Cellulose in atmospheric particulate matter

The spatiotemporal variations of free cellulose concentrations in atmospheric particles, as a proxy for plant debris, were investigated using a novel HPLC-PAD method. Filter samples were taken from nine sites of varying characteristics across France and Switzerland, with sampling covering all seasons. Concentrations of cellulose, as well as carbonaceous aerosol and other source-specific chemical tracers (e.g. Elemental Carbon (EC), levoglucosan, polyols, trace metals, and glucose) were quantified. Annual mean free concentrations correlated poorly between sites, even a ranges of about 10 km, indicating the 36 localised nature of the sources of atmospheric plant debris. With regards to these sources, 37 correlations between cellulose and typical biogenic chemical tracers (polyols and glucose) 38 were moderate to strong ( R s 0.28 – 0.78, p < 0.0001) across the nine sites. Seasonality was 39 strongest at sites with stronger biogenic correlations, suggesting the main source of cellulose 40 arises from biogenic origins. A second input to ambient plant debris concentrations was 41 suggested via resuspension of plant matter at several urban sites, due to moderate cellulose 42 correlations with mineral dust tracers, Ca 2+ and Ti metal ( R s 0.28 – 0.45, p < 0.007). No 43 correlation was obtained with the biomass burning tracer (levoglucosan), an indication that this 44 is not a source of atmospheric cellulose. Finally, an investigation into the interannual variability 45 of atmospheric cellulose across the Grenoble metropole area was completed. It was shown that 46 concentrations and sources of ambient cellulose can vary considerably between years. All 47 together, these results deeply improve our knowledge on the phenomenology of plant debris 48 within ambient air. tracers in ambient aerosol, collected at nine sites across both France and Switzerland. The objective of the study was to investigate the seasonal and geographical variability of atmospheric cellulose across sites of varying characteristics. Contributions of cellulose to the OM fraction of PM, and correlations of cellulose with tracers of characteristic sources were also completed, as well as an interannual study of cellulose concentrations at three sites within 133 the Grenoble (France) conurbation. This study, with the gathering of one of the largest data bases on atmospheric cellulose with more than 1500 samples, aims to provide a better understanding of this understudied component of atmospheric PM.


Introduction
remains key to uncovering more knowledge of its climatic and health effects, on both local and 62 larger scales (Nozière et al., 2015). Indeed, it has been hypothesised that our current    The concentration of "free" cellulose within the filter samples was determined following an 244 improved protocol based on the enzymatic procedure proposed by Kunit and Puxbaum (1996). 245 Free cellulose was extracted in an aqueous solution, which was then enzymatically hydrolysed 246 to glucose units using two cellulolytic enzymes. The glucose concentration was then quantified 247 by using an HPLC-PAD method. The hydrolysis step was the same as originally proposed, 248 however the enzyme quantities and analytical step have been modified in our protocol. The HPLC-PAD (Dionex DX500) was equipped with a Metrohm column (250 mm long, 4 mm 262 diameter), with an isocratic run of 40 minutes with the eluents A (84%, H2O), B (14%, 100 263 mM NaOH), and C (2%, 100 mM NaOH + 150mM NaOAc). Column temperature was 264 maintained at 30 °C. Eluent flow rate was 1.10 mL min -1 , and injection volume was 250 μL. 265 Each analytical batch contained six glucose and six cellulose hydrolysis standard solutions, 266 alongside unknown samples. Cellulose standards are used to calculate the cellulose-to-glucose 267 hydrolysis efficiency for each batch and are made from cellulose beads of 20 µm (Sigma 268 Aldrich, S3504). The final calculation of the atmospheric concentration of the free cellulose 269 takes this efficiency of conversion into account. The efficiency was variable between batches, 270 but was typically between 75 -94%, resulting in an average of 85 ± 8 %. The calculation also 271 subtracts the initial concentrations of atmospheric glucose of each sample, determined in 272 parallel with the aforementioned analysis of sugars and polyols. Finally, field and procedural 273 blanks are taken into account. The procedural blank results are greatly improved when the stock 274 cellulase enzyme solution is filtered to lower their glucose content. This is performed through 275 a series of centrifugal cleaning steps (n=10) by tangential ultrafiltration in a Vivaspin 15R tube 276 at 9000 rpm in Milli-Q water. Additional procedural information can be found in the 277 supplementary information (SI). 278

Cellulose Method Validation 279
This novel cellulose quantification method was subjected to a repeatability test, in order to 280 quantify the uncertainties with respect to glucose content within the filter punches. Briefly, a 281 high-volume sampler (Digitel DA80, 30 m 3 h -1 ) was used to collect PM10 onto a pre-fired quartz 282 fibre filter (Tissu-quartz PALL QAT-UP 2500 diameter 150 mm) on the roof of the laboratory, 283 and sampled a total of 615.1 m 3 of air on 15/03/2021. Ten filter punches of 21 mm were then 284 taken and subjected to the same cellulose-to-glucose enzymatic procedure as for normal 285 samples. It is important to state that we assume constant concentrations of both native glucose 286 and cellulose within the filter, as well as the same enzymatic cellulose-to-glucose conversion 287 efficiency for all ten filter punches. Each filter punch was then analysed three times using the 288 same HPLC-PAD method, to monitor repeatability in terms of both cellulose hydrolysis and 289 PAD glucose concentration measurements. Post hydrolysis, the total glucose content of the ten 290 filters was found. The variability (Relative Standard Deviation -RSD) was small, ranging from 291 0.7 -5.7 % for the three repeats of the same filter sample. The RSD of the glucose content 292 within the ten filter punches was calculated to be 9.9 %. For a 95% confidence in the 293 uncertainty estimate, the uncertainty in the measurement was therefore found to be 20% at a 294 maximum. 295 296

Limit of Quantification 297 298
In order to check for potential contamination of filters during transport, sampling and storage, 299 blank filters were taken across the nine sites. Within the Grenoble metropole, blank filters were 300 taken at Les Frênes and then applied to Caserne de Bonne and Vif (labelled QAMECS in Table  301 3). Further, blanks filters were taken at ANDRA-OPE on both PM10 and PM2.5 sampling days. 302 With regards to the Swiss sites (EMPA), blanks were taken from each sampling site and an 303 average glucose concentration taken from across the five locations. 304 305 Glucose concentrations calculated in the blanks were then subtracted from measured glucose 306 concentrations within each sample. After, any sample that then yielded a negative 307 concentration of glucose was deemed to be lower than the quantification limit (< QL), 308 representing 5.2 % of all samples. Table 3 summarises the concentrations of cellulose on the 309 blank filters, which has been converted from the blank glucose concentration and the average 310 sampling volume taken across the series. QL varied according to the site, from 0.53 to 13.4 ng 311 m -3 . In subsequent analyses of monthly, seasonal or annual concentrations (sections 3.1 -3.3 312 and 3.6), any sample that was deemed < QL was assigned a cellulose concentration of 313 [Blank]/2. This prevents an artificial increase in average cellulose concentrations.

316
In the following, cellulose concentrations are reported as "free" cellulose. The multiplication 317 factor of 1.39 derived by Tenze- Kunit and Puxbaum (2003) could have been used to derive 318 "total" cellulose. We chose not to do this, due to the large uncertainty in this ratio. From this 319 point onwards, "free" cellulose will be regarded as cellulose. 320  Table 4. An expanded version of Table 4, also including previous literature results, can be 331 found in Table S1 (SI). The evolution of cellulose concentrations across the respective 332 sampling periods for our study have further been included in the SI (Fig. S1).       Importantly, across the three sites, less than 30% of atmospheric cellulose was found within 373 PM2.5, on average. This large data set of size resolved cellulose concentrations confirms that 374 plant debris predominantly resides within the coarse aerosol mode. Thus, the remainder of this 375 work will solely discuss PM10 data to understand atmospheric cellulose and its behaviour. 376 Cellulose concentrations in Basel (suburban) also show minimal seasonality, but this may be 406 due to concentrations being too small to exhibit a full seasonal pattern. This is surprising, given 407 the close proximity of the site to a park-like area with trees and gardens. The lack of seasonality 408 in urban settings, however, is consistent with the findings of Caseiro (2008). Additionally, 409

Size distribution
Caseiro (2008)  Reasons for this are unclear, but this is suggestive of a source change in atmospheric plant 418 debris, or an additional source being present at some urban locations, that may mask the typical 419 seasonality. Given that these locations are urban in character, the weak seasonal variations may 420 be owing to anthropogenic activity. This will be investigated in section 3.5. 421

Contribution of Cellulose-C to OC 422 423
To determine the overall importance of cellulose contribution to PM, the percentage 424 contribution of cellulose-carbon to total organic carbon (Cellulose-C to OC) was determined. 425 The highest contributions to OC were typically found at rural sites, potentially due to fewer 431 local sources of OC in rural sites compared to more urban locations. In fact, the annual 432 contribution to OC found at Payerne (5.9 ± 4.4 %) is the highest found in literature. However, 433 the annual average for the urban site of Zurich is also in a high range, at 3.8 ± 2.9 %. Regarding 434 seasonal contributions, the rural sites in this study show a significantly different seasonal 435 pattern compared those found in the study by Sánchez-Ochoa et al. (2007). Here, we see a 436 noticeably smaller contribution of cellulose-C to OC during winter compared to summer. This 437 While seasonal contributions appear to be moderate in most cases, the contribution of cellulose-442 C within episodes can be much more significant. It is also worth noting that these contributions 443 to OC are derived from free cellulose concentrations. Thus, the contribution to overall OC will 444 be higher when considering total cellulose. At sites with typically lower seasonal contributions 445 (Basel, Bern, LF), the episodic contributions reached between roughly 4.1 to 6.3 %. However, 446 at the sites that illustrated a much higher seasonal average contribution to OC, the maximum 447 contributions during episodes were found to be between 16.1 % at Zurich and 19.7 % at 448 Payerne. These maximum contributions (detailed in Table S5)  Lastly, the contribution of coarse mode (PM with diameter less than 10 µm and greater than 455 2.5 µm) cellulose-C to coarse mode OC was evaluated at the three sites that completed both 456 PM10 and PM2.5 analysis (ANDRA-OPE, Payerne and Zurich). This can be seen in Table S6, 457 in the SI. As PM2.5 data for ANDRA-OPE was only available for the 2020 sampling campaign, 458 PM10 data from 2016 and 2017 was excluded. Table S6 shows a contribution of coarse 459 cellulose-C to be 3.16% at ANDRA-OPE, which is of very similar magnitude to that of the 460 overall cellulose-C contribution to OC. This is potentially due to the significant reduction in 461 cellulose source strength at the ANDRA-OPE site during the year of 2020, compared to the 462 years prior. This will be discussed in section 3.7. However, at both Payerne and Zurich, the 463 annual contributions to coarse OC are notably higher (11.02 % and 13.04 %, respectively) than 464 that of overall cellulose-C to OC (5.88 % and 3.76 %, respectively). From this data, we can see    Polyols are also used to provide tracer correlations with cellulose. These species are typically 494 used as markers of airborne fungi but have also been found to be present within leaves and 495 pollen (Medeiros, 2006). 496

497
As we can see in Table 5 cellulose-polyol concentrations. The stronger correlations at the rural sites indicate that a 506 significant portion of atmospheric cellulose, and thus plant debris, arises from biogenic sources 507 at these sites. As the values are typically below 0.7, this could suggest a different timing of 508 emissions between biogenic tracers and cellulose (e.g. meteorological conditions favouring 509 emission of fungal spores before plant debris). This is a distinct possibility, given that sampling 510 ranges between 3-6 days at the nine locations. Additionally, these moderate correlations with 511 biogenic tracers could be due to some input from other sources, but of a lower magnitude. By 512 contrast, the weaker correlations observed at most urban sites suggest that there remain other, 513 potentially more prominent, sources at play that determine atmospheric cellulose 514 concentrations. The two exceptions to this, LF and Bern, show that the sources of atmospheric 515 plant debris are not consistent within each designated site type. 516 517 It is noteworthy that the five locations that illustrate the strongest correlations with glucose and 518 polyols are the five out of the six sites in which the common, general-case seasonality is 519 observed. It is thus likely that this typical seasonality pattern is observed where the biogenic 520 source of plant debris is the most dominant.  (Table S7, SI). Of all sites, the Grenoble-based locations (Caserne de Bonne, 538 Les Frênes and Vif) were the only three to have greater than the thirty data points of 539 simultaneous cellulose and levoglucosan measurements needed for a robust correlation. None 540 of these three locations showed any correlation between cellulose and levoglucosan (Rs 0.05 -541 0.18, p 0.14 -0.74). In fact, the remaining six locations showed also very weak correlation, 542 except for the site of Bern, which showed a moderate correlation (Rs 0.49, p < 0.03). But, as 543 already mentioned, the relatively small wintertime dataset for these six other sites (n = 21 to 544 25) does not provide strong confidence in these results. Thus, we can state that the sources of 545 atmospheric plant debris do not include any significant input from biomass burning. given the lack of significant correlation with EC, it seems that a resuspension mechanism may 581 not include a vehicular input. Other anthropogenic resuspension mechanisms not related to 582 traffic may contribute; paper usage (e.g. newspaper and cardboard production) has been 583 mooted in previous literature (Caseiro, 2008  consistent across different regions and scales. This trend, or lack thereof, was expressed 607 numerically using correlation coefficients (R 2 ) of monthly concentration averages for the 608 groups of sites that were sampled at the same time. As shown in Table 6, the correlations 609 between sites within the Grenoble metropole (CB, LF and Vif) are low to moderate. This is 610 also the case for the Swiss sites, which span a much larger spatial range compared to the 611 Grenoble-based sites. The lack of a shared temporal variability seems to indicate that the major 612 sources of plant debris are most likely to be local to each site. It may also suggest that several The R 2 values in Table 6 (Table S3 and Table S4). Polyols are used as chemical tracers for fungal spores, a 632 very common class of PBAP, and here provide a near perfect example of a PBAP class 633 displaying homogenized concentration variations over time at a regional scale. This suggests a 634 single common source of polyols that is impacted similarly by external factors across all 635 locations, especially at short range e.g. within the Grenoble area. This was also suggested by 636 The stark contrast between the two sets of chemical tracers (cellulose vs. polyols) highlights 642 the rather local nature of atmospheric plant debris and its sources. Given that meteorology is 643 relatively consistent on a short to medium scale (< 200 km), it would be expected that plant 644 debris emissions would impact all sites of a given area similarly. However, heterogeneous 645 distribution of the diverse plant species at the city (or regional) scale might induce specific 646 temporal variations in the emissions of plant debris at the local scale. Therefore, the lack of 647 correlation in cellulose datasets may result from site-to-site differences in the dominant sources 648 (flora) or emission processes of ambient plant debris (Caseiro, 2008). 649  Seasonal and monthly average temperatures across the two sampling periods show some 675 differences, but the variation is slight (Fig. S4, SI). It is highly unlikely in this instance that the 676 large variations in the atmospheric cellulose concentrations were caused by ambient 677 temperature changes. This is further supported by the lack of change in seasonal average polyol 678 concentrations for the same sites, shown in  Table S9). Thus, it is likely that changes in atmospheric cellulose concentrations will have 690 resulted from changes in the source strength of plant debris, and not from a wider-scale 691 reduction in some or all other OC sources. 692 Given that these large interannual variations seemed to be predominantly limited to cellulose 693 and not the remaining sources of OC, it was necessary to evaluate the potential sources once 694 more. Following section 3.5, cellulose-tracer correlations were again produced using the same 695 characteristic source tracers for the two periods, to see if changes in cellulose concentrations 696 were consistent with variations in tracer correlations. These correlation coefficients can be seen 697 in Table 7 (Table S10 for full table). From the two sets of correlations, it is evident that the 698 between cellulose and Ca 2+ concentrations during the 20/21 campaign that was absent during 706 the previous series. This is particularly visible at Vif, but it is also a consistent trend across all 707 three sites. These findings suggest potentially two possible hypotheses. Firstly, the contribution 708 of plant debris arising from biogenic sources has been much weaker during the second 709    Further, we once again see a noticeable reduction in the contribution of cellulose-C to OC (%) 732 during the 2020 sampling period, compared to the two previous campaigns, especially during 733 summer and autumn (Fig. 9, numerical values Following these significant interannual variations within cellulose concentrations and 741 cellulose-C to OC, correlations of cellulose with source-specific tracers were completed to see 742 how the source of atmospheric plant debris has changes between the three sampling periods 743 (Table 8, p values in Table S13,  coordinator of the ANDRA-OPE site and atmospheric program, and provided the samples from 841 this site. CH is the head of the NABEL network in Switzerland, provided all samples from this 842 country and directed the program for this yearly sampling; SKG was the curator of the swiss 843 data. OF is responsible for the CARA program from the LCSQA in France, and provided partial 844 funding for sample analysis at LF site. CT was responsible for the sampling by Atmo-AURA 845 at the 3 sites in the Grenoble area. All authors reviewed and commented on the manuscript. 846