the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 21 Aug 2026
Molecular fingerprints reveal traffic PM2.5 complexity beyond a controlled gasoline tailpipe molecular profile
Abstract. Controlled gasoline tailpipe profiles may not fully represent traffic particulate matter with aerodynamic diameter ≤2.5 μm (PM2.5) under real road conditions. The molecular compositions of tunnel and controlled tailpipe PM2.5 were compared using high-resolution mass spectrometry. Formulas common to both profiles were concentrated at low m/z and low to intermediate carbon numbers and were dominated by CHO and CHON. In contrast, molecular formulas unique to the tunnel showed greater formula richness and had higher proportions of CHOS and CHONS formulas than those unique to the controlled tailpipe profile. Tailpipe coverage ratios were high for CHO and CHON formulas in both ionization modes (0.77–0.82 and 0.84–0.92, respectively) but much lower for the combined CHOS and CHONS class (0.10–0.15). Hierarchical clustering resolved a dominant shared CHO/CHON baseline and two smaller domains concentrated in tunnel samples, characterized respectively by organosulfur formulas and by candidate formulas associated with reported tire wear chemicals together with a low molecular mass CHN response in ESI+. Thus, the controlled gasoline tailpipe profile captured the dominant shared CHO/CHON molecular baseline but incompletely represented sulfur-rich and other chemically distinct molecular domains retained in tunnel aerosol. Comprehensive molecular characterization of traffic PM2.5 therefore requires controlled tailpipe measurements together with molecular profiles representing non-tailpipe vehicle materials and near-road processing.
- Preprint
(1201 KB) - Metadata XML
-
Supplement
(965 KB) - BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-4948', Anonymous Referee #1, 20 Sep 2026
-
RC2: 'Comment on egusphere-2026-4948', Anonymous Referee #2, 21 Sep 2026
Review of “Molecular fingerprints reveal traffic PM2.5 complexity beyond a controlled gasoline tailpipe molecular profile” by Li et al.
The authors presented a very comprehensive characterization of molecular composition in aerosol particles from controlled tailpipe and tunnel samples, using high-resolution mass spectrometry. The key finding is that tunnel samples contained more complex organics, especially sulfur-containing ones, as compared to tailpipe samples. Multiple parameters were used to demonstrate the differences, and interpretations and implications were also given. The study was well designed and manuscript clearly written. I therefore recommend Minor Revision with comments as below.
- One main drawback, as the authors noted somewhere in the manuscript, of this study is that the electrospray ionization (ESI) method is preferable for more polar components, while it might miss some non-polar compounds. Please make a note or two on this limitation somewhere in the Summary and implication section.
- For the ESI+ and ESI- modes, there might be some overlapping components. Please specify how they were treated in counting the number of compounds, as well as estimation of concentration (or intensity).
- There are a few places where the authors used many parallel/possible sources to describe certain compound classes, e.g., in L225. These are not very informative. Please use more specific possible sources, backed by evidence.
- L229: how is the NOx condition affecting the “tunnel processing” of organic components? More N-containing species? In fact, this term of “tunnel processing” is not very clear. Is it prompt oxidation? Or other physical or chemical processes? Please specify.
- It would be good to have more discussion on where do those sulfur-containing species come from in tunnel samples, and how will they affect aerosol properties such as volatility, hygroscopicity, toxicity etc.
Citation: https://doi.org/10.5194/egusphere-2026-4948-RC2 -
RC3: 'Comment on egusphere-2026-4948', Anonymous Referee #3, 29 Sep 2026
This study compares 28 tunnel PM2.5 samples with exhaust collected from one gasoline passenger car at idle, 20, and 40 km h−1 using UHPLC–ESI–HRMS. The formula-level contrast is potentially useful: CHO and CHON signals overlap substantially, whereas sulfur-containing signals overlap much less. The manuscript also makes welcome efforts to state that the car is a single-vehicle benchmark, to label tire-wear assignments as candidates, and to provide clustering sensitivity analyses.
My principal concern is the step from a detected molecular formula to a source or process. Identical formulas can represent different compounds, ESI peak areas are not particle mass fractions, and the tunnel and dynamometer samples differ in vehicle mix, collection, and probably matrix and detection opportunity. These limitations do not erase the measured contrast, but they constrain what its coverage ratios and clusters can demonstrate. The paper needs a clearer analytical audit, a reproducible statistical workflow, and narrower source-specific conclusions before it is suitable for publication.
Major comments
1. Define the comparison and its scope
Methods 2.1–2.2, lines 70–90; Introduction, lines 55–66. The three dynamometer filters are three operating conditions of one car, not independent vehicles or replicated runs. The tunnel filters include a mixed fleet (approximately 80% gasoline by the authors' count) and 28 collection periods. Please report the test vehicle's model/year, engine and injection system, emission standard, particulate filter and aftertreatment, odometer, lubricant, sampling date, duration at each speed, exhaust dilution ratio, collection volume and filter load. Clarify the tunnel filter size and the stated 4.3 cm² extraction aliquot. Provide the fleet composition and relevant ventilation or background context for the tunnel periods. These details are needed to separate a difference from this particular car and protocol from a general property of gasoline exhaust. Recast claims about 'tailpipe coverage' and traffic-wide representativeness accordingly. If no further vehicles are available, this should remain a case study with a well-defined comparator.
2. Reframe the tailpipe coverage ratio
Section 3.3 and Fig. 4, lines 260–285. TCC means that a formula occurred in at least one tunnel filter and at least one of the three dynamometer filters. TCR is the pooled tunnel ESI peak area of these formulas divided by the pooled area of TCC plus tunnel-only formulas, within a mode and elemental class. It is therefore a signal-overlap statistic, not a tailpipe-derived fraction of PM2.5, organic carbon, or even an identified-compound abundance. A CHO or CHON formula shared across samples need not be the same isomer or come from the same source. Likewise, a low sulfur TCR (0.10–0.15 combined) does not establish that the missing sulfur came from tire wear or in-tunnel chemistry; other fleet exhaust, fuel and lubricants, background aerosol, and resuspension remain possible. Please rename or explicitly qualify the metric in the abstract, figures, and conclusions. Show the distribution of sample-wise TCRs with uncertainty that respects day-level grouping, together with sensitivity to detection frequency, signal threshold, and each of the three CD conditions.
3. Audit the apparent tunnel-only richness
Methods 2.1–2.4 and Fig. 4a. There are 28 tunnel filters but only three CD filters, with different samplers and likely different particulate loads. The tunnel-only counts (1,108 ESI− and 2,189 ESI+) are sensitive to sampling effort, detection limits, extraction recovery, and matrix effects. Three tunnel field blanks are described, but CD field blanks, procedural blanks, analytical replicates, pooled quality-control injections, recovery checks, and run-order or batch-drift controls are not reported. Please document the blank-subtraction rule and non-detect definition for both sample sets, the time and conditions of storage of the 2019 tunnel filters, and the MZmine thresholds and adduct/isotope handling. Compare richness after matched sampling effort and a common signal threshold (for example, accumulation or rarefaction curves and a load-matched subset). Show whether the main sulfur and coverage contrasts persist after this audit. A neutral formula count must not be presented as a count of identified molecules.
4. Limit tire-wear attribution to what the measurements identify
Section 3.3–3.4, Fig. 5, and Supplement Text S1 and Table S7. The 62 'Primary TWP' and 24 'Secondary TWP' candidates are selected from elemental constraints, correlations with reference-formula time series, and CH2 Kendrick coherence. Neither a reference formula nor a homologous series establishes a compound identity or a tire source; common traffic timing can generate the correlations. C3 contains 10 candidate Primary TWP assignments among 50, while C4H11N and C4H10N2 alone account for 41.29% and 21.69% of its ESI+ peak area in Table S7. These low-mass N formulas are not source-specific. The reported one-feature removal only tests C4H11N. Several high-ranked 'Primary TWP' candidates in C3 are CHO formulas, which underscores the non-specificity of the label. Please provide chromatographic retention and MS/MS or authentic-standard evidence for a small set of influential markers, ideally alongside tire, lubricant, road dust, and exhaust reference materials. If these are unavailable, label the categories as formula associations throughout, remove any implied primary/secondary transformation assignment, and avoid treating C3 as evidence of a tire-wear source. Repeat the clustering and peak-area summaries after excluding both dominant low-mass N assignments.
5. Resolve the sulfur chemistry before inferring mechanism
Sections 3.2–3.4, especially lines 199–205 and 344–355. CHOS and CHONS formulas do not by themselves establish organosulfate functional groups, sulfur-bearing lubricants, a tire source, or secondary formation. The low overlap and the C2 enrichment support a descriptive statement about sulfur-containing formulas in these extracts. Please report which specific high-signal C2 features drive this result and examine their chromatograms, mass errors, isotope patterns, and diagnostic MS/MS fragments where available. Relate them to measured inorganic sulfate, EC/OC, and tunnel operating conditions with uncertainty and appropriate day-level treatment. An independent source test, such as matched lubricant/tire/road-dust spectra or sulfur isotope composition of suitable separated fractions, would strengthen attribution, but the present conclusions should not depend on an unmeasured pathway. Distinguish oxidation, partitioning, and true secondary synthesis rather than grouping all differences as 'near-road processing'.
6. Correct and stabilize the statistical workflow
Methods 2.5; Supplement Texts S2–S3, Tables S4, S5, and S9. Welch tests compare 28 tunnel samples with only three non-independent operating conditions from one vehicle. Within-mode relative areas are compositional; a change in one abundant feature changes other relative areas without a change in their absolute loading. Zero replacement and the fixed 10⁻⁹ pseudocount can drive large fold changes at low signal. More immediately, Text S2 says fold changes use arithmetic group means, while Eq. S2-1 displays medians and even repeats 'log2 FC' on the right-hand side; Fig. S5 says all 5,666 assignments were tested, while the text says testing was restricted to assignments detected in at least 10 tunnel samples. Please reconcile these statements, specify the family used for each FDR adjustment, and publish the feature-by-sample matrix and executable analysis. Recompute the contrast under a consistent fold-change definition, plausible pseudocounts, detection thresholds, and alternative normalizations. Treat the CD condition comparison descriptively unless its uncertainty can be estimated from independent runs or vehicles.
7. Show what clustering adds beyond the selection step
Section 3.4 and Supplement Text S3, Tables S6–S11. HCA uses 709 features already selected for a strong tunnel/CD contrast, so the resulting tunnel-enriched domains are not independent confirmation of that contrast. The seven daily values are smoothed over three days, which can increase similarity among features responding to shared day effects. The reported 82–84% mapped agreement under descriptor rescaling is dominated by the very large C1 group; adjusted Rand indices fall to 0.463–0.499 (Table S9). Report C2 and C3 membership overlap separately under rescaling, removal of smoothing, and day-block resampling, along with uncertainty in the day-level profiles. State the domain's magnitude as well as its formula count: from Table S11, C1 accounts for about 98.2% of the mean normalized ESI+ signal and 95.4% of ESI− signal within the selected HCA assignments (these are neither mass nor all detected ions). This context is essential when discussing the smaller domains.
Minor Comments:
Section 3.1, lines 185–195: 'emergent' and 'attenuated' use ratios above 10⁴ or below 10⁻⁴ with a nominal area of 1 for nondetections, but each speed has one filter. Describe these as detection/response contrasts and provide raw area and blank distributions for the influential features; avoid implying repeatable speed effects.
Section 3.2, lines 215–240: requiring detection in all 28 tunnel samples makes the 'persistent core' especially dependent on abundance and detection threshold. Show sensitivity at 25/28 or a similar prevalence threshold. A non-significant ANOVA of CoreFrac is not evidence that its contribution is invariant.
Supplement Table S2 and manuscript lines 230–242: rainy, cloudy, and sunny each correspond to one day with four within-day filters. The text acknowledges this; keep day-specific language everywhere and avoid weather-effect inference from these ANOVAs.
Figure 6 and Table S7: explain the different peak-area denominators for class, domain, mode, and all-assignment comparisons directly in the legend. Include the unscaled and normalized values for the influential features in a machine-readable table.
Citation: https://doi.org/10.5194/egusphere-2026-4948-RC3
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 141 | 60 | 86 | 287 | 48 | 79 | 75 |
- HTML: 141
- PDF: 60
- XML: 86
- Total: 287
- Supplement: 48
- BibTeX: 79
- EndNote: 75
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This study utilizes ultra-high-performance liquid chromatography-high-resolution mass spectrometry (UHPLC-HRMS) to compare the molecular composition of PM2.5 in a tunnel environment with that of exhaust from a single controlled gasoline vehicle on a chassis dynamometer (CD). The results indicate that while the controlled tailpipe emissions can represent the dominant CHO/CHON molecular baseline in tunnel aerosols, they fail to comprehensively cover the sulfur-rich molecular communities (CHOS/CHONS) retained under real-world road conditions, as well as specific components associated with tire wear particles (TWP) and low-molecular-weight CHN. Overall, the study provides rich data and employs methods such as hierarchical cluster analysis (HCA), volcano plots, and Kendrick mass defect (KMD) network analysis to deeply mine the HRMS data, presenting clear arguments. However, the manuscript requires further improvement regarding the representativeness of the experimental design, the rationality of the statistical analysis, and the limitations of the qualitative results. Specific comments are as follows:
Major comments:
1. Lines 70–90 in the main text indicate that the exhaust samples were derived from three operating conditions of a single gasoline vehicle, whereas the tunnel samples represent the combined effects of a mixed fleet and the tunnel environment. Even though approximately 80% of the vehicles in the tunnel were gasoline-powered, it cannot be assumed that this single test vehicle represents the actual emission profile of the gasoline fleet. Therefore, the molecular formulas "detected in the tunnel but not in the test vehicle" could originate from non-tailpipe emissions or environmental transformations, but they could equally originate from the tailpipe emissions of other vehicles. The authors are suggested to clearly define the scope of conclusions that can be supported by this single-vehicle exhaust control.
2. The tunnel samples were collected using a high-volume sampler at approximately 1.13 m³/min for 3 hours per sample, whereas the exhaust samples were collected using three parallel MiniVol samplers at 5 L/min. Differences in the total air volume, particle loading, filter size, and matrix between the two sampling systems could reach orders of magnitude. Furthermore, Lines 91–94 state that "a quarter of each quartz fiber filter (4.3 cm²)" was used for extraction. This area aligns with the 47 mm exhaust filter, but it is unclear if it corresponds to the high-volume tunnel filter. If the mass of particles entering the extraction and injection processes differs significantly, the greater number of T-O formulas and higher "formula richness" observed in the tunnel samples might partially reflect differences in detection limits rather than true chemical diversity.
3. The controlled exhaust molecular profile in this study was derived solely from chassis dynamometer tests of a single passenger car using RON 92 gasoline under three steady-state operating conditions (0, 20, 40 km/h). The fleet composition in the real tunnel is highly complex; Figure S1 in the supplement shows that the fleet includes gasoline cars, gasoline motorcycles, dual-fuel taxis, and electric vehicles. The steady-state emission profile of a single vehicle (lacking transient characteristics of acceleration/deceleration and mixed-vehicle types) is highly inadequate as a representative baseline for comparison. It is recommended that the authors discuss this limitation more deeply in the Discussion section and emphasize in the Conclusion that building a comprehensive traffic PM2.5 molecular library in the future will require integrated data from more vehicle models and more complex operating conditions.
4. The manuscript alternates among the terms "formula," "formula assignment," and "formula-ionization-mode assignment," but it does not explain how cases where the same neutral formula corresponds to multiple retention time peaks are handled. It is also unclear whether TCC (Tunnel-CD Common) matching is based solely on the molecular formula or if it requires alignment of both m/z and retention time. Structural isomers with the same molecular formula may have different sources; calculating the intersection based solely on the formula would overestimate the true chemical overlap between the exhaust and the tunnel samples. The authors need to clarify whether the comparative unit is the molecular formula or the chromatographic feature.
5. Tunnel samples encompass background air inputs, diesel/other vehicle emissions, lubricating oil, brake and tire wear, road dust resuspension, tunnel wall reservoirs, and in-tunnel chemical processes. The manuscript lacks background samples from outside the tunnel and does not include reference materials for tires, lubricating oil, road dust, or exhaust from other vehicles. Therefore, being "not detected in the CD profile of this single vehicle" merely indicates that these compounds are not covered by this limited reference profile; it does not prove they originate from non-tailpipe emissions or near-road transformations. Sulfur-containing formulas could also be collectively influenced by fuel/lubricating oil, other vehicle types, analytical blanks, and matrix effects.
6. The TCR (Tailpipe Coverage Ratio) is essentially "the proportion of the tunnel HRMS peak area corresponding to molecular formulas detected at least once in the single-vehicle exhaust." This may not be equivalent to the actual mass contribution or source contribution of the exhaust to the tunnel PM2.5. The authors should clarify this distinction.
7. The premise that "a controlled single-vehicle exhaust profile cannot fully represent actual traffic aerosols" is inherently a reasonable expectation. The manuscript needs to further highlight findings that go beyond this basic expectation. Please clarify what new scientific insights the clustering analysis provides compared to a direct molecular profile comparison.
Minor comments:
1. Line 40: “These processes collectively reduce the specificity of bulk chemical indicators and limit the effectiveness of traditional receptor models based on averaged aerosol composition metrics.” The meaning of this sentence is somewhat unclear, and the logical flow of the context needs to be better organized.
2. Could you please provide specific details about the test vehicle, such as the vehicle model, age, mileage, emission standard, engine load, and engine thermal state (e.g., hot/cold start)?
3. The Experimental Methods section mentions that 3 field blanks were used to correct the tunnel samples. Please clarify whether blank filters were simultaneously collected for the dilution channel or background air during the chassis dynamometer tests and used for background subtraction.
4. In the peak area bar charts in Figure 1, the data bars for the CD tests are multiplied by a scaling factor of 0.2 (x 0.2). It is recommended to briefly explain in the main text or the figure caption why this scaling factor was applied (e.g., because the CD sampling concentration was much higher than the tunnel environmental concentration) to prevent misunderstandings.