Virus removal vs. subsurface water velocity during slow sand filtration

At the end of the 19th century, when the groundwater level of rapidly expanding cities began to decline due to increasing demand for potable water, replenishing groundwater by infiltrating surface water into the underground started to be implemented as a solution (Richert 1900; Schübel 1936).

In the beginning, no experimental evidence was available for setting infiltration conditions – soil quality, water velocity and residence time of the water underground – that would produce an acceptable microbial quality of the exfiltrate. Experimental studies were then set up for gaining information on microbial removal (Table 1). As a result, 40–100 days underground residence time was recommended by Gross (1929) and 60 days by Knorr (1951b). Finally, Germany adopted 50 days residence time as sufficient for removing pathogens (DVGW 2006). As of 2006, numerous countries have adopted minimum residence time for groundwater recharge and riverbank filtration (RBF), mostly around 50 days (Chave et al. 2006). No rules have been issued with respect to the pore water velocity (PWV), i.e. of the interstitial velocity of the water underground.

Acording to the colloid filtration theory (CFT) (Yao et al. 1971; Tufenkji & Elimelech 2003; Tufenkji 2006), removal of colloids by soil filtration is more efficient when a lower PWV is chosen. A number of observations corroborate this prediction for viruses. Funderburg et al. (1981) found that removal of the phage φχ174 was directly proportional to the retention time in columns. Vaughn et al. (1981) did not find any retention of poliovirus at velocities of 25 m/d, yet the retention at 1 m/d was substantial. In columns percolated with suspensions of Poliovirus 1 and Echovirus 1, Wang et al. (1981) established a direct relationship between virus removal and column retention time, a relationship that fitted well with the predictions of the CFT (Schijven & Hassanizadeh 2000). Pang (2009) found a strong inverse relationship between the removal of viruses and PWV.

In a number of cases, even low PWVs yielded rather poor virus removal. At PWV of 0.33–0.56 m/d, Van der Wielen et al. (2008) observed adsorption-removal rates of 0.02 and 0.07 log₁₀/d, respectively, for the phages MS2 and φX174 (0.06 and 0.08 for adsorption plus inactivation-removal, respectively). Pang (2009) has reviewed the removal rates (log₁₀/m) of 86 viruses, including phages, reported in the literature. Among experiments conducted in sandy aquifers, two out of five would have undergone a removal of less than 5 log₁₀ over 50 days. In gravel aquifers the ratio was six out of 24, and in karst aquifers one out of nine. Such removal degrees are deemed insufficient if RBF was the only pathogen-removing procedure to obtain drinking water from raw water (WHO 2008). It is therefore important to find conservative estimates of the relationship of PWV to virus removals since in many locations underground passage of surface water is now the only procedure for eliminating pathogens to a tolerable level, for example, in Berlin (Grohmann & Petersohn 2000).

The present study had the following goals:

• To isolate a phage with a long survival time and a small adsorption tendency to coarse sand as a poor adsorbent matrix. This phage was to be used to establish a quantitative relationship between its removal by filtration in coarse sand and the PWV used during infiltration.

• Using observed removal data, a mathematical model was to be constructed to estimate PWV-dependent virus removal expected after underground passage of 50 days.

Clay-free, coarse sand used in all experiments has been described elsewhere (Bauer et al. 2011). Tap water of the city of Berlin was used for the percolation. This water is not chlorinated and had a conductivity of 705 μS/cm and a pH of 7.3.

Among many others phages, the phage strain 241 was isolated from the Teltow River, a fecally polluted stream in Berlin. One of the two criteria for its selection was that it should have long survival in a water-saturated sandy environment. To test for survival, the isolates, raised in Escherichia coli strain WG5, were suspended in sterilized water-saturated sand, 50 mL of the slurry was placed in a column, and the column outlet plugged. At different times, up to 1 mL of gravity-obtained eluate was collected and the phage concentration determined. The second criterion, low adsorption to coarse sand, was estimated by sorption isotherm experiments. A 10 mL phage suspension containing approximately 10³ phages/mL was incubated with 1 g coarse sand at room temperature in a shaker. After 2 hours incubation, the phages remaining in the liquid phase were determined. The sand type used was identical to the one used in the column experiments presented in this paper. Among 61 isolates tested for survival and adsorption, coliphage 241 was chosen for its long survival in the sandy environment and low adsorption to coarse sand. The long survival should mean that the removal observed after its passage through sand columns was attributable mostly to adsorption to the sand particles. The low adsorption tendency should ensure that the results and conclusions obtained with this phage were conservative with respect to other viruses.

For the analysis of coliphage 241, USEPA Method 1601 (2001) with the E. coli strain ATCC700078 as indicator strain was used, but instead of taking the agar formula recommended there, double-concentrated modified Scholtens-Soft-Agar (DIN-EN-ISO-10705-2 2002) was added to the sample before plating. The detection limit of the method was 1 plaque-forming unit (pfu)/100 mL.

Three different series of column experiments were carried out. The experiment series, which differ with respect to column length, PWV, and phage input, are outlined below.

Six acrylic glass columns with a diameter of 15 cm and a length of 120 cm were filled with a slurry of the sand described above. After upwards saturation, tap water was percolated top down at a filtration velocity of 0.30 m/d corresponding to a pore velocity of 1 cm/d for 2 days to stabilize the hydraulic conditions. The columns were connected in series with silicon tubing of an internal diameter of 4 mm, so as to allow top down flow in all columns. Hollow needles perpendicularly placed into the column walls and ending at the length axis of the column allowed sampling thanks to small faucets located at the outer extreme of the needles. Sampling was also carried out at the outlet of each column. In this way, the following filtration distances were monitored daily: 30, 50, 70, 90, 120, 180, 240, 300, 360, 480, 600, and 720 cm. Sampling volumes of 100 mL were taken slowly, each sampling taking 30–40 minutes in order to avoid too great a disturbance of the columns' flow rate. The reservoir containing the phage suspended in tap water was located approximately 4 m above the entrance of the columns and was connected with a silicon hose with the first column. The PWV of 1 m/d was maintained by the hydraulic pressure exerted by the reservoir and was regulated by a peristaltic pump at the outlet of the last column. The pump's function was not to increase, but to slow down the flow rate imposed by the hydraulic pressure. The phage concentration in the reservoir was monitored daily and a concentration of 10⁵–10⁶ pfu/100 mL was maintained, adding more phages if required.

The normality of the groups of values for each filtration distance was checked using three measures: skewness and two normality tests – Kolmogorov–Smirnov (K–S) and Shapiro–Wilk (S–W). In addition, the log₁₀ transformed values were checked in the same manner. The K–S and S–W test results showed a statistically significant departure from normality using both the raw and transformed data (p < 0.05). However, the log transformed values showed greatly reduced skewness (0.052 vs. 5.622). This indicates that log transformation does improve the normality, which was also evident in corresponding normal-probability plots. Thus, the log transformed data showed a better approximation to normality than the untransformed data. The subsequent analysis for significance was carried out with the log transformed values using the Tukey multiple comparison test.

The inlets of acrylic glass columns with a length of 120 cm as described for the previous experiment, filled and treated in the same manner, were connected to a reservoir containing a phage suspension. The columns were run in parallel. Different PWVs (4, 2, 1, 0.5, 0.25, and 0.1 m/d) were adjusted with peristaltic pumps at the outlets of the columns, and sampling was carried out daily. In the first series of the experiment, the duration was 15 days (Experiment A in Table 2 for PWVs of 4, 2, and 1 m/d). In the second series, the duration was 28 days (Experiment B in Table 2 for PWVs of 1, 0.5, 0.25, and 0.1 m/d). To keep the same sample size for all PWVs within one series of experiments, we took the single observations from days 4 to 15 in Experiment A (12 values) and from days 14 to 28 in Experiment B (15 values) for calculation of the average removals.

After the experiment for estimating the dependence of the phage removal on the PWV was finished, the columns were run without phage input at a PWV of 2.6 m/d with tap water over 26 days. The exfiltrates were collected in a pool every 24 hours, and the phage concentration in these pools was determined. The number of plaque-forming units eluted was calculated from the concentration in the 12 L pools collected daily. During Saturdays and Sundays the rinsing was stopped and the columns were held under water-saturated conditions, so that the active rinsing period was 18 days.

The removal of biogeochemical components during bank filtration has been the subject of several modeling studies. An example study is presented by Holzbecher (2006) and Holzbecher et al. (2006) showing the interaction between dispersion and degradation processes. Here we focus on adsorption as the major process for the removal of phages during sand filter passage.

Chatterjee & Gupta (2009) have presented evidence for nonlinear removal if the colloid is present in aggregates of different sizes. According to Redman et al. (2001), even mono disperse suspensions of colloids may consist of different populations with sufficient heterogeneities, in size or electrostatic charge, to cause different adsorption kinetics (Redman et al. 2001). In these cases the filter coefficient λ of the colloid suspension is not a fixed value, but follows a distribution. Tufenkji & Elimelech (2003) have analyzed several distribution alternatives and found that the power law distribution fits best to many published results of adsorption kinetics.

For we use the following functions:

• normal distribution with mean value and standard deviation σ;

• bimodal distribution with a low value k_low and a high value k_high, and corresponding standard deviations σ_low and σ_high, and fractions f_low and f_high;

• power law distribution with coefficient A and exponent −b: ⁠.

The computations, described in the following ‘Results and discussion’ section, based on Equation (5), were performed using MATLAB^®.

Phage 241 was chosen from 61 strains isolated following the procedure mentioned in ‘Material and methods’. This phage had the most appropriate combination of long survival and poor adsorption to coarse sand. The particles of strain 241 were round, with a diameter of c. 25 nm, and had no tail (Figure 1, inset). When suspended in a slurry of sterile sand and tap water, the phage concentration slightly increased during the first day presumably due to a disintegration of clumps, and afterwards started to decay. The decay rate then became approximately linear on a log scale, and the decay time (T₉₀), expressed as the time elapsed for a decrease of one log₁₀ unit, was calculated to be 40 days. This phage was used for all further experiments to obtain conservative estimates of removal by sand filtration.

A suspension of phage 241 was applied to the inlet of six columns connected in series at a PWV of 1 m/d over 39 days and samples were taken daily. The average concentrations of phage 241 found at different infiltration distances are shown in Figure 1 (circles). The average concentrations were corrected for the decay of the phages during the infiltration period, assuming the T₉₀-value of 40 days obtained in the survival experiment. To make sure that the concentration measurements at all distances had the same number of single observations, the averages of Figure 1 were built starting with the values obtained 7 days after initiation of the experiment. Consequently, each average consists of 32 single observations (30 at distance of 240 cm).

The phage showed an initial removal of approximately 1.2 log₁₀ during the first 120 cm (corresponding to approximately 1 log₁₀/m) of infiltration. During further infiltration the removal was much lower, amounting to approximately 0.72 log₁₀ between 120 and 720 cm, or 0.12 log₁₀/m (Figure 1). However, according to the statistical analysis outlined above, the averages at filtration distances greater than 90 cm were not significantly different among themselves (p > 0.05), with the exception of the value at 120 cm vs. 300, 600, and 720 cm, respectively (p < 0.05).

A fast initial rate of viral removal, followed by a slower rate at larger distances of the filtration path, has been frequently reported (Schijven & Hassanizadeh (2000) and the literature therein). Work by Schijven (2001) showed that soil heterogeneities might be responsible for this effect, but in our case the columns were used only 2 days after they were packed and it is therefore very improbable that build-up of heterogeneities in the columns was the cause of the nonlinear behavior of the phage removal. The removal results of this experiment were used for constructing the model below.

The removal of the phage 241 in columns run at different PWVs is shown in Figure 2. Removal after 1 m infiltration was 0.6 log₁₀, at a PWV of 4 m/d. Decreasing the PWV to 1 m/d did increase the removal degree, but only moderately, from 0.6 to 0.98 log₁₀ units. Further decrease of the PWV from 1 to 0.1 m/d caused a considerable increase of the removal from 1.01 to 6 log₁₀ units (Table 2). As the columns were 1 m long, the column run at a PWV of 0.1 m/d yielded phage concentrations no sooner than 10 days after the start of the experiment. Even after this time, the exfiltrate concentrations of the phage were below the detection limit (1 pfu/100 mL) in more than half of the samples. Therefore, at 0.1 m/d only a tentative log₁₀-removal, averaging 6 log₁₀ was estimated, by considering all concentrations below the detection limit as 1 pfu/100 mL, compounding all concentration values between days 14 and 28 and dividing them by the total volume of all samples. This calculation probably underestimates the true removal.

First we examined the 6 m column experiment, in which a velocity of 1 m/d was used. Using all averages (circles) shown in Figure 1, removal curves were simulated after applying all three distributions assayed – Gaussian, bimodal, power law – to p(λ). All three combinations used showed overestimations of the removal after 2 m infiltration (not shown) when compared with the empirical results. To obtain more conservative removal predictions at infiltration distances beyond 2 m, we fitted the parameters using the points symbolized with empty circles in Figure 1, i.e. the measurements obtained at larger infiltration distances. The results of the parameter estimation from this calibration are given in Table 3. The coefficient A in the power law distribution can be calculated from s by ⁠. The best-fit curves for the three distributions are plotted against measurements in Figure 1. The least squares residuals, listed in the last column of Table 3, show that the bimodal distribution is the nearest approach.

The calibrated model was first used for a theoretical evaluation of velocity-dependent removal vs. distance. Velocities were selected in the range between 0.25 and 8 m/d. We show the results for the bimodal distribution in Figure 3.

In a second step, the model, calibrated for the velocity of 1 m/d, was used to compare modeled results at different PWVs with the results experimentally shown in Figure 2 and Table 2. Figure 4 shows the results for the bimodal distribution of the removal. The model makes acceptable predictions for velocities equal or higher than 1 m/d, but for velocities lower than 1 m/d the model grossly underestimates the removal. While this situation suggests that the model needs refining in order to explain the underestimation of removal at PWVs smaller than 1 m/d, the model's predictions are conservative and confer an extra margin of safety.

The curve for the bimodal distribution in Figure 5 shows that for a PWV of 1 m/d, the infiltration would remove viruses by a factor of approximately 10⁻⁴. Higher PWVs result in less removal.

This result is approximately in line with the findings of Van der Wielen et al. (2008) if their results are converted as removal degree per day and extrapolated to a 50-day residence time.

WHO (2008), under the assumption of a concentration of 10 virus/L in the raw water (with rotavirus as a model organism), recommends (for high income countries) drinking water treatments reduce the viral concentration by a factor of 10^−6.495. The resulting concentration of 10^−5.495 virus/L corresponds to a disease burden of 10⁻⁶ disease adjusted life years per year, if 1 L water is consumed per day (WHO 2008). According to the results of the bimodal distribution shown in Figure 5, slow sand filtration (SSF) would not meet WHO requirements if carried out as the sole water treatment procedure. Either the underground residence time of the water would have to be extended beyond 50 days, or additional water treatment procedures would have to be implemented. Alternatively, the concentration of viruses in the surface water taken for infiltration would have to be reduced below 0.1 virus/L.

However, the removal degrees shown in Figure 5 have to be considered as a conservative estimate because other causes of removal (e.g. content of clay materials of the soil, known to bind viruses much better than sand (Sobsey et al. 1980), predation or enzymatic degradation by fungi or bacteria present in the soil) have been purposely avoided in the experimental design of this paper, and therefore have not been considered in the model. The removal of viruses by RBF carried out in a natural environment is probably much more efficient than that observed here with coarse sand free of clay, and with water poor in nutrients taken for elution. Sprenger et al. (2014) has followed the removal of indigenous somatic coliphages during bank filtration in the river Yamuna in Delhi (India). They found that the phages were removed by 3.3 log₁₀ after only 1 m infiltration and by 4.31 log₁₀ after 3.8 m. After approximately 119 days infiltration to a distance of 50 m the removal factor was higher than 10⁶. In that study the soil contained 1% of clay minerals. Natural microbial flora in the river bank presumably also had a positive influence on the removal of the phages. However, when artificial sand filters are operated for the first time, in which a soil flora is not yet present, the limitations of removal shown in Figure 3 should be taken into account.

An important question during SSF and RBF concerns the detachment of viruses associated to the soil particles, especially when the PWV increases and re-adsorption of the released virus is slowed down or absent. A fraction of the viruses hitherto adsorbed to soil might then be released and be eluted with the filtrate. Derx et al. (2013) have simulated PWV in river banks during raising of the river water level due to flooding. They concluded that removal of the viruses might be seriously impaired during flooding due to higher PWV, which might rise to 25 m/d.

We have quantified the escape of phage 241 into the liquid phase if PWV increases above 1 m/day, a PWV above which phage attachment is only moderate (Table 2). To achieve this, the columns of Figure 2(b), which had been run with a phage suspension over 18 days, were subsequently run at a constant PWV of 2.6 m/d without addition of phages over another 18 days, and the escaped phages in the exfiltrate were determined.

The average percentage of phage 241 released per day was 0.0026% of the phages attached to the sand (Table 4). According to this small percentage, the number of phages attached to, and the daily rate of phages detached from, the columns remained practically constant over the 18 days of the experiment (Figure 6).

The data allow estimation of the detachment constant governing the release of the phages from the sand, applying the equation: dn/dt = k_det × No, where dn/dt represents the number of phages released per day and No the total number of phages adsorbed to the sand. Values for k_det of 8.3 × 10⁻⁶ to 7.1 × 10⁻⁵ were obtained, depending on the column, yielding a mean for the four columns of 2.6 × 10⁻⁵ (Table 4). Consequently, if a sand filter, or a river bank, hosts 10⁵ sand-associated viruses, only approximately 2.6 viruses/day will be released into the liquid phase.

Other studies have also found attached viruses to be very slowly released (Bales et al. 1997) and generally the attachment constants to be much higher than the detachment ones, at least in the first meters of infiltration (Schijven & Šimunek 2002; Schijven et al. 2002). This situation allows the design of strategies for minimizing viral breakthrough during sudden increases of PWV or even for the operation routinely followed in RBF water works. Generally it will consist of halting the abstraction of water from the wells for a certain period of time in order to reduce or stop altogether the PWV, allowing even viruses with a low attachment constant to become associated with the soil particles. Resuming the abstraction rate will allow only a very small percentage of the viruses to reach the wells.

Conservative estimates predict virus removal by approximately a factor 10⁴ after 50 days infiltration in sand void of clay at a PWV of 1 m/d. At higher PWVs the removal degree is lower. Removal can be acceptably simulated by a mathematical model assuming that the adsorption coefficient of the viruses follows a probability distribution. At PWVs lower than 1 m/d, virus removal might increase dramatically, but simulation by the model underestimates the removal. While the results presented indicate that the model needs refining in order to explain the underestimation of removal at PWVs lower than 1 m/d, the model's predictions are conservative and confer an extra margin of safety.

As the detachment constant k_det has been found for this particular phage – and has been found by others for other viruses – to be very low, an intermittent operation of the water abstraction in RBF or SSF plants, with temporal interruptions of water abstraction, is expected to effectively allow attachment of viruses during the standstill periods. Resuming water abstraction will allow detachment of only a very small fraction of the viruses.

The authors thank Dr Rudolf Lutz, Max-Planck-Institut für Molekulare Genetik, Berlin, for the electron microscopy work and Dr Mark Wyer, Aberystwyth University, for the statistical analysis. This investigation was carried out as a contribution to the Natural and Artificial Systems for Recharge and Infiltration project (NASRI project) with the financial support of the ‘Kompetenzzentrum Wasser Berlin’ (KWB).