Metal Binding Reactor (MBR)

The second component, of our applied design is the metal binding reactor (MBR). The metal binding reactor is a piece of cylindrical housing apparatus designed to contain our genetically modified bacteria. The water, now free of large particulates, is fed into the top of the MBR. It travels down through the central pipe, and slowly rises back through the outer pipe, in which the bacteria are contained. The E. coli are grown on polypropylene scaffold torus structures (plastic pan scourers), using a chlorhexidine gluconate surfactant (peppermint flavoured mouthwash) which stresses the bacteria and promotes the development of a biofilm. Here, we have optimised the flow rate to ensure that the bacteria have a high probability of binding to the targeted metal ions. Additionally, we have created a mathematical model to inform the scalability of this aspect of the filtration system as it is unique to our project. Click here to find out more about our model.

As a proof of concept experiment, we tested the ability of type I pili in E. coli MG1655 to bind to mannose in a metal binding reactor (MBR), as this could be investigated at the same time as our metal binding pili were being developed. The bacteria formed biofilms on polypropylene torus scaffold structures within the MBR, and water containing a known concentration of mannose was passed through the reactor. The concentration of mannose removed from the water was measured using High-Performance Liquid Chromatography (HPLC). This would show whether this design of reactor is a feasible choice for our purpose.

Ideally, we would have tested the metal binding performance of our modified strains of E. coli in the MBR, but due to time constraints this was not possible.

Figure 1: Metal binding reactor (MBR) containing polypropylene torus scaffold structures, on which E. coli MG1655 formed biofilms.

Key Results from Design of Experiment (DOE)

In order to optimise the parameters of the MBR, we used a Design of Experiment (DOE) approach. DOE is a systematic, statistical method which allows the relationship between factors affecting a process and the output of that process to be determined. In a more traditional One Factor at A Time (OFAT) approach to optimisation experiments, factors are examined individually, with optimisation of one factor needing to be complete before a second factor can be examined. In contrast, in a DOE approach multiple factors are altered simultaneously, allowing the researcher to look at interactions between factors and making the optimisation process more efficient.

In this case, we used DOE to optimise the MBR to remove the highest amount of mannose possible. JMP software was used for all DOE work. We used a definitive screening design with 10 runs. Table 1 lists the experimental runs conducted and the results. Two responses were chosen to characterise the success of mannose extraction. These were:

  • The percentage of mannose removed from the water.
  • The change in concentration from the mannose solution prior to entry in the reactor.

Three factors were investigated to optimise the responses:

  • The initial mannose concentration.
  • The flow rate through the reactor.
  • The tori number.

Tori Number Flow Rate (L/s) Initial Mannose
Concentration (g/L)
Removed (%)
Mannose Concentration
Change (g/L)
9 0.0542 9.79 1.2 0.117
1 0.141 10.62 9.0 0.957
9 0.141 0.73 15.8 0.115
9 0.0542 0.78 22.1 0.172
5 0.0976 6.57 18.5 1.218
1 0.0542 16.67 26.1 4.344
1 0.141 0.672 7.0 0.005
9 0.141 10.14 31.6 0.321
9 0.0542 9.70 2.2 0.022
1 0.0542 0.78 11.6 0.010
Table 1: showing the runs for the second order interaction design. The responses were measured after 1L of mannose solution had passed through the MBR.

We analysed the data from our experiments using a Partial Least Squares SIMPLS model with no validation. From this we found the most important factors were: the interaction between the number of torus scaffold structures and the pump's flow rate into the MBR, the initial mannose concentration, and the interaction between the number of torus scaffold structures and the initial mannose concentration. These results are shown in the variable importance profile in Figure 2. A prediction simulator was used to predict the factor settings which would give the best possible MBR performance; these settings are shown in Table 2 and 3.

Figure 2: variable importance plot (VIP) showing the most important factors that effect the responses. The significant factors are those above the cutoff threshold (dotted red line) (Akarachantachote, 2013).

Factor Optimised Value
Tori Number 9
Flow rate (L/s) 0.141
Initial mannose concentration (g/L) 1.67
Table 2: showing the values of factors to optimise the performance of the metal binding reactor.

Response Predicted Value
Mannose removed (%) 26.4
Mannose removed (g/L) 3.52
Table 3: showing the values of response and the predictive values from the optimisation.

Ideally we would like to have conducted further iterations of experiments the accuracy of the optimised factor predictions and improve them further, but due to time constraints this was not possible. To see more about our design of experiment and the design evaluation we conducted to select designs please click here


The Evolution of the Metal Binding Reactor

The idea of using an MBR was introduced to us after a visit to Taunton Aquarium Centre click here to find out more. We visited here to obtain expertise on water filtration on small scales, where they have a variety of bioremediation techniques for removing harmful substances, such as nitrates and phosphates, from water to keep fish tanks healthy. They recommended a fluidised media reactor, which is a cylindrical vessel, containing media for bacterial growth, with the ability to be easily scaled up to suit a range of scenarios. In this case they used NP-Bacto-Pellets (Tropic Martin Perfect Marine) as media, which hosts nitrate metabolising bacteria that are already present in the water. They recommended that we use silica sand in our experiments as the sand still functions in low pH water, such as the water found in the Wheal Maid site where one of the lagoons was at pH 2.80±0.03. Click here to find out more about the primary data we collected. We used the idea of a fluidised media reactor to create our metal binding reactor.

Following these discussions we investigated the use of silica sand (Sigma-Aldrich) in a metal binding reactor. We sieved the sand so that we had three different size categories of sand: 150-180μm, 180-250μm and 250-425μm in diameter. This was in order to investigate how the size of sand grain affected the ability for E. coli to reduce mannose concentration in the DOE. When running the MBR with sand as media we discovered that when water was pumped through, a large amount of sand ran out of the MBR. This is not sustainable as the mass of sand in the reactor steadily decreased. We decided that we must consider implementing some of the following factors to successfully fluidise the media:

  • Larger media – as this requires larger flow rates to fluidise.
  • Lower flow rates – this would reduce the sand that flows out the MBR.
  • Finer sponges – to prevent sand escaping the MBR.

Figure 3: Jake B and Rachel using the metal binding reactor, containing silica sand as media.
Video 1: Silica sand being used as the media in the MBR. The sand is shown to be totally fluidised and is escaping through the outlet pipe (and the inlet pipe)

The most convenient of these options was to use a finer sponge that would not be able to pass sand through. For this we used cellulose sponges, placing a thin disk at the bottom and at the top of the MBR. Due to the properties of the cellulose sponge, it created a resistance to the flow rate, which in turn reduced the flow rate, addressing two of the three factors we listed above. When pumping water through the MBR with this new set up, we saw an improvement in the sands ability to fluidise, but the large resistance to flow caused by the sponges often led to reliability issues with the pump, sometimes leading to no water outflow. We observed the fluidising of the sand in some regions, but could not make the fluidising uniform due to the cellulose sponge not being uniform in shape, causing flow to arise mainly from one side.

Video 2: silica sand in the MBR with the use of the cellulose sponge. Sand can be seen to be partially fluidising.

While investigating the behaviour of the silica sand in the MBR, we also looked into the ability of MG1655 to form biofilms and grow on the surface of the sand. We did this by growing MG1655 on silica sand overnight in an LB medium. We then prepared the samples for SEM, imaged them and compared the growth to that we observed under the same conditions on NP-Bacto-Pellets, as well as negative controls for both. Initial imaging of these were unsuccessful as the overnights had been left for five days before they were eventually imaged, causing the bacteria to die from lack of nutrients, as well all the samples getting contaminated. This imaging process was then repeated and produced a density of 0.0088±0.0014μm-2 for sand and 0.2±0.4μm-2 for the NP-Bacto-Pellets. The large range in value for the NP-Bacto-Pellets was due to one image having a large number of MG1655, while the rest all had none or very few. Initially it appears that less biofilms are able to be formed by MG1655 on the sand and the NP-Bacto-Pellets, however it is not certain due to the large uncertainty of the latter.

Figure 4: SEM images of MG1655 growing on the surface of NP-Bacto-Pellets (Tropic Martin Perfect Marine) (left) and silica sand (Sigma-Aldrich) (right).

Our struggles with silica sand led us to visit the Plymouth Marine Laboratory (PML) click here to find out more, who have expertise mainly in the marine environment, but we had arranged to speak specifically with a research group who have experience with cultivating biofilms and bioremediation in the past. We were recommended to grow the bacteria on polypropylene scaffold structures (plastic pan scourers), using a chlorhexidine gluconate surfactant (peppermint flavoured mouthwash), to stress the bacteria to induce biofilm production. This led us to move away from using particulate media as a growth surface. Armed with this information we loaded the metal binding reactor with polypropylene scaffold structures that we modified into tori (see Figure). When water was pumped through the MBR the issue of media escaping was no longer an issue at any of the flow rates the pump produced. This led us to make the decision that we would focus on these scaffold structures instead of silica sand, providing we could successfully grow biofilms on the structures.

In order to do this we investigated a range of factors that would affect growth conditions:

  • Length of time left in incubator.
  • Angular frequency of incubator.
  • Concentration of surfactant.

Testing the first two factors required a method of measuring the biofilm growth on the tori. The simplest method for this is to have an observer view the tori after growth to gain a qualitative opinion on biofilm production. Although this lacked scientific rigour, it was the only viable method to produce results quickly due to time restraints. These factors were deemed less important than the final factor, so we deemed a less rigorous method acceptable. We determined that incubating the tori for 22 hours was most practical for our experimentation, and gave suitable time for biofilm formation. Secondly we found that incubating the tori statically produced better biofilms.

For the third factor, we tested a range of concentrations of surfactant and measured the biofilm growth using two methods. Firstly, we imaged the tori with SEM before and after they had been used in the MBR. This meant the surface density of bacteria on the tori could be measured. Should there be a similar number density in these samples, it demonstrates the bacteria remained adhered to the surface after fluid had been passed through, and that biofilm formation was robust. However, due to the large uncertainty previously seen from the surface density of bacteria on other mediums, the images alone may not come to a conclusion. The mannose concentration in the water was then measured to indirectly investigate the biofilm growth. The larger the concentration reduction, the greater the surface density of MG1655 on the sponge, so the more biofilm growth. Should both these results support the same conclusion we will find an optimal concentration of surfactant to use.

One of the SEM images can be seen in Figure 5 with the results of the number density of MG1655 at different surfactant concentrations before and after use in the MBR is shown in table 4. The data from here shows that the bacteria density is largest for the 0.1% surfactant concentration, and smallest at 5%. It also shows the reduction in bacteria density following being used in the MBR. All three of these reduction values have an uncertainty larger than the value itself suggesting that it is not statistically justifiable that the density changed significantly before and after use. This suggests the MG1655 remain adhered to the polypropylene surface. However due to the large uncertainty it is also possible that a significant proportion of the bacteria were removed from the surface following use. Thus further work must be conducted to investigate this.

Figure 5: times 5000 magnification SEM image of MG1655 growing on polypropylene scaffold structures at 0.1% of chlorhexidine gluconate surfactant.

Surface Density
Concentration of Surfactant (%) Before (μm-2) After (μm-2) Reduction (%)
0.1 (8.8±1.4)x10-3 (7±4)x10-3 20±60
1 (1.4±1)x10-3 (5±5)x10-4 60±130
5 (5±9)x10-5 (3±5)x10-5 50±260
Table 4: showing the surface density of MG1655 E. coli at different surfactant concentrations before and after use in the MBR.

The mannose concentration reduction was measured using high-performance liquid chromatography (HPLC), which is discussed in more detail in the next section. The result of measuring the reduction at different surfactant concentrations can be seen in Figure 6. The reduction was found by measuring the initial concentration of the mannose before running through the MBR, and measuring the final concentration of the mannose after 1L had passed through the reactor. Figure 6 conclusively shows that 0.1% surfactant produced the greatest reduction in mannose concentration so was the most effective at inducing biofilm growth on the polypropylene scaffolds out of the three surfactant concentrations investigated. It is likely that the higher concentrations of surfactant at 1% and 5% began to cause cell death.

Figure 6: graph showing how the concentration of chlorhexidine gluconate surfactant, used to induce biofilm formation on the polypropylene scaffold structure, is related to the reduction in mannose concentration after 10g/L mannose solution was passed through the metal binding reactor. The R-squared of these data points is 0.997.

Although the data points on Figure 6 suggest a linear fit, it cannot be assumed. We were led to believe there would be a peak in the curve showing an optimal concentration of surfactant as too low a concentration would not stress the bacteria to produce biofilms while too high a concentration would kill the organisms. This peak is likely to lie in the region of 0-1%, and a more detailed analysis of this region would determine this. However for the purpose of this experiment, we decided to use 0.1% surfactant for the DOE, due to time constraints.

Design Evaluation

Different experimental designs were created by varying the level of interaction between factors and the number of experimental runs in each design. These designs were then compared against each other in an attempt to find which experimental design was the most likely to accurately determine the effect of each of our factors of interest. The key areas of Design Evaluation that were compared are: power analysis, predicted variance profile, predicted variance surface, estimation efficiency and the colour map on correlations.

We looked at several designs by looking at second and third order interactions of the factors. Each of these designs were evaluated to see which would model the data to produce a reliable optimisation. Due to time constraints we decided to look at second and third order interactions for 10 experimental runs only. The main difference between these two designs came from power analysis, which is shown in figure 7. A larger power means the design is more likely to identify a true signal. Figure 7 Shows that the design with second order interaction (left) has larger values for this compared to third order interaction (right). This holds when the estimated root mean square error (RMSE) is increased, suggesting the second order interaction design will be best.

Figure 7: Power analysis for the second (left) and third (right) order interactions between the factors. Larger values for the power analysis show the design is better for this design evaluation.

The remaining areas of the Design Evaluation rendered similar results, as shown in Figure 8 showing the predicted variance surface. A flatter, more symmetric profile, is most desirable. For this reason we decided to use the second order interaction design.

Figure 8: The predicted variance profile for the second (left) and third (right) order interactions between the factors. The flatter and more symmetric the profile is the better the design is in terms of predicted variance profile.

High-Performance Liquid Chromatography (HPLC)

In order for DOE to be successful, a reliable and accurate method is required to measure the concentration of mannose before and after the solution has passed through the MBR. We decided to use high-performance liquid chromatography (HPLC) which compares the absorbance of an unknown sample with a known standard or callibration curve. We did this initially for mannose to make a calibration curve and understand the concentration range we can use in the DOE. We would have liked to choose mannose concentrations to reflect the metal ion concentrations found on our field trip to Wheal Maid click here to find out more. This found that the majority of the metal ion concentrations, above the drinking water standard, in the water fell between 0.03-50mg/L.

However due to instrument failure, the results from the field trip were delayed. For this reason we considered the concentration of metal ions from other water sources that have suffered from acid mine drainage. The Philippines for instance, at the Lafayette mine, found creeks to have metal ion concentrations of cadmium to be 0.811mg/L, whereas the Philippines National Standards for drinking water have a maximum concentration for cadmium being 0.003mg/L, almost 300 times larger than the safe levels. A key component of this experiment will be to see if the concentration of mannose can be reduced by a similar order of magnitude.

We initially decided to produce a mannose serial dilution from 40g/L to 0.3125g/L and run it through the HPLC. From this we will be able to establish the accuracy of the HPLC from the calibration curve it produces, and whether it is possible to run lower mannose concentrations through it. Three technical repeats were taken from this to give the results shown in Figure 9A representing data from one of the technical repeats. The protocol for the HPLC can be found here: HPLC protocol.

Figure 9: (A) Graph showing absorbance against retention time for each mannose concentration, shown in the legend. (B) Graph showing the calibration curve produced when the peak areas are plotted against the corresponding mannose concentration from (A). The R-squared value of this data is 0.99998, which supports the errors being too small to be observed on the graph

This calibration curve was then used to determine the concentration reduction in mannose at different surfactant concentrations, in table 1. It was also used in the DOE to determine the change in mannose concentration. The high R-squared value shows that the HPLC is a reliable method for calculating the mannose concentration, which is key for the success of DOE. The results of this are discussed in the following section.


Akarachantachote, N., et al,. Cutoff Threshold of Variable Importance in Projection for Variable Selection, International Journal of Pure and Applied Mathematics, Volume 94 No. 3, p. 307-322 (2014)