Method development and optimization

To develop and optimize a method for the separation of polar substances, different factors such as the stationary phase, eluent composition including different buffer concentrations and their pH, as well as gradients were tested. Method development was initially carried out using QC samples, as well as influent and effluent samples spiked with standards. The influence of each parameter was then evaluated and quantified using a scoring system. Considering the intended coupling of this separation method with NTS, we aimed to develop a suitable approach for multicomponent analysis, hence, to detect and separate as many compounds as possible. The investigations were initiated using ACN and ultra-pure water as the mobile phase system, following the recommendation of the column manufacturer and supported by numerous studies demonstrating the successful separation of polar compounds via LC-MS [18, 29, 31, 33].

Stationary phase

Two stationary phases for the separation of the polar substances were tested: the Trinity Acclaim mixed mode column from Thermo Fisher, and the SeQuant ZIC-HILIC from Merck. According to recent studies, both columns have shown a great potential in separating ionic and polar compounds [20, 33]. These columns, however, are unfortunately prone to high column bleeds [17]. Prior to method optimization, a test run with a QC sample was conducted with both of these columns employing methods proposed by Montes et al. [33] (for MMLC) and Boulard et al. [18] (for ZIC-HILIC) to observe their initial performance.

Mixed mode column

The column was equilibrated as recommended by the manufacturer. Despite a relatively long equilibration period, the first measurements revealed extremely high amounts of column bleeding. The column bleeds had a negative influence on the performance of the HRMS and caused issues with the ion source. Due to ion suppression, most of the reference standards were detected with extremely low intensities. This also caused low robustness in the measured intensities. The total ion current (TIC) of one measurement is shown in the supplementary material, Fig. S2. It was decided to eliminate this column from further studies to avoid problems with the HRMS.

ZIC-HILIC column

The same procedure as above was carried out for the ZIC-HILIC column. High column bleeds were at first observed with this column as well; however, after sufficient equilibration, they were reduced to an acceptable level. Further optimization experiments were only performed on the ZIC-HILIC column.

pH of the buffer system

Buffer pH values of 7.5, 6.0, and 4.5, along with the pH value of the non-adjusted ammonium acetate solution (6.8), were examined in positive mode. During this investigation, an isocratic method with a mobile phase 90/10% ACN/H2O + 20 mM buffer concentration was used.

The achieved retention times remained consistent across different pH values. At lower pH levels, an increase in the FWHM was observed. Peak intensity was highest for most substances at pH 6.8 and 6.0 but notably decreased at both higher and lower pH values, resulting in decreased sensitivity. For instance, melamine exhibited an intensity of 5.3E + 04 at pH 6.8, followed by 3.0E + 04 at pH 6.0, while at pH values 4.5 and 7.5, the method yielded intensities of 1.8E + 04 and 9.0E + 03, respectively. This was not the case for chlormequat and metformin, which showed slightly higher intensities at pH 6.0, representing the highest values recorded for both compounds. The pH 6.8 yielded in narrower chromatographic peaks, with lower FWHM values, particularly in the case of later eluting compounds, such as gabapentin. As a result, the pH value of 6.8 (the initial pH value of the ammonium acetate buffer) was selected for further method development. This value yielded the best results with the NTS workflow. For certain compounds, such as aspartame, the observed sensitivity at pH 7.5 and 4.5 was significantly reduced, leading the feature extraction software to classify these signals as background noise. Additionally, at these pH values, the software was unable to extract gabapentin, due to the extreme broadness of the XIC. Additional information regarding the impact of pH values on intensities, RT, and FWHM are provided in Fig. S3, depicting the XICs of selected compounds at the investigated pH values.

Concentration of the ammonium acetate buffer

For the examination of the buffer ionic strength, buffer concentrations ranging from 5 to 20 mM were evaluated. These were tested under isocratic conditions in positive mode. Unlike pH values, buffer concentration significantly affected chromatographic separation for some substances. The retention times of early eluting substances were not strongly influenced by this parameter. On the other hand, later-eluting substances demonstrated a significant increase in retention time with decreasing buffer concentration. This was particularly the case for strongly basic or cationic compounds like chlormequat and metformin. For the latter, retention times of 18.4 min at 5 mM, 12.7 min at 10 mM, and 9.2 min at 20 mM were observed. However, with a decrease in the buffer concentration, an extreme increase in the FWHM and decrease in the intensities were observed, which caused challenges in the NTS workflow. Looking closely at metformin again, a FWHM of 0.6, 0.4, and 0.25 was calculated for 5, 10, and 20 mM buffer, respectively. In conclusion, a buffer strength of 20 mM gave the most promising results. Extracted ion chromatograms (XIC) of exemplary substances measured at 5, 10, and 20 mM buffer concentration are illustrated in the supplementary material, Fig. S4.

Final method

Following the establishment of the mobile phase composition, 18 further methods were tested, where the oven temperature, the flow rate, and gradients were optimized. A detailed table including these parameters can be found in the supplementary material (Fig. S5 and Table S2).

Of the 18 tested methods, numbers 2, 3, and 14 delivered promising results based on the scoring system mentioned in the “Evaluation and selection of the most suitable method” section. While all three methods exhibited satisfactory performance, it was observed that method 3 demonstrated a slightly better separation of the cationic/basic substances in the complex influent and effluent matrices. This method also showed the highest compatibility with the MZmine3 (NTS) workflow, exhibiting the highest recall rates across quality control (QC), influent, and effluent samples when processed through the software. Consequently, method 3 was selected as the preferred method for conducting further experiments. It is noteworthy that the reference standard glyphosate was not detected with either method. Additionally, despite optimization efforts, the separation of the substances 1,3 di-o-tolylguanidine and 3-amino-1,2,4 triazole remained a challenge. Substances acquired in negative mode, such as cyanuric acid and 5-fluoroacil, showed relatively lower intensities, due to the lower sensitivity of the HRMS instrument in this mode. The final method is also mentioned in the “Instrumentation and acquisition” section of “Materials and methods.”

XICs of the selected substances obtained with method 3 in positive and negative modes are visualized in Fig. 3. As expected, opposed to the classic C18 method, compounds with a higher log D value such as cardiol and caprolactam (RT: 1.5 and 1.9 min, respectively) were eluted earlier than melamine and metformin (RT: 7.2 and 10.3 min, respectively), two examples of compounds with low log D values. In general, with the established method basic and amphoteric substances showed the highest retention (minutes 6–10), whereas acidic substances were eluted earlier (minutes 1.5–3).

Fig. 3

Selected extracted ion chromatograms (XICs) of reference standards in QC samples acquired with the final method (method 3) in ESI (+) and (−) modes

Repeatability and reproducibility

Prior to applying the developed method to real data and WWTP monitoring, it was crucial to assess the repeatability and reproducibility of the method in both positive and negative modes. This was particularly important since HILIC separation methods are usually associated with a lack of robustness, especially over longer periods of time [42]. Given that this method should be applied for NTS workflows, it was of great interest to focus on the robustness of the RT values which play a major role in the alignment of measurements. Intensity values are another important factor, as these values are used for trend analysis or quantitative screening analyses. Given the dependency of the NTS workflow on intensity values rather than peak areas, %RSD of the peak areas was not considered.

Repeatability or intraday precision of the reference substances was carried out in solvent and two further matrices, namely QC, influent and effluent (n = 10). The % RSD of the substances’ intensities and RTs were calculated and are presented in Fig. 4.

Fig. 4

Boxplots representing the range of calculated % RSD for a RT and b intensity of the compounds in positive and negative modes (n = 10) to determine the repeatability of the separation method (method number 3). The repeatability experiments were carried out in solvent (QC) and two wastewater matrices

In general, the average % RSD of the standard compounds’ RT was < 0.3 in the solvent and the wastewater matrices. This low value is satisfying for a HILIC separation method. As Fig. 4a and b suggest, the highest %RSD was interestingly observed in the QC. A closer look at the RT %RSD of each substance in Fig. S7(B), acephate showed high retention time shifts in the influent matrix, whereas caprolactam showed near zero %RSD in all samples.

Although the %RSD of the intensities were substantially higher compared to the RT, they still had an average value of < 5 in wastewater matrices and the solvent, which is a satisfactory outcome. Similar to the RT study, acephate was detected with significant shifts in its intensity across all three samples (Fig. S7(A)).

Reproducibility or interday precision was carried out only with the QC. The experiment was conducted exclusively in solvent due to the uncertainty surrounding the consistency of influent and effluent matrices over the course of 10 days. Daily samples have fluctuating matrices, and the chemicals within these samples are prone to degradation over time. This inconsistency and instability would interfere with the experiments and introduce uncertainty to the results.

The QC sample was measured once a day over a period of 10 days. The RT and intensity %RSD of the substances are presented in Fig. 5. There were no major shifts in the RTs of the substances, with the average of %RSD at 1. However, the intensity values exhibited less stability, with an average %RSD of 6. The intensities of specific substances, such as acephate, thiourea, and aspartame were responsible for this discrepancy (Fig. S8). It is important to note that the performance of the HRMS system also plays a role in the obtained results. To ensure this matter, multiple internal calibration of the HRMS within each sequence was carried out.

Fig. 5

a Boxplots representing the calculated % RSD of a RT (orange) and b intensity (blue) to determine the reproducibility of the method over 10 days in QC samples

In conclusion, the developed method showed acceptable repeatability and reproducibility in the case of most substances. For some compounds such as acephate and thiourea however, high %RSDs were observed.

Linearity and limits of detection

Considering that this method is to be applied for trend analyses and potential quantitative screening, it was crucial to evaluate the linearity and LOD of the method. This study aimed to provide a general understanding of the method’s detectability LOD across a range of different compounds, helping us evaluate its overall performance. Although future analyses will be performed on wastewater samples, it was decided to conduct the linearity and LOD experiments using QC samples. This was due to the fact that the influent wastewater matrix may already contain the target substances, potentially causing uncertainties or interferences in the investigation. In general, the established method provided good linearities and low LODs for most of the substances, the results of which are presented in Table 1.

Table 1 The linearity presented through the correlation coefficient (R2) of the calibration curve and LOD of the reference compounds were investigated with the established method. Compounds measured in negative mode are indicated with (−). The LODs of 1,3-di-o-guanidin and saccharin (−) are estimated at ≤ 0.1 μg/L based on the achieved S/N

The method was further assessed by calculating the LOD at a S/N of 3, with the lowest examined concentration at 0.1 μg/L. Overall, twelve compounds presented an LOD of ≤ 1 μg/L. Specifically, at a concentration of 0.1 μg/L, 1,3-di-o-guanidine and saccharin both presented intensities with S/N ratios ≫ 3. In contrast, the method achieved a high LOD of 40 µg/L for acephate and cyanuric acid.

Furthermore, an 8-point calibration curve was constructed for each substance covering a concentration range of 1 to 100 μg/L. However, for substances with higher LODs, calibration curves were constructed using fewer points. Analysis of the correlation coefficients (R2) derived from these calibration curves revealed satisfactory linearity for all substances, with coefficients around 0.99. An exception was observed for cyanuric acid and picrylsulfonic acid where the R2 value fell to 0.91 and 0.69, respectively. The linearity range varied between substances. While some showed linearity covering the full tested concentration range and possibly even beyond, others had a narrower linear range. This study demonstrates that while a multicomponent method for polar compounds is feasible, it faces challenges with some substances due to differences in generated signal intensities across the compounds.

NTS of polar substances in industrial wastewaterCreation of a chemical fingerprint for the junctions

Firstly, one 24-h sample from each junction, one influent, and one effluent sample were measured with the routine RPLC and with the established ZIC-HILIC method. NTS was conducted on both datasets, and the feature lists were compared by matching exact masses within a deviation of ± 3 ppm (Fig. 6). Between 12 and 29% of features were separated successfully with both columns. This outcome was also well expected, as the Restek Aqueous C18 column has polar functional groups offering a better retention of some polar components. Nevertheless, in all five junctions, some features were exclusively detected using the ZIC-HILIC column, highlighting the need to incorporate a method targeting polar compounds into the routine NTS monitoring of industrial wastewater.

Fig. 6

NTS was employed to measure one single 24-h composite sample from each of the five junctions A–E, the influent (IN) and the effluent (EF) with both the C18 (orange) and the established ZIC-HILIC (blue) method. The Venn diagrams display the percentage of features detected only with the C18, the percentage of features detected only with the ZIC-HILIC, and percentage detected with both columns. These outcomes are calculated based on only the exact masses with a deviation of ± 3 ppm

The effluent sample exhibited the highest percentage of features detected exclusively with the ZIC-HILIC method, accounting for 44%. In this sample, 27% of the overall features were detected by both methods. This outcome was anticipated, considering that the effluent matrix contains persistent polar substances that may be generated during the treatment process.

With the ZIC-HILIC column, 455, 480, 906, 818, and 1137 features were detected for junctions A through E, respectively. An in-house database was created, consisting of a library for each junction, listing the m/z, RT, and an ID for every feature indicating the name of the junction where the feature was detected. Therefore, a “chemical fingerprint” was created for A through E, aiming to track the features in the influent and effluent back to their point of origin.

Exploration of the chemical space

The influent and effluent of an industrial WWTP were monitored over a 10-day time frame. Consecutive 24-h composite samples were measured with the ZIC-HILIC method. The NTS workflow was carried out and features were extracted. As an additional filtering step, all features with a retention time of < 2 min were removed from the feature lists, to focus on substances with higher polarity. This reduced the number of features in the influent by 63% and in the effluent by 40%. It is important to note that the quality of the generated data was assessed prior to data analysis, with the detailed results presented in Sect. 6 of the supplementary material.

The influent and effluent data matrices were subjected to PCA individually to visualize the data and to identify any events or patterns within this time frame. The score plots are visualized in Fig. 7a and b.

Fig. 7

Ten consecutive 24-h composite influent and effluent samples were measured with the established ZIC-HILIC NTS approach. Features were extracted, preprocessed, and subjected to PCA. PCA score plots of a influent and b effluent samples over 10 days are presented

The score plot of the influent suggests that the features in these samples remained mostly consistent in composition and behavior from days 1 through 5, indicating stability over this period. Additionally, a grouping tendency or a “pattern” was observed every 2 days suggesting a potential temporal trend or correlation. For example, data points corresponding to days 1 and 2 demonstrate proximity to each other, as do data points representing days 3 and 4. The score plot also indicates that starting at the 6th day, there were changes in the composition of the influent. These changes were further intensified on days 7 and 8. However, towards the end of the investigation period, particularly on days 9 and 10, the composition began to resemble that of day 6. Upon closer examination of the dataset, it was apparent that from day 6 onwards, a cluster of new features emerged in the influent. The number of these newly emerged features almost doubled on days 7 and 8, accompanied by an increase in their intensities compared to day 6. The source of this irregularity was further investigated and pointed to junction C within the industrial park.

Interestingly, the observed pattern and the significant change in the influent’s composition on the 7th and 8th day were also reflected in the effluent, however with a 1-day delay. Hence, days 8 and 9 of the effluent appeared to have a distinct content compared to the other days.

Removal efficiency of the investigated WWTP

The data matrices of the influent and effluent were aligned, to create one feature list with their respective intensity profiles across both sampling sites. The removal efficiency of the WWTP was estimated by calculating the average intensities of the features in the influent and effluent over the 10-day period. The results are depicted in Fig. 8. Around 90% of the compounds were fully eliminated by the WWTP. The other substances were removed to a large extent and detected in the effluent at trace intensities. These results demonstrate the effective and strong performance of the WWTP, especially in treating industrial polar substances [8, 43, 44]. Just as in the case of all WWTPs, biological TPs were also detected in the effluent, which are a result of the treatment [45,46,47].

Fig. 8

Distribution of removal rates for detected features in the WWTP. The removal rates are based on the intensities of the features detected in the influent and effluent. The bar graph illustrates the number of features corresponding to their respective removal rates

These findings are specific to this 10-day period. Highly sensitive targeted analytics, the state-of-the-art monitoring method today, is used continuously to monitor the efficiency of this WWTP.

Investigation of poorly removed features

Five of the substances with the lowest removal rate were prioritized and their intensity trends in the influent samples over the 10-day investigation period were plotted (Fig. 9). Feature F293 had the lowest removal rate of only 2%. The intensity plot of F293 indicates higher concentrations released in the wastewater between days 6 and 9. A plausible explanation for the low removal rate is that the WWTP microbiology may not be adapted to degrade this substance, especially when it is introduced in sudden high quantities. Similar observations were made for features F250 and F126, which had slightly higher removal rates of 5% and 6%, respectively. In the case of these features, extending the study duration could provide further insights into whether their poor removal is due to sudden high concentrations or if the WWTP microbiology is inherently unsuitable for their degradation. The degradation rate is specific to the physicochemical properties of the substance and the microbiology of the WWTP. Therefore, these substances may achieve higher removal rates over time as the WWTP adjusts to this change. Features F486 and F174 were present in high amounts in the influent throughout the investigation. Although these were partially removed, they can be considered as persistent polar compounds.

Fig. 9

The features with the lowest removal rates and their intensity trends in the influent samples. These could be pinpointed to their emission source within the industrial park using the created chemical fingerprints (“Creation of a chemical fingerprint for the junctions” section)

To take a step forward in reducing the amount of these substances in the wastewater, the sources of these compounds were investigated by employing junction chemical fingerprints. Two of the features, namely F293 and F250, could be exclusively traced back to junction C. F126 was detected at higher intensities (+E06) in samples from site D. It was also present in junction E due to the merging of wastewater from D into site E. Although F486 was detected in C and E, it is most probably released from one of the E plants, as the intensities in the E samples are higher by almost a factor of 10 in comparison to C. Interestingly, F174 was detected in all junctions with the exception of A. The next step involves identifying and notifying the specific chemical plants responsible for these emissions.