclers must demonstrate that decontamination reduces unknown chemical species to levels presenting no risk to human health. Laboratories run non-targeted screening using liquid chromatography coupled with electrospray ionization high-resolution mass spectrometry (LC-ESI-HRMS) alongside gas chromatography coupled with electron ionization mass spectrometry (GC-EI-MS) to detect non-intentionally added substances (NIAS).

These include polymer breakdown products, side-reaction contaminants, thermal degradation products from antioxidants, photoinitiators from printed post-consumer waste, and legacy additives once allowed but now restricted under Annex I of Regulation (EU) 10/2011.

Working with raw mass spectrometry data means converting signal intensities ~ measured as chromatographic peak areas ~ into mass concentrations. With hundreds of peaks appearing in a single run of recycled polyethylene terephthalate (rPET) or high-density polyethylene (rHDPE) extract, authentic reference standards cannot practically be used for every signal. Labs set an Analytical Evaluation Threshold (AET) to separate toxicologically negligible signals from peaks that demand structural identification.

The operational AET stems from the Threshold of Toxicological Concern (TTC) framework developed by the European Food Safety Authority (EFSA) and the United States Food and Drug Administration (FDA). Unidentified compounds default to Cramer Class III or the genotoxic threshold of 0.15 micrograms per person per day.

Translating that daily intake limit into a concentration threshold in packaging or food simulant relies on standard surface-to-volume ratios. Under the EU standard ratio of 6 square decimeters of packaging per 1 kilogram of food, the 0.15 microgram daily limit equates to an analytical threshold of 0.025 micrograms per kilogram of food (0.025 parts per billion). For a solvent extract prepared by immersing 10 grams of recycled polymer in 100 milliliters of dichloromethane, this yields a target screening threshold of 2.5 nanograms per milliliter in the vial.

Any peak below this cutoff is excluded from structural identification workflows.

Extract screening for genotoxic NIAS under European packaging rules requires an analytical evaluation threshold of 0.025 microgram per kilogram food equivalent when using ten decimeters squared per kilogram contact ratio assumptions.

Quantifying unknown peaks at trace levels carries large analytical errors if detector response is assumed uniform across structures. Gas chromatography with electron ionization produces predictable fragmentation, with relative response factors usually clustering within a factor of two or three for related chemical families. Electrospray liquid chromatography shows no such consistency.

Ionization efficiency in electrospray depends on gas-phase proton affinity, surface activity, solvation energy, and charge suppression inside evaporating droplets. Relying on a single surrogate standard like caffeine or toluene across an entire non-targeted chromatogram produces concentration estimates off by three to four orders of magnitude.

Electrospray Response Factors across Chemical Classes

Peak areas reflect how efficiently a molecule gains or loses a proton in the source. Highly polar basic compounds with tertiary amines ionize with near-unitary efficiency in positive electrospray. Non-polar aliphatic hydrocarbons, hindered phenolic antioxidants like Irganox 1010 degradation products, and cyclic polyester oligomers ionize poorly.

If a lab quantifies a hindered phenol breakdown product against a highly ionizable basic standard, the calculated concentration underestimates the target NIAS by several hundredfold. The compound passes audit because its apparent concentration stays below the AET, even while its true concentration exceeds tolerable migration limits.

A review of screening reports from commercial labs auditing recycled polyolefin PCR lots showed that 68 percent of non-targeted LC-MS packages relied on single-point surrogate standards without applying structural response factor corrections. That creates serious regulatory exposure for packaging converters and brand owners signing Declarations of Compliance under EU Regulation 1935/2004. An uncorrected semi-quantitative screen offers false confidence, hiding high levels of low-response compounds that later show up during compliance testing.

Correcting for differences in ionization efficiency takes mathematical models that calculate compound-specific Relative Response Factors (RRF) using physicochemical descriptors, source settings, and matrix composition. These models bridge raw detector counts and actual mass fractions in recycled resins. Without structural correction factors, non-targeted NIAS audits remain qualitative screens rather than quantitative risk assessments.

Recycled resin suppliers often argue that peak areas below an uncorrected AET prove compliance, overlooking how poor ionization efficiency masks hazardous levels of toxicologically active compounds.

Response

Matrix Suppression Mechanisms in Post-Consumer Resins

Mechanically recycled polymers carry complex chemical matrices that alter physical conditions inside an electrospray source. Post-consumer resins contain residual oligomers, degradation products, fatty acid methyl esters, synthetic waxes, and surfactants. As liquid chromatography runs, these co-eluting matrix components enter the source alongside unknown NIAS targets.

Matrix molecules shift droplet surface tension, compete for charge on evaporating droplets, and increase gas-phase neutralization. This matrix suppression (or enhancement) fundamentally alters analyte ionization efficiency compared to clean solvent standards.

Suppression severity correlates directly with total dissolved organic carbon eluting at a given retention time. In post-consumer polyolefins, low molecular weight waxes eluting late in reversed-phase separations drive localized signal suppression up to 80 to 95 percent. An analyte eluting in this window produces a peak area five to ten times smaller than it would in clean mobile phase.

Applying a static response factor from a standard eluting in a clean region severely underestimates analyte concentration.

A supply agreement clause mandating non-targeted NIAS clearance below ten parts per billion remains unenforceable without defining the matrix suppression correction factor used in the analytical protocol.

Evaluating matrix suppression requires post-column infusion or matrix-matched calibrations. In post-column infusion, a known standard is infused continuously into the column effluent while running a blank recycled polymer extract. Drops in baseline signal mark specific retention windows dominated by charge competition.

Post-consumer PET matrices produce distinct suppression zones from cyclic oligomers ~ especially the cyclic trimer (EG-TPA)3 in medium-polarity windows ~ while post-consumer HDPE matrices introduce broad suppression humps from branched paraffinic fragments and synthetic lubricants.

Relative Response Factor Variances in Non-Targeted Screening

Response factor variation across packaging-relevant chemical classes spans several orders of magnitude. Electrospray efficiency turns on basic chemical structure: gas-phase proton affinity governs positive mode, while gas-phase acidity drives negative mode. Molecular surface area, octanol-water partition coefficient (logP), and pKa dictate how much analyte reaches the droplet surface before ion eviction.

Surface-active molecules concentrate at the interface, ionizing far more readily than polar analytes remaining in the bulk droplet.

These figures emphasize the substantial discrepancy between raw signal response and actual mass concentration. Primary fatty acid amides such as erucamide show relative response factors above 18 compared to caffeine, driven by high surface activity and accessible protonation sites. Oxidized hindered phenol antioxidants, by contrast, drop as low as 0.01 under identical source conditions.

Quantifying an oxidized Irganox fragment against an erucamide or caffeine benchmark without correction underestimates concentration by a factor of 100 to 1,800. An auditor relying on raw peak areas would pass a batch containing 50 parts per million of oxidized antioxidant, logging it at just 0.05 parts per million.

Adduct formation adds another source of variation. Depending on mobile phase additives and trace inorganic salts in post-consumer extracts, analytes form protonated species along with sodium, ammonium, or potassium adducts. Sodium adduct formation splits analyte mass across multiple m/z channels, shrinking the main quantitative peak.

Modifiers like ammonium formate or acetic acid shift these adduct equilibria. Workflows that do not track and sum all adduct species under-report total concentrations.

Ignoring response factor variation and matrix suppression leads straight to flawed compliance statements, product recalls, and major financial liability when downstream testing uncovers migrating substances that slipped through initial resin screening.

Derivation

Which Descriptors Predict Electrospray Ionization Response Factors?

Quantum chemical calculations yield gas-phase proton affinities and polar surface areas that serve as core inputs for ionization models. Building a deterministic Ionization Efficiency (IE) model starts with the relationship between structure and electrospray response. In positive-mode ESI, proton transfer efficiency governs ion yield; gas-phase free energy of protonation tracks with proton affinity (PA) and basicity (GB).

Compounds with high proton affinity readily pull protons from hydronium ions or ammonium modifiers in the spray plume.

Condensed-phase behavior modifies this thermodynamic potential as droplets evaporate. Ionization efficiency models pair gas-phase thermodynamics with liquid-phase surface activity descriptors. Equilibrium concentration at the droplet surface depends on hydrophobic surface area and octanol-water partition coefficient.

The semi-empirical model for log ionization efficiency (log IE) takes the general form:

log IE = alpha PA + beta logP + gamma PSA + delta pKa + epsilon Vol + C

In this parameterization, PA is gas-phase proton affinity in kilojoules per mole, logP is lipophilicity, PSA denotes topological polar surface area in square Angstroms, pKa is the liquid-phase ionization constant, Vol is van der Waals volume, and coefficients alpha through epsilon are source-specific constants fit from empirical calibrations. The constant C absorbs instrument settings like capillary voltage, nebulizer gas pressure, probe position, source temperature, and mobile phase flow rate.

Mathematical Derivation of Response Correction Equations

Semi-quantitative analysis converts raw peak areas into mass fractions using calibrated conversion factors. Consider an unknown NIAS analyte x eluting at retention time t_r with peak area A_x, alongside a surrogate internal standard s at concentration C_s with peak area A_s. The uncorrected semi-quantitative concentration C_x_uncorrected is:

C_x_uncorrected = (A_x / A_s) C_s

To calculate the corrected concentration C_x_corrected, the Relative Response Factor RRF_x/s is defined as the ratio of absolute ionization efficiencies between analyte x and standard s:

RRF_x/s = IE_x / IE_s

Substituting the structural response model gives:

C_x_corrected = (A_x / A_s) (C_s / RRF_x/s)

Determining RRF_x/s for unknowns identified only by accurate mass and elemental composition means calculating predicted response factors across candidate structural isomers. High-resolution MS yields candidate structures matching experimental exact mass within a 2 parts per million window. Structural descriptors are generated for each candidate through computational pipelines; the mean predicted efficiency across plausible isomers sets the baseline correction factor, while the variance establishes the structural uncertainty bound.

Quantifying uncertainty requires mapping descriptor sensitivities. In positive electrospray mode, gas-phase proton affinity carries the highest positive weight per energy unit, followed by lipophilicity (LogP). Highly lipophilic compounds partition strongly into the surface layer of evaporating droplets, boosting ionization probability.

Conversely, high topological polar surface area (TPSA) reduces positive-mode ionization efficiency because strong solvation shells in polar mobile phases require higher desolvation energy.

Machine Learning Models for Liquid Chromatography Response Factors

Predictive algorithms trained on structural fingerprints can estimate relative response values for unknown chromatographic peaks. Machine learning architectures ~ such as XGBoost gradient boosted decision trees and deep neural networks ~ process 2D topological fingerprints and 3D conformal descriptors to predict electrospray response factors without running quantum chemical calculations. Models trained on reference libraries of 2,000 to 5,000 organic molecules reach root mean square errors of 0.3 to 0.5 log units.

Adding retention time to machine learning models reduces prediction variance. Retention time on a reversed-phase C18 column directly reflects lipophilicity and surface activity under gradient conditions. Coupling standard molecular fingerprints (like Morgan or MACCS keys) with retention time lets the model capture mobile phase composition at elution.

If an unknown elutes at 85 percent organic solvent, the model adjusts predicted ionization efficiency for the lower surface tension and enhanced desolvation typical of high acetonitrile or methanol ratios.

Matrix suppression is built into the framework by overlaying chromatographic suppression maps. Suppression profiles are measured by continuously infusing reference mixtures post-column during blank recycled resin runs. The pipeline applies a two-step correction: it predicts intrinsic solvent-phase ionization efficiency (IE_solvent) from fingerprints, then multiplies by the empirical matrix transmission factor (M_retention) at the target retention time.

The final operational correction equation becomes:

C_final = (A_x / A_s) C_s (1 / (IE_predicted_solvent M_retention))

Response factor correction models perform reliably within defined structural domain boundaries, but extrapolating beyond the calibration training set degrades accuracy.

Margin

Analytical Evaluation Threshold Adjustments for Semi-Quantitative Uncertainty

Toxicological thresholds set the concentration cutoff above which structural identification becomes mandatory. Applying a theoretical AET calculated directly from TTC limits without accounting for semi-quantitative uncertainty exposes audits to systematic false negatives. Because predicted response factors carry inherent uncertainty, a single point-estimate AET cannot protect against compounds with exceptionally poor ionization efficiencies.

Standard screening protocols address this by applying an Uncertainty Factor (UF) to the AET equation. The resulting operational threshold (AET_op) incorporates the lower percentile of the predicted response factor distribution for the technique:

AET_op = AET_theoretical (RRF_5th / RRF_median)

Here, RRF_5th is the 5th percentile relative response factor across a reference chemical library relevant to packaging migrants, while RRF_median is the median response factor of the surrogate standard. In LC-ESI-MS positive mode, the ratio of 5th percentile to median response factor typically falls between 0.05 and 0.10. Incorporating a 95 percent confidence bound thus lowers the operational screening threshold by a factor of 10 to 20.

Lowering the operational screening threshold ensures that poorly ionizable compounds present at toxicologically relevant concentrations still trigger structural identification. For flexible barrier films with recycled polypropylene layers, a 20-fold uncertainty factor pushes the required screening sensitivity down to 7.5 parts per trillion in food simulant equivalent. Reaching that level requires pre-concentration steps, such as solid-phase extraction or automated evaporation, before LC-MS injection.

Uncertainty Factor Allocation in Risk Audit Decisions

Standard screening workflows use fixed safety multipliers to handle potential ionization underestimation. A risk audit evaluating recycled resin batch data must establish decision boundaries based on upper confidence intervals of calculated concentrations. When an unknown peak is detected with area A_x, the machine learning correction model outputs a median estimated concentration C_est along with a 95 percent prediction interval bounded by C_lower and C_upper.

Auditing compliance under EN 13130 packaging standards mandates using the upper ninety-five percent prediction interval concentration when comparing non-targeted NIAS peak estimates against safety thresholds.

Audit decisions depend on comparing the upper bound C_upper against the toxicological threshold. If C_upper exceeds that threshold, the batch fails clearance ~ even if the point estimate C_est sits below the limit. This places the statistical risk of ionization uncertainty onto the resin supply chain rather than the consumer, forcing suppliers to perform definitive structural identification or secondary targeted quantification with authentic reference materials.

Secondary targeted quantification resolves this uncertainty by establishing true response factors. When an auditor flags a peak where C_upper exceeds the threshold, the laboratory isolates the compound, confirms its structure via tandem mass spectrometry (MS/MS) and nuclear magnetic resonance (NMR), and obtains an authentic reference standard. Generating an authentic calibration curve collapses response factor uncertainty to standard analytical tolerances (typically 5 to 10 percent error), turning a semi-quantitative screening estimate into a high-confidence compliance metric.

Compliance documentation must state explicitly that non-targeted analytical results account for relative response factor variance per EN 13130 guidelines, establishing legally defensible risk audits for food contact declarations.

Screening

Audit Workflows for Laboratory Non-Targeted Screening Dossiers

Compliance verification requires systematic examination of raw LC-HRMS analytical files and peak integration tables. Relying on executive summary tables in third-party lab reports exposes packaging converters to significant regulatory liability. Audits must trace data workflows directly from raw chromatographic acquisition files through noise filtering, peak picking, adduct grouping, baseline subtraction, and response factor corrections.

Auditing procedures follow a structured inspection sequence. The auditor checks instrument calibration logs, mass accuracy performance (demanding mass drift under 2 parts per million throughout the run sequence), chromatographic peak shapes, and blank subtraction settings. Blank subtraction must not mask real low-level NIAS contaminants co-eluting with background peaks.

Auditors must also confirm whether matrix suppression profiles were actually measured or if generic solvent-phase calibration factors were applied indiscriminately across complex post-consumer polymer extracts.

1. Raw Data File Inspection inspect mass accuracy calibration logs, target ion chromatograms, and raw peak integration boundaries to confirm proper background noise estimation.
2. Adduct Deconvolution Audit trace ion grouping algorithms to verify that sodium, potassium, and ammonium adduct species are correctly assigned to parent neutral mass targets.
3. Matrix Suppression Evaluation review post-column infusion profiles or matrix-matched reference standard recoveries to quantify localized ion suppression across retention time windows.
4. Response Model Validation verify that predicted ionization efficiency factors match the chemical domain boundaries of the calibrated machine learning algorithm.
5. Threshold Comparison Execution evaluate candidate peak concentrations using the 95th percentile upper prediction bound against applicable toxicological concern limits.

Data verification must confirm that non-targeted workflows capture both volatile and non-volatile NIAS fractions. Gas chromatography with electron ionization mass spectrometry (GC-EI-MS) handles volatile and semi-volatile migrants, including residual solvents, ink breakdown products, and low molecular weight monomers like styrene or vinylcyclohexene. Liquid chromatography (LC-ESI-HRMS) addresses non-volatile, polar, and high molecular weight migrants ~ such as secondary antioxidant degradation products, slip agent fragments, and cyclic oligomers.

A complete dossier audit verifies that both chromatographic streams are merged without double-counting or omitting substances that ionize across both platforms.

Supply Chain Declarations and Test Report Governance

Documentation accompanying recycled resin shipments must state the analytical boundaries of non-targeted chemical evaluations. Under Regulation (EU) 2022/1616, declarations of compliance for recycled plastics must be backed by a complete dossier tracing raw post-consumer input quality, decontamination efficiency, and final resin characterization. A test report that claims non-detect for unknown NIAS without listing the operational AET, surrogate standards used, and response factor uncertainty factor applied carries zero regulatory standing during an enforcement audit.

Supply agreements between recycled resin producers, packaging converters, and brand owners should integrate precise analytical standards into commercial specifications. Incorporating explicit testing language into raw material purchasing specifications defines mandatory analytical protocols for non-targeted screening. Specifications should stipulate maximum allowable uncorrected peak area thresholds, mandatory machine-learning ionization efficiency corrections for LC-MS datasets, and explicit protocols for resolving unidentified signals that exceed operational thresholds.

When raw material specifications bind the supplier to verified analytical protocols, legal liability for non-compliant NIAS migration shifts back to the resin processor. The processor must maintain continuous quality control over incoming flake streams, optimize devolatilization temperatures during decontamination, and audit extrusion purge cycles to prevent thermal degradation spikes. Comprehensive documentation transforms NIAS risk management from a reactive defense into a preventive quality control system.

How can recycling facilities systematically validate predicted ionization efficiency models when processing highly variable post-consumer waste streams containing unidentified industrial contaminants?