Overcoming Matrix Effects in Non-Target Screening: A Novel Strategy for Urban Runoff Analysis

Abstract

Matrix effects (MEs) challenge the reliability of target and non-target screening (NTS) in urban runoff analysis, particularly due to sample heterogeneity, which makes pooled samples inadequate for method development, validation, and ME corrections. This study investigated MEs in 21 urban runoff samples from different catchment areas, showing high variability in signal suppression (0–67% median suppression at 50× relative enrichment factor, REF). Runoff collected after prolonged dry periods (“dirty” samples) required enrichment below REF 50 to avoid suppression exceeding 50%. In contrast, “clean” samples had suppression below 30% even at REF 100. To correct residual MEs, a novel Individual Sample-Matched Internal Standard (IS-MIS) strategy consistently outperformed established ME correction methods, achieving <20% RSD for 80% of features through analysis of samples at three REFs as part of the analytical sequence to match features and internal standards. In contrast, internal standard matching with a pooled sample resulted in only 70% of features meeting this threshold (<20% RSD). Although IS-MIS requires additional analysis time (59% more runs for the most cost-effective strategy), it significantly improves accuracy and reliability, making it a viable and cost-effective solution for large-scale urban runoff monitoring. Furthermore, IS-MIS generates valuable data on peak reliability through measurements of signal intensities across multiple REFs, which can be used to remove “false” peaks and improve data preprocessing and method development in NTS.

Introduction

Urban runoff is a major source of chemical pollutants to freshwater, marine, and groundwater systems, originating from sources such as construction materials, vehicle traffic, urban green spaces, and wastewater. − These pollutants are often discharged directly into recipient ecosystems, either deliberately or as overflow during heavy rain events, where they pose risks to both the environment and human health. Effective and reliable monitoring of urban runoff’s chemical composition is therefore essential to inform water management and treatment strategies.

Liquid chromatography–mass spectrometry (LC-MS) is a widely used technique for analyzing the chemical composition of runoff water. Ionization is typically done with electrospray ionization (ESI), enabling the detection of a broad range of polar and semipolar compounds. While alternative ionization methods, such as atmospheric pressure chemical ionization (APCI), are also used, their application is limited by a narrower range of ionizable compounds, especially for polar compounds that are relevant in water samples. For targeted analysis, LC-ESI coupled with triple quadrupole MS (QqQ) provides highly sensitive (usually in the ppb or ppt range) detection of known analytes through selected reaction monitoring. In contrast, LC-ESI coupled with high-resolution MS instruments, such as quadrupole time-of-flight (qTOF) or Orbitrap, is better suited for suspect- and non-target screening (NTS). Their high MS sensitivity in both data-independent acquisition (DIA) and data-dependent acquisition (DDA) modes, combined with high resolving powers (10,000– 500,000) and mass accuracy, enables improved separation and identification of unknown compounds.

Despite its versatility, LC-ESI-MS faces significant challenges due to matrix effects (MEs) from coeluting matrix constituents that enhance or (more commonly for ESI) suppress analyte signals. − MEs are typically reduced by diluting samples to keep them within an acceptable range without compromising sensitivity. Remaining MEs, along with factors such as instrumental drift and variations in injection volume, are often corrected using isotopically labeled internal standards. The two approaches are combined, with target analysis relying more on internal standard correction, while dilution is favored in non-targeted analysis due to the difficulty of matching internal standards to unknown compounds. Alternative methods, such as correction based on the TIC, have also been developed but are not widely adopted.

Urban runoff variability further complicates analytical workflows. Unlike wastewater, with its predicable influent and effluent matrix differences and dilution needs, urban runoff is highly influenced by site-specific factors and precipitation dynamics. Rainfall frequency, timing, and dry periods between rain events substantially alter chemical composition due to pollutant accumulation during dry periods, even at the same location. , Accounting for this variability is important to determine appropriate relative enrichment factors (REFs), enabling an unbiased comparison of pollutant patterns and concentrations between samples with varying MEs. This also improves the concentration estimates and measurement precision.

Internal standard correction is typically done by matching internal standards with specific analytes. Isotope-labeled analogues are ideal but often limited by availability and the number of standards that can be spiked, as internal standards themselves can cause signal suppression. , This challenge is greater in suspect and NTS, where thousands of compounds are analyzed simultaneously. Internal standards are usually matched by retention time (Rt), assuming similar MEs for close-eluting compounds. However, research shows that structure-specific MEs plays an important role, ,− emphasizing the need for strategies that consider real sample behavior.

For this purpose, Boysen et al. (2018) proposed best-matched internal standard (B-MIS) normalization, where replicate injections of a pooled sample are used to optimize internal standard selection and reduce relative standard deviation (RSD). , Although effective, this strategy may introduce bias in heterogeneous samples due to unaccounted ME variability. For urban runoff samples, existing strategies for ME correction are therefore prone to biases and inaccurate results in both target and NTS analysis.

To address these limitations, we developed a novel strategy for the improved ME correction of urban runoff samples: individual sample-matched internal standard (IS-MIS) normalization. This correction strategy outperforms existing strategies for ME correction by effectively handling sample-specific MEs and instrumental drift. Our study demonstrates that IS-MIS is a robust, cost-effective solution for routine urban runoff monitoring.

Figure provides an overview of the experimental steps, starting with an initial screening to characterize urban runoff samples from a wide range of sources. Samples were analyzed in several dilutions to identify the optimal dilution and evaluate the performance of different correction strategies, which are further described in the following sections.

1.

Experimental design and data processing workflow. (1) Initial analysis of urban runoff samples used to determine ME levels. (2) Dilution workflow to evaluate strategies for mitigating ME. (3) Determination of optimal dilution, with example data showing the suppression of 6PPD-Quinone-d5 between samples at varying REFs. (4) NTS steps: (I) data preprocessing to reduce feature count; (II) feature assessment based on fit across REFs, with examples a high reliability (R ² = 1.0) and low reliability (R ² = 0.34); and (III) evaluation of ME correction strategies.

Materials and Methods

Chemicals and Standards

A standard mix (StdMix) of 104 runoff-relevant pesticides, pharmaceuticals, rubber, and industrial compounds (5–250 μg/L) was prepared in methanol, along with an internal standard mix (ISMix) of 23 isotopically labeled compounds covering a wide range of polarities and functional groups (0.04–1.9 mg/L; details in Table S1). Both mixes were stored at −18 °C. For sample dilution, the ISMix was diluted 20-fold to achieve concentrations of 2–95 μg/L.

LC-MS grade methanol and water (Honeywell Riedel-de Haen, Germany), formic acid (Fisher Chemicals), and Milli-Q water (Ultrapure, >18.2 MΩ/cm) were used for sample preparation. LC-MS solvents included LC-MS grade water with 0.1% formic acid and acetonitrile with 0.1% formic acid (Chemsolute).

Sampling and Sample Preparation

Twenty-one runoff samples were collected during summer and autumn 2023 in Santander (Spain), Odense (Denmark), and Copenhagen (Denmark). Sampling sites were selected to cover different types of catchment areas, including roofs, roads, inner-city and suburban areas, and sewer overflows (Table S2). At each site, 250 mL subsamples were collected at 0, 15, 30, 45, 60, 75, 90, and 120 min after the start of the rain events. Field blanks were prepared approximately every tenth sample using a similar procedure but with 2 L LC-MS grade water containing 300 mg/L CaCl₂ to mimic runoff ionic strength.

Samples were stored in cold, dark conditions for up to 3 days before lab processing. Subsamples were combined into composite samples, and 20 mL aliquots were transferred under magnetic stirring. Standard parameters were measured: turbidity with a pHhotoFlex Turb (WTW) system, pH and conductivity with a multimeter, and dissolved and total organic carbon (DOC and TOC) using a Shimadzu Total Carbon analyzer. Organic carbon was measured after filtration through 0.45 μm Q-max PES filters (DOC) or 5 μm PVDF filters (TOC).

For chemical analysis, composite samples were processed using a method based on Tisler et al. (2021) for wastewater samples, though ion exchange sorbents were excluded, and elution was done with methanol. Briefly, the pH was adjusted to 6.5 with formic acid, and the sample was filtered through 0.7 μm glassfiber filters (Whatman). The filtered samples were then processed with multilayer solid-phase extraction (ML-SPE) using 250 mg Supelclean ENVI-Carb columns with the addition of 550 mg 1:1 Oasis HLB and Isolute ENV+ sorbents. Elution was done with 11 mL of methanol, followed by preconcentration to REF 500 via evaporation to a final 2 mL volume using a Biotage TurboVap at 40 °C with a nitrogen flow of ∼0.5 L/min.

Instrumental Analysis and Data Acquisition

Analysis was performed with an Acquity Ultraperformance Liquid Chromatograph coupled with a Synapt G2S qTOFMS (Waters, Taastrup, Denmark). Separation was performed on a BEH C18 column (100 × 2.1 mm, 1.7 μm; Waters) with gradient elution at a flow rate of 0.3 mL/min. The gradient started at 1% B, held for 1 min, increased to 30% B after 3 min, further to 99% B at 16 min, and maintained until 21 min before returning to initial conditions (1% B) with a 5 min equilibration. The MS was operated in MS^E mode (DIA, all-ion fragmentation), alternating between low energy (MS1) scan and high energy (MS2) scan using a collision energy ramp of 10–40 eV with a scan time of 0.35 s from 50 to 1200 Da. ESI was used with capillary voltages of 1 kV in the ESI+ mode and 2.5 kV in the ESI– mode. Triplicate injections of 2 μL were performed in both ionization modes. To evaluate the system performance, a quality control (QC) sample, prepared by combining equal amounts of all runoff extracts, was injected for every eight injections throughout the analytical sequences.

Peak integration of target analytes and internal standards was performed using TargetLynx (Waters, version 4.1) with integration parameters optimized for each analyte based on peak shape and m/z. A mass window of 10–20 mDa and a Rt window of 0.2 min were used, and peak areas were manually inspected to ensure accurate integration.

Feature detection, extraction, and filtering for NTS analysis was done using MSDial version 4.92. In both ESI+ and ESI–, the minimum peak height was set to 2000, with an MS/MS cutoff of 200. Accurate mass tolerances were set to 0.01 Da for MS1 and 0.05 Da for MS2. Only features with peak areas exceeding ten times the maximum peak area in methanol spiked with internal standards (neat blank) were kept for further data preprocessing to exclude contaminants in solvent and ISMix.

Initial Screening of Runoff Samples

The initial screening of runoff samples was done as part of the D4Runoff project (https://d4runoff.eu/). Samples were spiked with internal standards and analyzed at REF 50, except for two combined sewer overflow samples (OD7 and SA5), which were diluted to REF 25 because of suspected high MEs due to the wastewater content. Quantification of MEs was performed by comparing internal standard peak areas in the samples to those in field blanks.

Dilution Experiment

To further investigate MEs and internal standard matching, six samples were selected based on their suppression levels and typical urban runoff sources (roads, roofs, inner-city and suburban areas, and combined sewer overflow). A pooled sample (Runoff_pooled), prepared by combining 50 μL extracts from all 21 runoff samples, and neat blanks were also included for analysis.

Extracts of the selected samples were spiked with StdMix and diluted to a REF of 250. Spiking ensured analyte presence at low concentrations (0.5–20 μg/L at REF 50) to minimize signal suppression introduced from spiking itself and increase the possibility of remaining within the linear range. Analytes were spiked to test IS-matching of targeted analytes, which is beyond the scope of this study, focusing on correction in NTS.

REF 250 samples were further spiked with ISMix (final concentration 2–95 μg/L) and diluted in 20× diluted ISMix solution (same concentrations as above) to prepare sample extracts at REF 100, 50, 25, 12.5, 6.25, and 3.125. These REF values were chosen based on the initial screening, which suggested that REF values above 100 caused excessive signal suppression (see Figure ).

2.

Boxplots showing the signal suppression of internal standards in runoff samples. Top and bottom edges indicate 25th and 75th percentiles, respectively, whiskers show the range of values not considered outliers, and outliers (observations with values more than 1.5× the interquartile range away from box edge) are marked with red +. Boxes indicate high-suppression samples (red) and low-suppression samples (green). Samples were analyzed at REF 50, except OD7 and SA5, which were analyzed at REF 25 (marked with *).

NTS Workflow for ME Correction

For NTS, dilutions to REF values of 12.5, 25, 50, and 100 were used, covering typical enrichment ranges. Feature peak areas were determined using MSDial, except for LOESS (LOcal regrESSion) normalization, where peak heights were used. For all internal standards, peak areas were scaled to their average value in neat blank samples analyzed in the same sequence. This step, while not strictly necessary for this study, is important for transferability between batches, since it effectively transforms internal standard signals to a measure of signal suppression rather than absolute signals and thereby removes variation caused by interbatch differences in signal intensities.

Features detected in fewer than 20% of the injections were excluded to reduce noise and focus on features detected in multiple samples. Features present in fewer than three of four REF dilutions in individual samples were also discarded.

Statistical Evaluation of ME Correction Approaches

Feature signal intensities (peak heights or areas) were normalized to their respective REF values. To distinguish MEs from random variation, the coefficient of determination (R ²) was calculated by regressing normalized feature peak areas against log-transformed REF values using MatLab (version 2024a). Panel 4 of Table illustrates this step with two example features: one with R ² = 1 (high peak reliability) and another with R ² = 0.34 (low reliability). In-sample RSD was used to evaluate correction strategies, measuring variation across REF values after correction. This quantified the effectiveness of each correction strategy in mitigating MEs, instrumental drift, and other factors affecting peak intensities. For comparison, RSDs were also calculated without correction.

1. Summary of the Data Correction Strategies .

Correction Strategy	Description	% Extra Runs
1. LOESS normalization	Drift correction using LOESS in MSDial, based on QC samples analyzed every eight injections.	0% (baseline; 1 QC per 8 injections)
2. Rt-based internal standard matching	Internal standard assigned to non-target features based on the closest Rt to capture chromatographic coelution patterns.	0% (no additional runs required)
3. Pooled sample-based internal standard matching	Internal standard assigned to non-target features based on the lowest RSD observed in pooled runoff samples.	15 extra injections (3 injections at five different REFs)
4a. IS-MIS with four REFs	Internal standard selected based on the lowest RSD with all replicates across REF 12.5–100.	265% (3 injections at 3 REFs per sample)
4b. IS-MIS with two replicates	Internal standard assigned using two replicates from each REF, with the remaining replicates reserved for validation.	176% (2 injections at 3 REFs per sample)
4c. IS-MIS with two REFs	Internal standard matching performed with REF 50 triplicates and additional samples at REF 25 and 100, using one replicate per REF; remaining injections used for validation.	59% (1 injection at 2 REFs per sample)

^aThis table outlines the data correction strategies compared in this study. Strategy 1 applies external drift correction without sample dilutions. Strategy 2 selects internal standards based on the closest Rt, while strategy 3 uses the internal standard that minimizes the Runoff_pooled feature RSD. Strategies 4a, 4b, and 4c are variations of the IS-MIS strategy, each requiring different percentages of additional runs beyond triplicate sample analysis.

Four correction strategies were tested (Table ). LOESS normalization for external drift correction based on QC samples was done in MSDial (strategy 1). The remaining strategies were used to match analytes and internal standards based on Rt (2) or RSD using either pooled samples (3) or IS-MIS (4a–4c). Three IS-MIS strategies, varying by the percentage of additional runs required (265%, 176%, and 59%), were compared. The extra runs were calculated as the additional requirements to triplicate injections of samples and one QC sample per 8 injections, but not counting additional runs such as calibration standards, blanks, and other samples typically included in NTS. To assess precision, five resamplings were performed for IS-MIS strategies 4b and 4c with different combinations of injections for matching and validation of the obtained correction results.

Results and Discussion

Factors Affecting Signal Suppression in Runoff Samples

The MEs of internal standards at REF 50 (REF 25 for OD7 and SA5) varied substantially between runoff samples, as illustrated in Figure . Notably, three samples, namely, OD1, CO1, and CO2 (highlighted with a red box), had significantly higher signal suppression compared to the others, with median suppression levels of 60%. In contrast, seven samples (highlighted with a green box) showed minimal suppression with median values below 10%. Similarly, there were significant differences between internal standards with varying degrees and ranges of MEs for different compounds (S6).

Urban runoff samples show significant variability in the MEs between sites and rainfall events. Persistent site-specific differences could not be assessed because each site was sampled only once; however, the observed variability appears largely independent of catchment characteristics. For example, OD1 and OD2, collected from similar residential areas, showed median suppression levels of 67% and 8%, respectively, suggesting that factors beyond the catchment type, such as environmental conditions and rainfall patterns, play a key role.

To date, no studies have directly explored the relationship between catchment area and rain events and ME. However, previous research has reported high variability in pollutant concentrations at the same site. , A meta-analysis by Mutzner et al. (2022) estimated that a median of six rain events is needed to estimate site-specific pollutant concentrations within an acceptable error range. This indicates that ME variability is driven by both inter- and intrasite conditions.

A key factor affecting MEs in our study was the antecedent dry period (ADP) before sample collection. As shown in Figure , the samples with the highest MEs (CO1, CO2, and OD1) were all collected after prolonged dry periods. The correlation between ADP and ME (R ² = 0.93) is consistent with previous research that identified ADP as a major factor in pollutant buildup on urban surfaces and in runoff pollutant concentrations. −

3.

Median suppression of internal standards plotted against antecedent dry period (ADP) for all 21 runoff samples analyzed during the initial screening.

Pochodyła-Ducka et al. (2023) reported a significant relationship between ADP and heavy metal concentrations in stormwater (p < 0.01 for Ni, Zn, Cr, and Fe during winter and <0.05 for Pb and Ni during summer). They attributed this to pollutant accumulation during dry periods and subsequent wash-off during rainfall. This phenomenon is also emphasized in hydrological pollution transport models.

The correlation between ADP and ME was supported by a similar relationship between ME and DOC levels (R ² = 0.87; see S3). This suggests that prolonged dry periods lead to accumulation of organic matter, which contributes to high ME, though additional data are required to accurately determine the correlation. In contrast, no similar correlations were observed with turbidity (R ² = 0.26) or conductivity (R ² = 0.02) (see S4 and S5). A possible explanation is that factors such as road salting and suspended particulates influence these parameters without directly impacting the ME, especially after sample preparation with SPE. For turbidity, aggregation, and settling of particles between sampling and measurement might also have affected the results.

These results suggest that ADP can be a practical indicator for selecting the appropriate REF during sample analysis, providing a simple and efficient method to predict MEs in heterogeneous runoff samples. Adjusting REF selection based on the expected matrix composition from ADP could reduce the need for reanalysis due to inappropriate dilution. However, this approach is not suitable for samples with high wastewater content, such as combined sewer overflows, where ADP has little impact on pollutant levels.

Optimizing Dilution for Urban Runoff Analysis

The effect of REF on MEs was assessed through a dilution series. As shown in Figure , the median signal suppression increased significantly at higher REF values, particularly beyond REF 6. The relationship was an approximately linear pattern on a log-transformed scale, indicating that doubling the REF nearly doubled the ME. This effect was evident across all samples but was most pronounced in high-suppression samples OD1 and CO1, where median suppression at REF 50 reached 60% (OD1, standard deviation = 19%) and 55% (CO1, standard deviation = 19%).

4.

Median signal suppression (%) of internal standards at various REF values observed in dilution experiments. The x-axis displays log-transformed REF values for improved readability. The data show that suppression increases with REF, particularly beyond REF 6. High-suppression samples (OD1 and CO1) show steep increases, reaching around 50% suppression at REF 50, while low-suppression samples (OD2, OD7, and CO3) and the pooled sample maintain lower suppression levels between 20% and 30%.

In contrast, low-suppression samples (OD2, OD7, and CO3) and the pooled sample showed lower suppression levels, ranging between 20 and 30% median suppression at REF 50. The observed variation in MEs highlights the need to tailor the REF selection to the individual sample matrix composition. High-matrix samples require lower REF values to minimize suppression and maintain analytical accuracy, while low-matrix samples can be analyzed at higher REF values without compromising data reliability.

Pollutant analysis in water samples uses a wide range of REF values. Tisler et al. (2021) achieved optimal REFs at 10 and 50 for non-target analysis of wastewater influent and effluent, respectively, with a 2 μL injection volume, keeping MEs below 20% for most analytes. In contrast, Backe and Field (2012) used a much higher REF equivalent to REF 450 (with 2 μL injection) to analyze selected analytes using large volume injection and different SPE sorbents. This resulted in signal suppression exceeding 20% for some compounds, though PFAS showed a minimal ME.

Our results suggest that REF values over 100 are likely to induce considerable MEs, even in relatively “clean” urban runoff samples. Since MEs are highly dependent on the specific instrument, sample preparation, and analytical conditions, the choice of methods will have a significant impact on amount of ME. Particularly, ML-SPE has previously been shown to extract a wider range of compounds compared to single sorbent methods, which could lead to an increase in the observed ME. This exemplifies a general dilemma in NTS between broad chemical analysis and robustness, which increases the need to carefully assess suitable REFs for comprehensive NTS workflows.

Evaluating Internal Standard Matching for NTS Features

The cumulative proportion of features with an RSD below a given value is shown in Figure . Internal standard correction outperformed LOESS normalization, which provided only minor improvement over no correction. This result is as expected, as LOESS corrects for instrument drift using QC samples but does not address sample-specific ME.

5.

Cumulative proportion of features (y-axis) with RSD values (x-axis, %) for the four overall correction strategies, with RSD calculated for each feature across four sample dilutions (REF 12.5–100). IS-MIS (all injections) achieves the lowest RSDs, outperforming pooled sample-matched and LOESS corrections, with the latter providing limited improvement over no correction. The shaded area for IS-MIS with two REFs and two replicated indicates the range between minimum and maximum RSD values across five resamplings. The vertical dashed line at 20% RSD marks a common reliability threshold.

Kilpinen et al. (unpublished work) reported similar findings, but with less improvement. Their study demonstrated a median RSD reduction from 19% to 12% with internal standard correction. In contrast, our results show a more substantial decrease from 33% to 6% when comparing LOESS to the best-performing IS-MIS (14% with pooled sample-matching). The greater improvement in our study is likely due to the wider REF range used (REF 12.5 to 100), compared to their narrower range (REF 5 and REF 50), which may have mitigated MEs in their samples.

All three IS-MIS strategies outperformed pooled sample-matched and Rt-matched internal standard correction. Rt-matched correction gave the worst performance with only 54% of features below 20% RSD, whereas pooled sample-matched correction improved this to 58%. In contrast, IS-MIS is tailored to the unique matrix composition of each sample, leading to more accurate corrections. This resulted in 90% of features below 20% RSD when using all four REFs (strategy 4a in Table ). However, this strategy requires all replicates, potentially overfitting the model and tripling the number of runs.

The other IS-MIS strategies used a subset of injections for matching and the rest for validation, avoiding overfitting. These strategies achieved similar performance (80% of features <20% RSD), while reducing the number of extra runs considerably, down to 59% when using two REFs (strategy 4c in Table ).

These findings demonstrate that IS-MIS effectively corrects for MEs while minimizing the need for additional work. Furthermore, the consistency across resamplings (Figure ) demonstrates the robustness of the IS-MIS strategy for determining the best internal standards for non-target features, making it a viable solution for overcoming MEs in challenging matrixes such as urban runoff. For practical use of IS-MIS, the injection volume can be varied instead of REFs to reduce the amount of work and avoid issues with insufficient extract volume for preparing multiple dilutions. This requires a simple adjustment to account for changes in internal standard concentrations, which remained constant across the REFs in this study.

Peak Reliability Information

To validate our findings, it was important to determine if variation between REF values was due to random error or systematic trends. For this purpose, R ² values were calculated for each feature against the log-transformed REFs (Figure ). Most features showed high R ² values (median = 0.98), indicated by the clustering of points on the right-hand side of Figure . This suggests that variation across REFs follows an exponential trend, consistent with the doubling of suppression at twice the REF, as discussed in the section Optimizing Dilution for Urban Runoff Analysis. The variation is largely driven by the ME, although instrumental drift and other factors also contribute, as shown by the slight improvement seen with LOESS normalization (Figure ).

6.

Scatter plot showing R ² values (x-axis) vs RSD (y-axis) of REF-normalized signal intensities for NTS features across four sample dilutions (REF 12.5–100), with each point representing a feature detected in at least three out of four REFs for a given sample. R ² indicates peak reliability, with high values (>0.9) indicating consistent changes in signal intensity across sample dilutions, and RSD shows variation across dilutions, which can either be from random uncertainty or an indication of ME for reliable features.

Figure reveals an additional benefit of IS-MIS, as analysis of multiple dilutions provides valuable data for distinguishing between reliable peaks and unreliable peaks based on their resemblance with expected compound behavior. , Reliable peaks are here defined as peaks showing increased intensity with higher REF values, allowing R ² and slope to serve as end points in method development, feature reduction, and prioritization in NTS.

Eliasson et al. (2012) demonstrated the usefulness of peak reliability data by maximizing the difference between reliable peaks (those with R ² > 0.9 for the dilution series) and unreliable peaks (R ² < 0.05) during method optimization using pooled plant tissue samples. For heterogeneous matrices like urban runoff, however, this strategy can provide critical sample-specific information on features not captured by pooled data. Peter et al. (2019) used multiple dilutions to urban runoff samples, reducing non-target features by ∼25–40% using thresholds of R ² ≥ 0.8 and slope ≥ 0.3 (log-transformed peak areas). This demonstrates how IS-MIS data, obtained through multiple sample dilutions, can be used to enhance feature selection and interpretation while also improving ME correction.

Conclusion

Urban runoff samples show major variability in MEs between sites and rain events, with median signal suppression ranging from 0% to 60% at REF 50. These variations show the importance of adjusting analytical workflows to account for the unique characteristics of each sample to ensure robust and accurate analysis with both target and NTS.

Our results demonstrate that IS-MIS offers a practical and scalable solution for addressing MEs in heterogeneous samples, such as urban runoff. IS-MIS achieved superior correction of non-target features, with 80% of features showing RSDs below 20%. This performance exceeds that of alternative correction strategies by addressing sample-specific matrix variability. Although IS-MIS requires additional analysis time (59% extra runs with two REFs), the improved data quality justifies this effort, particularly in large-scale monitoring programs that depend on reliable data with minimal reanalysis.

In addition to correcting the ME, IS-MIS provides key data on peak reliability through multiple sample dilutions, improving the reliability of feature selection. This improves the reliability of NTS for routine monitoring, specifically for large sample sets, where variable matrix compositions can otherwise hinder accurate analysis. While the approach was developed for urban runoff, it can also be implemented for other sample types, including wastewater and other environmental matrices, as well complex biological samples such as urine and blood samples. Future studies will be relevant for testing the approach across different sample matrices

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.analchem.5c01303.

• List of chemical standards, information about urban runoff samples, correlation between signal suppression and dissolved organic carbon, correlation between signal suppression and turbidity, correlation between signal suppression and conductivity, and signal suppression for individual internal standards (PDF)

The authors declare no competing financial interest.