HuNoV GII.4 was obtained from the Catholic University of Korea (Seoul, South Korea). HuNoV GII.4 stock samples were diluted in RNase-free water (Qiagen, Hilden, Germany) and vortexed briefly. The viral supernatant suspension (6.88 ± 0.18 log₁₀ genomic copies/mL) was stored in 500 μl aliquots at −80°C until use. For quantitative analysis, viral genomic RNA concentrations were determined by reverse transcriptase-quantitative polymerase chain reaction (RT-qPCR) analysis, following the protocol described in ISO 15216-1:2017 (ISO 15216-1:2017, 2017), as previously described by Lee et al. (2018). The entire process of preparing the viral stocks was carried out with strict adherence to biosafety considerations (Mariita et al., 2022). The entire process of preparing the viral stocks was carried out with strict adherence to biosafety considerations (Mariita et al., 2022).

Brine was collected daily for 5 days from a kimchi-manufacturing company located in Gwangju, Korea. The microbial and physicochemical properties of the collected brine samples were analyzed and their characteristics were profiled. All microbial and physicochemical parameters were selected according to standard methods for examining brine water (APHA, 2019). The brine water profiling was based on eight parameters, namely the pH, nephelometric turbidity units (NTU), chemical oxygen demand (COD), biochemical oxygen demand (BOD), dissolved oxygen (DO), total coliform (TC) level, Escherichia coli level, and HuNoV level. The salinity of all experimental brine water samples was 14 ± 1.1%. Samples were denoted according to different NTUs (8, 16, 24, 32, 40, 48, and 56 NTUs), which corresponded to the number of reuses. KimchiTown Co. (Gwangju, Korea) kindly provided used brine water samples from seven groups (numbered according to their consecutive number of reuses; 0, 1st, 2nd, 3rd, 4th, 5th, and 6th), which had been previously evaluated as negative for HuNoV GII.4 based on the ISO 15216-1:2017 method. Each experimental setup was designated as one experimental batch, and 300 ml of brine water samples artificially inoculated with HuNoV GII.4 was prepared for each experimental batch. HuNoV GII.4 was resuspended in 300 ml of each brine water sample (denoted as 0, 8, 16, 24, 32, 40, 48, and 56 NTU samples), and 1 ml stock suspension of HuNoV (containing 6.88 ± 0.18 genomic copies) was inoculated individually into 300 ml of each batch (used brine water samples), corresponding to approximately 6.90 log₁₀ genomic copies/batch of HuNoV GII.4.

In this study, UVC LED irradiation (peak wavelength of approximately 275 nm) treatment was conducted to inactivate HuNoV GII.4 in used brine water. Electronic printed circuit boards (PCBs), based on UVC LED modules (SeoulViosys Co., Seoul, Korea) corresponding to the peak wavelength (approximately 275 nm), were utilized (Figure 1). UV LEDs with emissions at 275 nm and an optical power output of 60 mW at a current of 30 mA and voltage of 35 V were used. For the UVC LED treatment, the PCB connected to the LED-UVC was fixed on the top of the frame, and each 300 ml brine water sample was placed 5 cm away. The brine water samples that were inoculated with HuNoV were divided into two groups. One group was subjected to UVC LED treatment without stirring. The other group was treated with a UVC LED while stirring at 100 rotations/min (rpm) using an impeller (Supplementary Figure S1). The intensities of the UVC LED modules were measured with a spectrometer (AvaSpec-ULS2048-USB2-UA-50; Avantes, Apeldoorn, Netherlands). Using this experimental setup, the modified irradiance intensity of the UVC LED was 20.27 ± 1.17 μW/cm². The treatment dosages were determined based on preliminary experiments. The inoculated brine water samples were treated at dosages of 0, 5, 10, 15, and 20 mJ/cm² for all inactivation treatments.

To analyze the inactivation kinetics of UVC LED treatment against HuNoV GII.4 in used brine water following irradiation, the resulting data were analyzed using the GInaFIT software (version 1.6), a freeware tool (Geeraerd et al., 2005). The inactivation plots fit well with the log-linear model (Eq. 1) and the inactivation rate constant (k_inact) was calculated (Eq. 2).

where Log₁₀R is the infectivity reduction on a log₁₀ scale; N is the infectious HuNoV titer (log₁₀ genomic copies/μL) after UVC LED treatment; No is the initial infectious viral titer (log₁₀ genomic copies/μL); k_inact is the pseudo-first-order inactivation rate constant at a given inactivation dose (mJ/cm²). The maximum inactivation rate constant (k_max) was calculated using a specific dose (D) of UVC LED irradiation, expressed as the mJ/cm².

The induced k_inact was calculated as a function of turbidity using SigmaPlot (version 14, San Jose, CA, USA). The interrelationship between the sample turbidity and inactivation rate was used to illustrate the exponential one-phase inactivation kinetics of HuNoV (Eq. 3). The developed regression model was combined with a logistic linear model to predict HuNoV inactivation as a function of turbidity (NTU) and the UVC LED irradiation dose (D).

where T_NTU is the NTU of a brine water sample, k₀ is the k_inact at an NTU of 0, a is the k value at infinite turbidity, and b is the rate constant. From the above regression equation, the following equations were derived to predict HuNoV inactivation as a function of the T_NTU and UVC LED D values.

For HuNoV GII.4, with UVC LED irradiation, but without shaking (0 rpm),

For HuNoV GII.4 with the UVC LED irradiation and shaking (100 rpm),

We investigated whether the developed model could predict HuNoV GII.4 inactivation based on turbidities other than those used experimentally. The turbidity of the brine processing water samples, which were provided from KimchiTown Co. (Gwangju, Korea), for further experiments was 28 and 36 NTU. The turbidities were within the range previously used to develop the model. The accuracy factor (A_f) and bias factor (B_f) were used for validation purposes and were calculated using the following equations:

where Y_model is the modeled value, Y_obs is the observed value, and n is the number of experimental replicates. The root-mean-square error (RMSE) of the mathematical kinetic model prediction with respect to the estimated variable Y_model was defined as the square root of the mean squared error (Eq. 6):

where Y_model is the modeled value, Y_obs is the observed value, n is the number of observations, and k is the number of estimated parameters in the model. The residual sum of squares (RSS) was determined as the sum of the squared distances between the predicted values and empirical data:

where Y_model is the modeled value, and Y_obs is the observed value. The Akaike information criterion (AIC) formula (Eq. 8) was used for the sum of squares optimization:

where n is the number of observations, k is the number of estimated parameters in the model, and SS is the sum of the squares.

Duplicate samples were tested for each brine water sample, and the experiments were repeated in triplicate. For statistical analysis, one-way analysis of variance was used to compare differences among mean values using the SPSS Statistics software (version 19, IBM Corp., Chicago, IL, United States). The threshold for statistical significance was set at p < 0.05. The experimental results were determined as log₁₀ genomic copies/μL, and regression analysis was performed using the SigmaPlot software (version 14.0).

To evaluate the efficiency of MBS/PMA/RT-qPCR, we first investigated the recovery rates from processing water artificially inoculated with HuNoV GII.4. The mean recovery rates of HuNoV GII.4 between the RT-qPCR method alone, as a negative control, and PMA/MBS-qRT-PCR were compared. The mean recovery quantities of HuNoV determined using the MBS/RT-qPCR and MBS/PMA/RT-qPCR assay were 91.74 ± 1.3% and 88.4 ± 2.5% for NoV (Supplementary Figure S2). The use of 200 μM PMA had no significant effect on the quantitation of HuNoV GII.4 virus particles (p > 0.05).

The microbial and physicochemical properties of the collected brine samples were analyzed, as shown in Table 1. As the frequency of brine use increased, the pH and DO values decreased significantly, whereas the turbidity, BOD, and COD increased significantly. As the number of uses increased by one, the turbidity increased proportionally. In terms of microbiological changes, E. coli and HuNoV were not detected in any of the samples, and TC was detected at a level of approximately 2 log₁₀ colony-forming units (CFUs)/mL in brine water used consecutively for five or six times.

The virucidal effect of UVC LED irradiation was affected by increased turbidity when the impeller system was not used to stir the samples (0 rpm; Table 2). Regarding the virucidal effect of UVC LED irradiation in used brine water samples, the HuNoV concentrations after UVC LED treatment at 5 or 10 mJ/cm² showed maximum decreases of 3.44 and 3.76 log₁₀ genomic copies/μL, respectively, in 8 NTU water samples. Overall, as the brine water turbidity increased, the reduction in HuNoV caused by UVC LED treatment decreased. This was especially true at the maximum turbidity (56 NTUs), where the water samples contained 0.37 or 0.51 log₁₀ genomic copies/μL after exposure to 5 or 10 mJ/cm² UVC LED, respectively.

We anticipated that suspended solids could interfere with the virucidal activity by reducing the transmittance of UVC LED irradiation. Surviving populations of HuNoV GII.4 (log₁₀ genomic copies/μL) after 15 or 20 mJ/cm² UVC LED treatment were not detected in the 8, 16, and 24 NTU brine water samples, whereas the HuNoV GII.4 levels in 56 NTU brine water samples were detected by 2.40 (at 15 mJ/cm²) and 2.87 (at 20 mJ/cm²) log₁₀ genomic copies/μL. Interestingly, the virucidal activities of UVC LED treatment were higher when using the impeller system (set at 100 rpm) than when the impeller system was not used (Table 2). In particular, 10 mJ/cm² UVC LED treatment inactivated HuNoV GII.4 in brine water samples at an NTU of up to 32. Furthermore, HuNoV GII.4 in the 56 NTU samples was not detected after UVC LED treatment at ≥20 mJ/cm² with impeller system set at 100 rpm. Overall, the HuNoV values were reduced significantly more by UVC LED treatment with the impeller system (100 rpm) than without the impeller system, with all tested brine water samples. UVC LED disinfection can be sensitive to turbidity (Gullian et al., 2012; Carré et al., 2018). Gullian et al. (2012) demonstrated that an increase in turbidity in water samples from 10 to 16 NTU resulted in a 60% decrease in the antimicrobial activities of UVC LED treatment. Carré et al. (2018) verified that the antimicrobial activity of UVC LED irradiation was maintained without significant changes when the turbidity was below 5 NTU.

Moreover, the exposure frequency of disinfecting UVC LED light sources and UVT of the water sources also affect the virucidal efficacy. As the turbidity increases and UVT decreases, there is a limit to how deep a UVC LED light source can penetrate water, which is an important factor in sterilization. In particular, understanding the UVT of the processing water source is an important factor to ensure that the treated UVC LED dose is sufficient to inactivate pathogenic viral particles (Jarvis et al., 2019). In this regard, our experimental results demonstrate that the water stirring system significantly improves the UVC LED-disinfection levels caused by turbidity. UVC LED treatment is an energy-based disinfection process, where the microbial inactivation efficiency is determined by various parameters including the treated environmental condition and the applied dose of the UV light source (Gidari et al., 2021). Theoretically, the dose (mJ/cm²) of UVC LEDs is determined by the exposure time (s) multiplied by the intensity delivered to microbial cells (mW/cm²). Therefore, for UVC LED treatment-induced inactivation, the UV dose (rather than the exposure time) is the most precise measure of the efficacy of disinfection (Nyangaresi et al., 2018). Thus, regarding virucidal mechanisms based on UVC LED treatments, fatal damage to microbial cells occurs over a UVC LED range because the intercellular components of microbes (such as RNA and capsid proteins) can sensitively absorb UVC photons. Damage to viral proteins and/or viral nucleic acids is well known as a general mechanism of virus inactivation by UVC LED disinfection (Wan et al., 2020). For example, a UVC LED light source can disintegrate the viral capsid protein of murine norovirus-1 (Park et al., 2016). UVC LED treatment caused oxidative damage to the viral capsid protein, which was linked to reduced infectivity with HuNoV (Sano et al., 2010; Tanaka et al., 2018) and bacteriophage MS2 (Rule Wigginton et al., 2010). Poliovirus inactivation has been demonstrated to occur secondarily to viral protein–genome crosslinking induced by UVC treatment (Wetz and Habermehl, 1982). UVC has been shown to inactivate severe acute respiratory syndrome coronavirus 2 by destroying viral nucleic acids (Gidari et al., 2021).

Viruses are frequently classified as having single-stranded DNA, double-stranded DNA, single-stranded RNA, or double-stranded RNA (Kowalski, 2009). Such viruses show different susceptibilities to UV irradiation; therefore, virucidal effects vary depending on the type of genome involved. Generally, single-stranded viruses, including HuNoVs, are more sensitive to UVC LED treatment because of a lack of redundant genetic information in the second strand, which enables double-stranded viruses to repair the damage (Tseng and Li, 2005). In contrast, non-enveloped viruses (including HuNoVs) typically show more UVC resistance than enveloped viruses because lipids and capsid proteins in an envelope can be destroyed more easily than other parts of viruses (Perlmutter and Hagan, 2015). In general, inactivation of pathogenic microorganisms occurs in the UV wavelength range of 250–280 nm, and the maximum effect has been confirmed around 265 nm (Chatterley and Linden, 2010). Therefore, previous studies reported that relative germicidal effectiveness of UV-C energy based on microbial DNA damage depends on various wavelengths (Bintsis et al., 2000; Hijnen et al., 2006). Moreover, many factors influence the germicidal effect of UVC LED, and the main factor is the operating temperature of LED chips. As electric energy is applied to the LED, it is converted into heat and light. The heat generated increases the LED junction temperature, reducing the light output, which is a major factor in the sterilization effect. Therefore, it is important to properly manage the operating temperature generated by the UVC LED irradiation (Arques-Orobon et al., 2020).

For quantitative analysis, we performed molecular biology assays, unlike the cell culture technology used by other researchers. The overestimation of infectious viruses after disinfection treatment and the inability to differentiate between inactivated and infectious viruses is considered a main disadvantage of RT-qPCR (Karim et al., 2015). Therefore, the developed method, such as the MBS/PMA/RT-qPCR assay, could allow for more precise quantification of potentially infectious viral particles in environmental samples after disinfection (Lee et al., 2018). In particular, Karim et al. (2015) demonstrated that pretreatment with the intercalating dye PMA combined with RT-qPCR assay could potentially be used with all non-culturable or fastidious viruses, although specific conditions of inactivation must be investigated.

With the inactivation regression model of HuNoVs treated with UVC LED, we observed that increased turbidity in the used brine water samples decreased the virucidal effect of UVC LED treatment. The most noteworthy and interesting findings of this study are presented in Figure 2. Figure 2 presents the inactivation curves, which were drawn using a log-linear model. We confirmed that operating the impeller system during UVC LED treatment increased the virucidal effects obtained when disinfecting turbid samples. Our results indicated that the k_inact increased by approximately 2.15-fold when using the impeller system at 100 rpm with the 8 NTU samples (lowest turbidity studied) and that the k_inact increased by 1.69-fold even in the 56 NTU samples (highest turbidity studied). Importantly, the test group exposed to UVC LED with the impeller system set at 100 rpm showed a tendency toward a delayed decrease in the k_inact (Figure 2). Furthermore, the developed model can be used to predict the virucidal effect of turbidity-dependent UVC LED disinfection with brine water samples. In this regard, the developed model can be used to determine the processing conditions when considering the titers of target viruses, the water-sample turbidity, and the UVC LED dose. Universally, the k_inact has been mostly used as a mathematical parameter to explain pathogenic microorganism inactivation by UVC disinfection and has also been applied to compare the resistance of pathogenic microorganisms to UVC disinfection (Goldman and Travisano, 2011; Viana et al., 2013). In cases where the resistance to UVC LED treatment is high, the k_inact appears to be low, and in general, k_inact values appear to be significantly lower for infectious viruses than for pathogenic bacteria (Weiss and Horzinek, 1986).

The interrelationship between the k_inact and brine water sample turbidity was determined mathematically by regression analysis, and the fitness of the regression models agreed well with the exponential one-phase decay model (Figure 2). The five model-selection criteria used to determine the correspondence between the two models are presented in Table 3. The better the fit of the developed model, the lower the RMSE, RSS, and AIC. Thus, the lowest value obtained from the model-selection criteria implied a reliable predictive model.

With the model developed using the UVC LED and impeller system, the model-selection criteria (including the RMSE, RSS, and AIC based on goodness of fit) demonstrated better goodness-of-fit than the model without the impeller system (Table 3). Both models provided suitable fits based on small RMSE, RSS, and AIC values with both types of UVC LED treatment (i.e., with/without an impeller system). In terms of the A_f and B_f values, an A_f and B_f of 1 indicates an absolute agreement between the observations and predictions. An A_f value above 1 represents a less accurate average estimate, and generally the farther the B_f value is from 1, the more a model underpredicts or overpredicts the experimental data. The virucidal effects observed using UVC LED treatment without the impeller system were not a good fit with the predicted values, with A_f values ranging from 1.23 to 1.31 and B_f values ranging from 0.74 to 0.87. The data obtained after UVC LED treatment with an impeller system showed model fittings, with A_f values ranging from 1.08 to 1.12 and B_f values ranging from 0.88 to 1.07. These statistical analysis values demonstrate that UVC LED treatment conditions should be approached conservatively to inactivate HuNoV GII.4. Comprehensively, it is necessary to consider several additional factors because our statistical analysis of the virucidal effects against HuNoV was only based on water sample turbidities. For instance, various viral pathogens are necessary for additional evaluation because of their fundamental variations. In addition, it is necessary to examine correlations between changes in the virucidal effect and other parameters (beyond the NTU) such as the temperature, BOD, COD, and microbacterial communities, and it is necessary to determine specific conditions for the most efficient UVC LED treatment. In conclusion, a statistical technique using predictive modeling can predict the sterilization effect of UVC LED treatment as an approach suitable for safe disinfection of brine for reuse in the future.