A sensitive and robust method for automated on-line monitoring of enzymatic activities in water and water resources

The realisation of a novel concept for automated on-line monitoring of enzymatic activities in water was successfully demonstrated by long-term field testing at two remote Austrian ground water resources. The β-D-glucuronidase (GLUC) activity was selected as a representative enzymatic model parameter for the on-line determination. But the device can be adapted for any enzymatic reaction with diagnostic relevance for microbial water quality monitoring, as demonstrated for the β-Dgalactosidase activity. Automated filtration of volumes up to 5 litres supports sensitive quantification of enzymatic activities. Internet-based data transfer, using internal control parameters for verification and a dynamic determination of the limit of quantification, enabled robust enzymatic on-line monitoring during a 2-year period. A proportion of 5,313 out of 5,506 GLUC activity measurements (96.5%) could be positively verified. Hydrological (discharge, gauge, turbidity, temperature, pH, electric conductivity, spectral absorbance coefficient at 254 nm) as well as microbiological parameters (Escherichia coli, coliforms) were concurrently determined to characterise the investigated ground water resources. The enzymatic on-line measurements closely reflected the different hydrological conditions and contamination patterns of the test sites. Contrary to expectations, GLUC did not qualify as a proxy-parameter for the occurrence of cultivation-based E. coli contamination and warrants further detailed investigations on its indication capacity as a rapid means for microbial faecal pollution detection in such aquatic habitats. Microbial on-line monitoring is likely to become more important in the future, complementing existing surveillance strategies for water safety management. Further perspectives on the application of such analytical on-line technologies, such as their connection with event-triggered sampling and standardised diagnostics,


INTRODUCTION
The determination of microbial water quality is a key requirement for water safety management. Standard determination procedures still rely on cultivation-based methods. Unfortunately, these methods are laborious and time consuming and are not suitable for rapid water quality assessment (ISO ). In contrast, online monitoring of chemo-physical and hydrological water quality parameters has been applied frequently during the last decade. There is a significant lack of automated on-line systems for the detection of microbial parameters. Enzymatic activity determination has the potential for rapid measurements. For example, several publications from the last two decades have suggested the use of direct β-D-glucuronidase (GLUC) determination for monitoring faecal pollution in various water sources (Fiksdal et (Togo et al. ) quantification. However, conventional determination of enzymatic substrate hydrolysis requires a specialised laboratory structure as well as laborious sampling, manual sample handling and further processing procedures.
In this study, a novel automated device that is capable of measuring enzymatic activity in water by applying fluorogenic model substrates was rigorously tested. The device supports automated sampling of adjustable sampling volumes (100 ml up to 5,000 ml) and allows online quantification of enzymatic activities. Internet-based transfer to the user of the generated activity data, including internal quality control parameters, enables remote monitoring, with the final goal of supporting near-real-time water quality management. The device was permanently installed at the measuring site to allow on-line data generation. The GLUC activity was selected as a representative model parameter, but the device can be adapted for any enzymatic determination for which substrates are available.
The main focus of the study was a rigorous field evaluation of the following: (i) the operational applicability, (ii) the stability of measurement, and (iii) the reliability of data transfer of the newly developed measuring device. Over a period of 2 years (2010 to 2011), continuous GLUC data were automatically generated and analysed together with concurrently recorded physicochemical, hydrological and microbiological parameters. Laboratory testing of the biochemical measuring principle was also applied to verify comparability to standard enzymatic determination. The possibility to adapt the device to other enzymatic activities was verified by application of the β-Dgalactosidase assay. Because the Austrian water supply relies on almost 100% ground water, two representative but hydro-geologically different ground water sources were chosen as test sites.

MATERIALS AND METHODS
The automated enzymatic measuring device: the fluidic system and the determination process The central element of the automated measuring device (Coliguard EChs prototype, mbOnline, Austria) is the reaction chamber, where the investigated water sample is concentrated and the enzymatic activity in the water sample is estimated. The whole fluidic system is outlined in Figure 1.
The measuring process consists of several consecutive steps and starts with the initialisation of the instrument (checking the electrical components and flushing the sample and reagent lines). The sample is pumped through the reactor and the filter. Programmable sample volumes range from 100 ml to 5,000 ml (we used 1,000 ml). The filtration pressure is monitored. For incubation, a buffer is injected, and the pH is adjusted during continuous stirring of the chemical enzyme substrate. For GLUC activity determination, methylumbelliferyl-β-D-glucuronide (MUG) is added until a final concentration of 200 mg l À1 and 0.025% Triton-X-100 (w/v) is reached; the temperature is set to 44 ± 0.1 W C according to George et al. (). The total Figure 1 | The fluidic system of the automated measuring system for GLUC determination. The sample is aspirated through the input tubing (1) by the sample pump (3), which also measures the sampled volume and is calibrated for flow rates from 2-20 ml min À1 . During forward pumping, the sample is pumped through the reactor (5). The pressure is monitored by a sensor (4), which also regulates the flow rate. The reactor consists of a reaction chamber and a filtrate chamber, separated by a ceramic filter with a 0.45 μm pore size (7). The reaction chamber of the reactor comprises a heating unit, a temperature sensor, an agitation unit and a window for the fluorescence optic measurement.
Capillaries for reagents (6) are used to support the reaction chamber. The system valves (2) direct the flow in the system. A launching cleaning procedure is performed using heat incubation and a cleaning solution (8). The resulting waste is discharged through the waste line (9).
incubation time was set to 75 min (default). Degassing of the reactor limits the formation of air bubbles and reduces the risk of failure during detection. The subsequent enzymatic reaction in the agitated reaction chamber is tightly controlled based on the control parameters, as given in Figure 2. The increase of fluorescence is permanently monitored, and the slope of the signal in the steady state phase is used to calculate the enzymatic activity by least square linear regression analysis (see also Figure 1add, additional material; available online at http://www.iwaponline.com/wst/069/032.pdf). The enzymatic activity is expressed as methylumbelliferyl (MUF) production per time and volume (pmol MUF min À1 100 ml À1 ); GLUC activity directly corresponds to the hydrolysis rate of the MUG substrate (Fiksdal et al. ; Farnleitner et al. ). Further technical details can be found in the respective patent description of the automated measuring device (AT505306, Lendenfeld ; Austrian Patent Office). Each measurement is concluded with a cleaning procedure, which uses a cleaning solution and heat incubation of the reaction unit to eliminate all enzymatic activity of the fluidic components and tubes. Blank values are typically determined once daily to capture minor technical and methodological changes, and are performed using the same procedure used for standard measurements except that the filtration volume is set to a minimum of 7 ml.
The instrument communicates the enzymatic activity data and numerous instrument control parameters (Figure 2(b)) in predefined time intervals via General Packet Radio Service (GPRS) using a virtual private network (VPN) protocol to a server application, where data are stored. A user connects to the server application to view measurements and instrument data and to set schedules or to change instrument settings (Figure 2(a)). Data can be extracted from the database by external programs for further use.
As an additional microbiological parameter, the activity of the galactosidase enzyme (GAL) was also determined using a second automated enzymatic measuring device to complement the GLUC activity measurement. GAL activity was estimated as the rate of hydrolysis of the substrate 4-methylubelliferyl-β-D-galactopyranoside (MUGal) (Sigma-Aldrich, St Louis, USA) due to an increase of the product MUF similarly to how the GLUC measurement was performed. The biochemical conditions during the measurements were optimised according to George et al. ().
Laboratory experiments were also performed to assess the comparability of the enzymatic assays as applied in the automated device using a standard enzymatic assay.
Comparison of the GLUC activity assay was performed at 44 W C (EC 3.2.1.31 Sigma Aldrich, G7646). The sample was simulated by a GLUC solution with an activity of 5 U ml À1 . The Sigma Aldrich assay was conducted according to the manufacturer's instructions except that MUG was used as the substrate instead of phenolphthalein glucuronide. The released MUF was determined using high performance liquid chromatography (HPLC) (Agilent 1100 System, Vienna, Austria).

Establishing a limit of quantification
The principle of the limit of quantification (LOQ) is based on a statistical approach that determines when the measured enzymatic reaction becomes significant compared to the background signal at the start of the measurement (time zero). The approach follows the basic philosophy established for the field of chemical analysis (Mocak et al. ) but includes the aspects of continuous and linear signal increase (based on zero-order enzyme kinetics). By the least square method, a linear regression curve and its confidence limits are fitted to the enzymatic measurements. The measured enzymatic reaction (or the corresponding slope of the regression line) at a certain point in time is regarded as statistically significant when the average measurement exceeds threefold the standard deviation (3σ) in relation to the theoretical zero line of the reaction (as exemplified by k 2, Figure 2add (additional material; available online at http:// www.iwaponline.com/wst/069/032.pdf)). Regression curves with slopes below 3σ (as exemplified by k 1, Figure 2add, additional material) are considered non-significant. The threshold slope (k t ) is defined as the slope of the regression curve that equals 3σ. Threshold slopes are calculated by the software of the automated enzymatic measuring device for each enzymatic activity determination and are considered to be the LOQ. It follows from the continuous increase of the signal, that, in theory, any measured enzymatic reaction will eventually become significant. However, the measuring time is usually fixed at a maximum incubation time of 75 min, although this can be adjusted to longer time intervals if needed.
Chemo-physical, hydrological and microbiological field parameters All hydrological and chemo-physical data were recovered by in-field on-line sensors directly installed at both test sites. At the limestone karst aquifer spring 2 (LKAS2) site, conductivity, water temperature and electrical conductivity were registered with the data collecting system GEALOG-S from Logotronic (Vienna, Austria  The picture of the user interface shows that the user is able to monitor the registered GLUC and LOQ as well as the internal control parameters (flow rate, max. pressure, filtration duration, start signal, base signal with/without buffer, sample volume) and the fill levels of the respective reagents. Physicochemical, hydrological and microbiological parameters were recorded at both locations during the 2-year investigation period (Table 1add; additional material, available online at http://www.iwaponline.com/wst/069/032. pdf). Although median values for the SAC254 and the turbidity were quite similar, the LKAS2 site showed much higher variation. Microbial indicators accurately demonstrated this difference in the frequency and extent of surface contamination. For LKAS2, E. coli counts exhibited a broad range from undetectable to 435 MPN 100 ml À1 (in total 95 (46%) positive detections), whereas at the PGAW1 site, E. coli was not detected in 100 ml throughout the period of the investigation. Additionally, coliforms could only be detected on very rare occasions (n ¼ 6) at the PGAW1 site, with a maximum of 3 MPN (most probable number) 100 ml À1 . In contrast, concentrations of up to 1,100 MPN 100 ml À1 were detected at the LKAS2 site (in total 153 (74%) positive detections). Recovered data clearly highlight the contrasting contamination characteristics of the selected ground water sources, providing a basis for the rigorous field testing of the automated measuring device.

Statistical analysis and software
Time corresponding data were extracted from data material using relevant functions of Excel (MS Office 2007).
Statistical analysis was performed with SPSS Version 17.0. Correlation analysis was performed using Spearman's rank correlation analysis.

Laboratory testing: comparability to standard biochemical assays
Laboratory testing demonstrated that the automated enzymatic measuring device yields comparable results to those determined using the standard Sigma assay. For the short incubation time, the Sigma assay revealed a hydrolysis rate of 2.1 mg MUF l À1 (n ¼ 10; coefficient of variation (CV) ¼ 3.9%, HPLC). The automated measuring device also yielded 2.1 MUF mg l À1 (n ¼ 10; CV ¼ 4.6%). The measurements taken after 75 min of incubation time showed a concentration of 5.2 MUF mg l À1 (n ¼ 10; CV ¼ 7.6%; HPLC) for the Sigma assay and compared to the automated measuring device, which detected 4.8 MUF mg l À1 (n ¼ 10; CV ¼ 3.6%).

Field testing
Basic performance characteristics of the automated measuring device The automated enzymatic measuring device was able to generate and transmit GLUC activity data approximately four times per day from both locations during 2010 and 2011. The device at the LKAS2 site registered 2711 GLUC activity measurements throughout the whole period, among which 2603 (96%) could be verified based on the internally defined control parameters (e.g., filtration pressure, incubation temperature and signal increase after adding substrate). The majority of the verified data (n ¼ 1804 corresponding to 69%) was also above the LOQ and was subjected to further analysis. At the PGAW1 site, 2795 measurements for GLUC activity were registered, among which 2710 (97% of the measured data) could be verified based on respective control parameters, but only 468 (17% of the verified) measures exceeded the LOQ and could be used for further evaluation. The LKAS2 site showed a broad range of GLUC activities, ranging from <0.1 to 8.4 pmol MUF min À1 100 ml À1 . The PGAW1 site was quite constant, ranging from <0.1 to 0.7 pmol MUF min À1 100 ml À1 over the 2 years of testing.
As many as 1404 GAL activity measurements were generated at the LKAS2 site for 2011 (not possible for 2010), and 1340 (95%) of these could be verified by the respective control parameters. Of these, 1193 (89% of the verified) were above the LOQ, ranging from <0.1 to 3.7 pmol MUF min À1 100 ml À1 . At the PGAW1 site, 2741 measurements of GAL activity were registered for 2010 and 2011. Of these, 2650 (97%) could be verified, and 799 (30% of the valid) were higher than the LOQ.

Dynamics of enzymatic on-line data
Automated measurements revealed remarkable temporal GLUC activity patterns for the test sites at LKAS2 and PGAW1, as demonstrated by the representative time series for the winter 2010/2011 and summer 2010 seasons ( Figure 3). In winter, the GLUC activity at the PGAW1 site showed almost no variation and was at approximately the same level as that of the LOQ. The GLUC activity determined at the LKAS2 site during winter 2010/2011 was also comparable with the activity at the PGAW1 site. However, one flooding event, which was in accordance with increased discharge, SAC254 and turbidity (data not shown), reflected the scenario of aquifer contamination by surface runoff (up to 2.5 pmol MUF min À1 100 ml À1 ). During the summer months, the situation at the PGAW1 site did not change significantly, showing only slight variations in the GLUC activity. In contrast, the GLUC activity measurements at the LKAS2 site during the summer reflected the highly dynamic regime of this aquifer, with pronounced variations and maxima of up to 6.0 pmol MUF min À1 100 ml À1 .

Correlating enzymatic activities with environmental data
For the LKAS2 site, a pronounced correlation between the GLUC data and the SAC254 (n ¼ 1804; ρ ¼ 0.64; P < 0.001), turbidity (n ¼ 1804; ρ ¼ 0.69; P < 0.001) and discharge (n ¼ 1804; ρ ¼ 0.77; P < 0.01) could be observed; GLUC and GAL activity also revealed a high correlation (n ¼ 666, ρ ¼ 0.72; P < 0.001; data only available for 2011). Correlations observed between GLUC activity and concentrations of cultivation-based E. coli (n ¼ 113; ρ ¼ 0.53; P < 0.001) and coliforms (n ¼ 113; ρ ¼ 0.63; P < 0.001) were below the correlation levels observed amongst the other environmental parameters (i.e., SAC254, turbidity) and GLUC. The PGAW1 site showed no correlation or very low correlations between GLUC activity and the physicochemical or hydrological parameters (ρ < 0.3). Remarkably, GAL activity measurements at the PGAW1 site significantly correlated with GLUC activity (n ¼ 113; ρ ¼ 0.42; P < 0.001; data obtained for 2010). As E. coli and coliforms were not detectable except on very rare occasions (n ¼ 6) at the PGAW1 site, these data were not used for further statistical analysis. . Continuous series of automatically generated GLUC activity values, measured four times per day, could be determined and transferred online using the automated enzymatic measuring device. As much as 96.5% of the single measurements (5313 out of 5506) could be positively verified by the control parameters (such as filtration pressure and reaction temperature). Remarkably, the automated enzymatic measuring device could be successfully used under highly fluctuating hydrological conditions, as observed at the dynamic LKAS2 site. Automated enzymatic measurements were robust, even with high variations in suspended and dissolved materials and with turbidity levels of up to 4.5 FNU (formazin turbidity unit) and SAC254 values of up to 6.4 (abs m À1 ). A monthly service interval (to change the filter and refill the reagent) appeared to be sufficient to maintain successful long-term operation at these ground water sites. Successful detection of GAL activity demonstrated that the enzymatic range is not limited to GLUC and may be extended to other enzymatic activities of interest for which fluorogenic model substrates are available (Hoppe ; Noble & Weisberg ).

DISCUSSION
Reliable detection of enzymatic GLUC activity as low as approximately 0.01 pmol MUF min À1 100 ml À1 and up to 8.4 pmol MUF min À1 100 ml À1 proved to be achievable under the differing ground water test site conditions. The range of the observed GLUC activity appeared low compared to previously reported ranges, from a few up to 10 6 pmol MUF min À1 100 ml À1 (George et al. ; Farnleitner et al. ; Lebaron et al. ; Ouattara et al. ). However, the determined GLUC activity levels were reasonable, considering the very low to moderate faecal pollution occurring in the investigated ground water sources (Reischer et al. ; Kirschner et al. in press). The automated GLUC activity measurements accurately reflected the different hydrological situations and contamination patterns of the ground water test sites, as revealed by correlation analysis with concurrently determined environmental parameters. Water quality dynamics in the LKAS2 system were driven by discharge and surface runoff conditions and were reflected by bulk parameter changes, as demonstrated by SAK254 (used as a surrogate for dissolved organic carbon input), turbidity (used as a surrogate for particle input) and GLUC activity (used as a surrogate of particle-bound enzyme input). The strong effect of the discharge conditions on the spring water quality of the LKAS2 gas already been demonstrated (Farnleitner et al. , ; Reischer et al. ). Unlike the pronounced water quality dynamics at the karstic LKAS2 site, the alluvial porous PGAW1 site revealed very constant water quality, which was only extremely rarely influenced from the surface, as demonstrated by the measured physicochemical parameters as well as by applied microbiology standards. The situation at the PGAW1 was also accurately reflected by the automated GLUC activity determination, yielding very low and constant values, frequently below the LOQ. No relevant correlations were discernible amongst any of the physicochemical and hydrological parameters at these constant conditions. Only particle-associated enzymatic GAL and GLUC activities revealed a significant association. Although cross-interference between the GAL and GLUC substrate cannot be completely ruled out, the statistical relationship suggests a capacity of particle-associated enzymatic activities to sensitively monitor for surface influence at alluvial porous ground water sites. Automated filter enrichment before performing the enzymatic measurements supports highly sensitive detection for any subtle changes in microbial abundance or biomass in ground water. In this sense, GLUC and GAL activity (or that of any other microbial enzyme, such as esterase or phosphatase) may be regarded as a sensitive process parameter for on-line monitoring of microbial surface influence, complementing the traditional physicochemical parameters (such as SAC254 and turbidity). However, further investigations must be undertaken to evaluate this promising approach in more detail in such habitats.
The observed level of correlation between E. coli and GLUC activity (r ¼ 0.53) at the LKAS2 site was far below the expected level. The results do not indicate that GLUC activity can serve as a surrogate for cultivation-based bacterial standard faecal indicators in such habitats. This observation is in contrast to previously published studies on the indication capacity of GLUC activity, reporting tight associations between E. coli or thermotolerant coliforms and GLUC (Fiksdal et al. ; George et al. ; Farnleitner et al. ). However, it should be noted that previous studies were focused on catchments with recent influences from municipal sewage (human origin). The investigated alpine LKAS2 catchment shows a completely different faecal pollution characteristic because it is mainly influenced by ruminant faecal sources (Reischer et al. ; Farnleitner et al. ). Furthermore, a complex pollution pattern, ranging from old to recent faecal excreta input is expected, according to the prevailing sources and the hydrological runoff situation. The recovered data highlight the strong need for further detailed investigations on the actual faecal indication capacity of GLUC activity at complex catchments, such as the LKAS2 site, in order to evaluate whether GLUC adds useful information for water abstraction management. GLUC activity may be considered a conservative biochemical indicator that can detect culturable, viable but non-culturable (VBNC) and dead bacterial cells of faecal origin (Garcia-Armisen et al. ), leading to low correlations with bacterial standard indicators, especially with respect to aged faecal pollution. However, GLUC activity may also show associations with non-faecal sources, such as with particles from the soil matrix. Indeed, previous investigations reported the existence of non-faecal-associated GLUC activity such as that caused by algae interference (Davies et al. ).
Given the protected ground water habitat at the PGAW1 site, a correlation of GLUC or GAL activities with bacterial standard indicators was not expected. Most of the enzymatic measurements resulted in levels below or close to the LOQ and may be regarded as natural enzymatic background activities in alluvial ground water. Clearly, further basic knowledge on the occurrence and origin of very low-level enzymatic background activities in porous alluvial aquifers is needed to substantiate future monitoring applications.

CONCLUSION AND PERSPECTIVES
The results of this study highlight that automated on-line monitoring devices for microbial or biochemical parameters to support water quality monitoring is a realistic task. The application of such automated field systems will likely become increasingly important for sustainable and proactive water management in the near future (Noble & Weisberg ; Liu & Lay ; Hammes & Egli ; Connelly & Baeumner ). However, one of the basic requirements for such automated systems is its sufficient sensitivity and robustness (Noble & Weisberg ), which is still a major obstacle for many potential on-line sensor and detectors (Liu & Lay ; Connelly & Baeumner ). Nonetheless, increasingly advanced microbial on-line systems will become available, combining automated sampling procedures with efficient target concentration, purification and detection mechanisms (Noble & Weisberg ). It is important to note that many of the developed on-line technologies will be complementary to existing microbiological standard parameters; they are not designed to replace any of the proven surveillance practices. High-resolution online information on microbial water quality aspects can be considered a type of 'process parameter' that continuously monitors a water source or supply system's behaviour for any subtle changes. Such continuous on-line information will also offer the unique opportunity to include 'event-triggered' automated sampling activities. In the case of water quality changes, whether detected by physical, chemical or microbiological probing, samples will then be automatically taken for further standard microbiological analysis in the laboratory. We are thus anticipating an exciting future in which the realisation of water safety plans is based on 'intelligent' systems with a high level of interdisciplinary interaction.