Investigating Variability in Microbial Fuel Cells

The research presented, although carried out in the area of microbial fuel cells, reaches an important and broadly impacting conclusion that when using mixed inoculate in replica reactors under replicated conditions, different communities emerge capable of different levels of metabolism. To date there has been very little research focusing on this, or even reporting it, with most studies using duplicate or triplicate reactors, in which this phenomenon is not fully observed. ABSTRACT In scientific studies, replicas should replicate, and identical conditions should produce very similar results which enable parameters to be tested. However, in microbial experiments which use real world mixed inocula to generate a new “adapted” community, this replication is very hard to achieve. The diversity within real-world microbial systems is huge, and when a subsample of this diversity is placed into a reactor vessel or onto a surface to create a biofilm, stochastic processes occur, meaning there is heterogeneity within these new communities. The smaller the subsample, the greater this heterogeneity is likely to be. Microbial fuel cells are typically operated at a very small laboratory scale and rely on specific communities which must include electrogenic bacteria, known to be of low abundance in most natural inocula. Microbial fuel cells (MFCs) offer a unique opportunity to investigate and quantify variability as they produce current when they metabolize, which can be measured in real time as the community develops. In this research, we built and tested 28 replica MFCs and ran them under identical conditions. The results showed high variability in terms of the rate and amount of current production. This variability perpetuated into subsequent feeding rounds, both with and without the presence of new inoculate. In an attempt to control this variability, reactors were reseeded using established “good” and “bad” reactors. However, this did not result in replica biofilms, suggesting there is a spatial as well as a compositional control over biofilm formation. IMPORTANCE The research presented, although carried out in the area of microbial fuel cells, reaches an important and broadly impacting conclusion that when using mixed inoculate in replica reactors under replicated conditions, different communities emerge capable of different levels of metabolism. To date there has been very little research focusing on this, or even reporting it, with most studies using duplicate or triplicate reactors, in which this phenomenon is not fully observed. Publishing data in which replicas do not replicate will be an important and brave first step in the research into understanding this fundamental microbial process.

When only simple substrates such as acetate are being digested, pure cultures of known electrochemically active microorganisms (EAMs) have been used as inoculum to create these biofilms. The most known and extensively studied EAMs are Geobacter sulfurreducens and Shewanella oneidensis (3). G. sulfurreducens is an anaerobic bacterium, found naturally in soils where reduction of iron (III) oxide is required, whereas S. oneidensis is a facultative bacterium capable of anaerobically reducing metals, nitrates, and chromates (4). When the digestion of complex organics is required, as in many intended applications of METs, a pure culture biofilm containing only electrogens is unsuitable, as a mixed consortia of bacteria are needed to complete the fermentative and hydrolytic reactions. However, bioaugmentation of anodes with an enriched electroactive consortium in MFCs, prior to inoculation with raw wastewater, has been found to significantly reduce start-up times and increase coulombic efficiency (CE), compared to reactors without precolonized anodes (5).
It would be reasonable to assume that the use of a pure culture would consistently result in optimal performance when fed a solution of pure acetate. However, although pure cultures of EAMs have been successfully used within METs to generate current (6)(7)(8)(9)(10), it is recognized that mixed cultures tend to produce higher current densities than identical reactors inoculated with pure cultures. For example, G. sulfurreducens produced 22% less power than a mixed culture from anaerobic digester sludge (11), and S. oneidensis produced 56% less power than an air cathode MFC inoculated with wastewater (12). Thus, although the use of a pure culture can lead to a high CE with simple substrates, as there are fewer electrons lost to alternative pathways, using a mixed culture is necessary and often results in a more robust and stable community that is less prone to contamination (13).
The inoculation of METs with mixed cultures is typically done using wastewaters. Within wastewater, electrogenic bacteria are likely to be present at very low numbers, estimated to be around 17 viable electrogens per mL in a population of general bacteria of 10 6 (14) (i.e., 0.0017% of the population). However, when subject to the culture vessel of an MFC reactor, which favors electrogenic metabolism, these bacteria are typically able to colonize the electrode and generate current within a number of days. Despite the high level of stochastic processes involved in this inoculation, a wide variety of wastewaters and other inocula have been used around the world to successfully initiate current in METs. The amount of current, expressed as both the peak current, (the maximum amount of substrate that was converted to current at any one time) and the total current produced from all the substrate present, is also affected by the presence of the competing bacteria that do not possess extracellular electron transfer capabilities. For example, the presence of methanogens, specifically hydrogenotrophic and acetoclastic methanogens, can result in electron losses with the conversion of hydrogen or acetate into methane (15). Similarly, the presence of sulfur-reducing bacteria can result in sulfate reducing to hydrogen sulfide, which if deposited on the anode, can inhibit electrogenic growth (15). If denitrifying bacteria are present, electron losses are possible due to the reduction of nitrate (16). Thus, the microbial composition of the inocula is highly influential on biofilm formation and therefore MET performance, yet this composition is extremely varied even within the same inocula source.
Biofilm formation and microbial community composition can be influenced by several other factors, as well as the type of inoculum used. These include the anode potential (17), cell design (18), surface charge and hydrophobicity (19), and the temperature (20). The type of feed is also important, and there is evidence to show that the more complex the substrate fed to the MET, the higher the diversity of bacteria found in the biofilm (16). A change in the type of substrate midrun, such as a switch from acetate to glucose (21), or a change in the type of wastewater used (22), could also alter the microbial community (23). Many of these factors that affect biofilm formation are controllable when setting up laboratory-based systems; however, the components and subsequent effect from wastewater inocula is not.
Most published MFC studies use only duplicate or triplicate reactors run under different conditions. This means that the interreactor variability is generally not apparent and is therefore not investigated. In addition, publishing data that shows high variability between replicas can be difficult, meaning it is an under reported phenomenon. However, it can be seen in the data presented in some articles (21,22,(24)(25)(26)(27)(28), and has been reported in others (25,29). Koch et al. (30) ran 5 identical wastewater fed reactors, reporting that each exhibited a different performance in terms of coulombic efficiency (CE) (10% to 38%) and chemical oxygen demand (COD) removal (37% to 65%), which continued over six feeding batches. In a continuous flow pilot-scale MEC, hydrogen production from each electrode ranged from 38.5 to 175.6 mL/day when using six identical electrodes within one reactor (29). Additionally, power densities ranged from 0.49 to 0.88 mW/m 2 , with one electrode failing to start up at all. Similarly, in another pilot-scale MEC, average current production varied from 192 to 252 mA, and average hydrogen production ranged from 24 to 474.6 mL/day in three identically made 1 m 2 electrodes (25). Although not explicitly discussed, variability can be seen in other pilot scale studies, where electrode units tend to be stacked or replicated to make up a larger scale treatment unit. Current densities were shown to range from 2 to 5 A/m 3 in a 200-L modulated MFC treating municipal wastewater (27) and in a stacked MFC design treating residential wastewater, a range of 0.5 to 3.5 mA in current was seen between 9 electrode configurations (28). There was no pattern seen in performance with respect to the wastewater flow in any of the pilot-scale studies.
Understanding the root cause of this variability will be key to achieving improved biofilms, capable of high pollutant removal and electron transfer rates, boosting the performance of METs overall. It is likely that implementation of METs will involve multiple stacked or cassette-style units. Variability between these will reduce the overall performance and therefore increase costs. More importantly, it highlights our lack of understanding of, and our inability to control and engineer the biofilms which operate in these systems. It introduces risk, moving the technology further away from commercialization rather than toward it. Industry would expect METs to perform at a standard and predictable rate. It may be the case that high variability exists in all open microbial systems; however, the real-time and localized nature of how performance is recorded in an MFC (i.e., as current from an electrode) makes it more apparent. Investigating and quantifying this variability gives an opportunity to further understand the bacterial communities within METs and will be critical in advancing this technology.
This study investigates the variability in 28 replica MFC reactors operated under identical conditions. The initial goal was to identify the scale of this variability in terms of peak current produced and coulombs recovered from the same feedstock and inocula. Another experiment was then set up to see if this variability persists in three consecutive feed cycles using either just pure acetate and an acetate and wastewater mix. The reactors were then inoculated with the biofilms from previous high performing reactors to ascertain whether higher performing and more consistent biofilms could be created.

RESULTS
Variation from wastewater inoculation. Initially 28-replica air-cathode MFCs (type-A design) were inoculated with a wastewater-acetate mix and left in identical conditions until current production ceased. High variation in current was observed across 27 reactors (Fig. 1), with one reactor failing to start. Current was initiated in the reactors between 0 and 7 days, and in all except one reactor, this was followed by an exponential rise in current, indicative of the growth and metabolism of electrogenic bacteria. The peak current ranged from 0.05 to 0.28 A/m 2 , with a general trend that the reactors taking longer to initiate produced a lower peak current (r 2 = 20.501, P = 0.009). Total coulombs recovered in current ranged from 33 to 165 AÁs, out of a possible 233 AÁs available. Again, the reactors taking longest to initiate were able to recover less current (r 2 = -0.794).
The reactors were then fed a wastewater-acetate mix for three consecutive runs with an attempt to standardize the current. The wastewater used was "live," containing further inocula. The variability in peak current production persisted during these three runs (Fig. 2), with ranges of 0.26, 0.20, and 0.21 A/m 2 . However, the variability in total coulombs reduced, although it was higher in the middle run compared with the other two (ranges of 103, 139, and 78 AÁs). It can be seen in Fig. 2 that some reactors which initially produced high current saw a drop-off in performance through the three runs, while others saw an increase. It can also be observed that the reactors that produced no or very little current in the first inoculation did go on to produce current in later runs. However, after using Spearman's rank correlations to analyze the performance between each subsequent run, in terms of both total coulombs and peak current, there is a positive (P values between around 0.5 and 0.7) and significant (P values between 0.000 and 0.023) correlation. This suggests that the performance established in the first run persisted through three sets of refeeding and reinoculation.
After the three feeding cycles, two high current-producing anodes, two low currentproducing anodes and three medium-producing anodes were selected for DNA extraction and 16S sequencing. The results show that there is a change in community composition from the initial wastewater inoculum to the samples taken from all reactors (see Fig. S4 in the supplemental material).
Certain families of bacteria that were present in the wastewater disappear in the reactors, such as Bacillaceae, Planococcaceae, and Acetobacteraceae, while many other bacteria appear in the reactors were below detection in the wastewater inoculum (Fig. 3). There also seems to be some clustering with regard to the low, medium, and high cur-FIG 2 Current production for the initial three standardizing runs using a wastewater-acetate mix, with box plots to represent the variation in peak current and total coulombs. Each color represents a different reactor. After this first experimental round the reactors were then dismantled, cleaned, UV sterilized, and given replacement anodes before performing a repeat experiment in which they were reinoculated with a fresh wastewater-acetate mix. The variability was shown to be reproducible (Fig. 4). Peak current ranged from 0.04 to 0.36 A/m 2 and total coulombs ranged from 46 to 183 AÁs, which were similar values to those seen in the first inoculation. However, on this occasion, the time when current was initiated was more consistent, with all reactors starting between 2 and 4 days.
The current production curves during the second inoculation seemed to visually Investigating Variability in Microbial Fuel Cells Applied and Environmental Microbiology separate into three different groups in terms of the pattern of the curve (Fig. 5). Curve A showed a sharp rise, reaching a high peak around 0.31 A/m 2 and maintaining this for 2 to 3 days before an exponential decline. Curve B was similar, but reached lower peak currents of on average 0.24 A/m 2 , and typically remained at this level for 4 to 6 days before an exponential decline. Comparatively curves in group C started with the same sharp rise, but then seemed to stabilize at a much lower peak current of around 0.1 A/m 2 , remaining at this peak current for between 5 and 10 days. The average peak current and total coulombs for these different curve patterns groups are statistically different from each other (peak current P values of 0.000 and 0.002, total coulombs P values of 0.007 and 0.025 when comparing groups A and B, and B and C, respectively, with a two sample t test) as shown in Fig. 6.
The time of initiation, defined as the time current rose above 0.02 A/m 2 , was noted for each reactor. It was seen that time for initiation showed slight negative correlations with both total coulombs (r 2 = -0.618, P = 0.001) and peak current (r 2 = -0.473, P = 0.019), ( Fig. 7A and B). With group C removed, the strength of these relationships increased (r 2 = 20.724 and 20.781 for total coulombs and peak current, respectively). This suggests that those biofilms that started to produce current more quickly produced more current overall and at a higher rate. The rate of exponential increase was also calculated for each reactor and ranged from 0.0015 to 0.017 mA.s, which correlated slightly with total coulombs (r 2 = 0.653) and strongly with peak current (r 2 = 0.916) ( Fig. 7C and D). This indicates that those which had a higher rate of current increase in the exponential phase were able to reach a higher peak current and exploit more of the COD overall.
These reactors were then fed in three cycles of just acetate with no additional inocula (see Fig. S1). As with the first run, the range in peak current observed in these runs remained the same, and the range in total coulombs produced declined. Analysis of the Spearman's rank correlations in this case showed that there was not a significant relationship between the ranked values for peak current or coulombic efficiency between sequential feeding rounds, except between rounds one and two. There was less "carry forward" in performance behavior in subsequent runs where no additional inocula was provided, compared to the previous experiment where additional inocula was added.
Controlled inoculation. The next experiments assessed whether the biofilm of the reactors could be used to seed or reinoculate other reactors and produce the same levels of performance. The reactors were switched to the type-B design, due to easy access to the anode/biofilm, and 18 reactors were inoculated with a wastewater-acetate mix. All reactors produced a current, and again, different current production curves were seen (Fig.  8). Time to initiate ranged from 1.4 to 2.7 days, peak current ranged from 0.04 A/m 2 to 0.3 A/m 2 , and total coulombs ranged from 18.5 to 167 AÁs.
Following this first inoculation run, the effluent and biofilm from two high-performing reactors (R2 and R5, green dotted line in Fig. 8) and two low performing reactors (R6 and R11, red dotted line in Fig. 8) were removed. This was then used to reinoculate

Investigating Variability in Microbial Fuel Cells
Applied and Environmental Microbiology half of the reactors each (i.e., 9 "Good" and 9 "Bad" reactors) with a sterile wastewateracetate mix. Prior to reinoculation, the reactors were cleaned, sterilized, and had their anodes replaced with the fresh carbon felt. As is seen in Fig. 9, the variability persisted in both groups, though there was a distinct difference in the performance of these. For "Good" reactors, time to initiate ranged from 1.2 to 1.4 days, peak current from 0.04 to 0.25 A/m 2 , and total coulombs from 11.4 to 80.2 AÁs. For "Bad" reactors, time to initiate ranged from 1.6 to 2.0 days, peak current from 0.01 to 0.11 A/m 2 , and total coulombs from 1.2 to 24.3 AÁs (Fig. 9, red lines). This process was repeated, though on the second occasion, only one very high performing reactor was chosen (Fig. 9, green and red dotted lines). Good reactors recorded a time to initiate ranging from 1.1 to 2.8 days, peak current from 0.02 to 0.19 A/m 2 , and total coulombs from 3.5 to 62.0 AÁs (Fig. 10, green lines). For the bad reactors, time to initiate ranged from 2.6 to 3.5 days, peak current from 0.003 to 0.03 A/m 2 , and total coulombs from 1.0 to 7.3 AÁs (Fig. 10, red lines).
On the final run the reseeding was completed as before, but this time fresh wastewater (i.e., with new live bacteria present) and acetate mix was used alongside the inocula (Fig. 11). This time, although all the good reactors were quicker to start producing current than the FIG 10 Current production during inoculation of 18 type-B air-cathode MFCs using a sterilized RSLacetate mix. Green reactors were inoculated using reactor 2, and red reactors were inoculated using reactor 13 from the previous run (run 10 in Table S1).

FIG 9
Current production during inoculation of 18 type-B air-cathode MFCs using a sterilized RSLacetate mix. Green reactors were inoculated using reactors 2 and 5, and red reactors were inoculated using reactors 6 and 11 from the first inoculation run (run 9 in Table S1). bad reactors, variation increased, with much higher overlap seen when comparing peak current and total coulombs. For good reactors, time to initiate ranged from 0.8 to 1.2 days, peak current from 0.04 to 0.39 A/m 2 , and total coulombs from 9.6 to 89.2 AÁs (Fig. 11, green lines). For bad reactors, time to initiate ranged from 1.4 to 1.8 days, peak current from 0.04 to 0.20 A/m 2 , and total coulombs from 7.5 to 105.4 AÁs (Fig. 11, red lines).
In all three runs, good reactors inoculated quicker than bad reactors. When running a two-sample t test, a significant difference in Fig. 9 and Fig. 10 was seen when comparing good and bad reactors with respect to total coulombs (P values = 0.013 and 0.004, respectively) and peak current (P values = 0.049 and 0.002, respectively). However, in Fig. 11 the use of nonsterile wastewater resulted in high overlap between both groups, with P values of 0.250 and 0.220 for total coulombs and peak current, respectively. Although there were significant differences between good and bad reactors in Fig. 9 and Fig. 10, reactor performance was still highly variable within each group.
Samples of the original inocula and the good and bad inocula for each successive run were taken and 16S sequencing performed. Figure 12 shows that there is a shift in community composition from the original inoculum toward the fourth inoculation round. This shift entails an observable but not significant reduction in diversity and richness (see Fig. S6 in the supplemental material). The biofilms become dominated by bacteria within the class of Gammaproteobacteria, with a significant reduction in others such as Alphaproteobacteria, and Actinobacteria. In the second inoculation round the community of both the good and bad remain quite similar, with only a small change from the original wastewater. This goes on to produce reactors with quite mixed performance in terms of current (Fig. 9). However, those reactors chosen for the third inoculations have a very different composition, with the good being very different from the wastewater inoculum and the round-2 inocula. By the time of the fourth inoculation round, both the good and bad inocula become genetically similar, and in many reactors produce equivalent current output (Fig. 11).

DISCUSSION
Microbial fuel cells are a form of biotechnology that aims to rival current wastewater treatment methods. They rely upon the combined effect of different species of bacteria to remove pollutants while simultaneously recovering energy (16). Inoculation of MFCs is often done with a natural bacteria source such as domestic wastewater or soil. However, due to the lack of control of the high diversity of bacteria within wastewater, FIG 11 Current production during inoculation of 20 type-B air-cathode MFCs using a sterilized RSLacetate mix. Green reactors were inoculated using effluent and biofilm from reactors 2 and 3, and red reactors were inoculated using effluent and biofilm from reactors 18 and 19 from the previous run (run 11 in Table S1). and the low proportion of electrogens present (14), a natural variability occurs. This variability is often not discussed, or even obvious, due to the small number of replicates used.
Variability within MFCs is observed directly in the pattern of the current generated. We inoculated 28 replica reactors under identical conditions, at the same time, using the same wastewater and provided them with a simple acetate substrate as the feed. The results show that there was a high level of variability in the peak and total current produced, and a moderate level of variability in the time of initiation of current and the rate at which current increases. The variability occurred during the first batch run with the inoculation and persisted into subsequent runs regardless of whether these runs included fresh wastewater inocula, pregrown electrogenic inocula, or no inocula at all. This shows that despite the reactor configurations being identical, and internal resistances between the electrode being very similar, there are consequential differences in the biofilm formation process within the reactors.
Some of this variability could be a result of the random distribution of electrogenic bacteria present in wastewater inoculum. Yang et al. (40) found there to be a high and relatively consistent number of electrogens present in influent wastewater of 7.0 6 0.3 Â 10 5 per mL, which was measured across three samples. However, in a study which estimated viable electrogens present in wastewater (i.e., those able to initiate the generation of current), based on most probable number (MPN) statistics of the initiation of 17 reactors, this number was much lower and more variable and suggested between 5 and 50 electrogens per mL (14). This means that in our reactors containing 30 mL of wastewater there could be between 500 to 5,000 viable electrogens, and after 4 days of growth (based on a doubling time of 24 h [41]) there would be between 4,000 and 40,000 electrogens. A small variation in initial starting number of electrogens could result in very different biofilm populations. To reduce this effect, the inoculum was well mixed in a 50:50 ratio with acetate prior to being put in the reactors. However, if the variability was caused by random differences in starting numbers of bacteria, a histogram of the maximum current should have a normal distribution. It did not, suggesting a more complex reason for the variability. In addition, in the final experiments where the inoculum was produced by mixing up an existing anode into a solution of acetate, (i.e., the same acclimated bacterial community was redistributed into the reactors), high variability was still observed. Therefore, although differences in actual numbers of viable electrogens entering the reactors may cause some variability, other factors are also likely important. These may include the proportions of competitive bacteria which take up either resource (substrate) or biofilm space (or both), and symbiotic bacteria which may help (e.g., through the production of electron shuttles). It may also be the case that the very localized differences in bacterial proportions lead to different localized substrate gradients, which influence the rates at which bacteria grow and metabolize. Variability of current generation shown in our experiments matches observations of previous studies which have investigated microbial community assembly and bioreactor performance. When 3 parallel anaerobic digesters were seeded with the same digested manure inoculum and operated under identical conditions, their biogas production rates and microbial communities at steady state were significantly different (42). Similarly, high b-diversity was found between 14 identical microbial electrolysis cells (MECs) inoculated with the same wastewater source, which were divided into 4 distinct groups based on similarity in community structure and functionality (assessed on H 2 and CH 4 yields) (43). This aligns with our grouping of MFCs based on total charge recovered, peak current and Coulombic efficiency, and further suggests stochastic processes (e.g., initial colonization of the anode) are likely to have contributed a substantial role in the observed variability.
The establishment of a bacterial community in an MFC can be understood in terms of classic bacterial growth under favorable conditions (14). Typically, this contains four stages: the lag phase, exponential phase, stationary phase, and death phase, with the first two relating to the growth of a biofilm. The lag phase is equivalent to the time taken for initiation of current; it represents the period of time taken for the low number of electrogens in the inoculum to adjust to the environment and begin to establish an electroactive biofilm on the anode. The exponential phase is equivalent to the current production rate, which, assuming each electrogen can produce electrons at the same rate, shows the rate of growth of electrogenic organisms.
In our experiment, the lag phase took between 0 and 7 days in the first round and more consistently 2 to 4 days in the second round of inoculation. Reactors were held at laboratory temperature, which may have been different between each experiment, causing some of the difference between each run. This variability shows that despite the reactor configurations being identical, and internal resistances between the electrode being very similar, there are immediately consequential differences in the environment that the electrogenic bacteria must adjust to. There was a weak negative correlation between the time to initiate and both the peak current and the total coulombs produced in the feeding cycle. That is, the slower the bacteria adjust to the environment and begin to grow, the slower this growth rate will be and the less food these bacteria will be able to metabolize overall. This indicates that there is competing, nonelectrogenic bacteria in the system, which also grow and utilize substrate, preventing it from being converted to electrons. The low value of this correlation, especially when group C is included (which seem to grow and then be limited), suggests that there are other complex phenomena occurring.
There was a weak positive correlation between current production rate and total coulombs consumed, but a strong and significant positive correlation with peak current achieved. The biofilms that had a higher rate of current increase in the exponential phase were able to exploit more of the substrate overall and reach a higher peak current. This suggests that if electrogens were quicker to establish, they could be more dominant. However, in a small subset of reactors (group C in run 2), growth was initiated, and the growth rate was high, but then there appeared to be a sudden constraint with the current capped at a certain level. This indicates that the electrogens present are still able to metabolize and have access to substrate, but they can't grow, possibly due to competition for space on the electrode or some other limiting factor.
It was anticipated that with subsequent feeding cycles, variability would reduce as the electrogenic biofilm stabilized. It was seen that in terms of total coulombs recovered, (i.e., how much of the substrate ended up as current), there was a slight reduction in variability, but no reduction in the variability was observed in peak current. Interestingly, the performance in individual reactors that occurred in the initial run was perpetuated into subsequent runs when additional wastewater inocula was also provided, but much less so when no additional inocula was provided. This suggests that in the first inoculation cycle, the large diversity of bacteria added to the cell can acclimatize quickly into a relatively stable evolutionary state, and therefore, when the feed and inocula is repeated, there is little community shift (44). However, when the situation changes, for example, if the substrate is changed but no inocula is added in the feeding, the bacteria need to adjust to this new state, and they do so and then perform differently. Importantly, in either case, we did not observe a significant reduction in variability or improvement in performance. This is important for reactor start up, especially at scale and with real wastes, as it suggests that once formed, a biofilm will not change. Therefore, it could be argued that when a reactor is started with multiple electrodes, those which initially do not start well should be removed and replaced, rather than hoping for improvement.
Analysis of the 16S sequencing showed that even within 3 feeding cycles, the community in the reactors shifts significantly from the inoculum. Examination of the operational taxonomic unit (OTU) table shows that Geobacter, a common electrogen, is not observable in the wastewater inoculum and only present in relatively low numbers in the medium and high current-producing reactors, and not in the low current-producing reactors. This may indicate that with only three feeding cycles, this common electrogen has not been able to dominate the biofilms, suggesting preacclimatization with a pure culture could be a strategy for improved performance (45).
When comparing the community composition of the samples taken from the low, medium, and high current-producing biofilms there is some clustering, as the samples group together in different regions of the phylogenetic tree, and within the PcOA plot. This indicates there is a difference in the microbial community makeup of those reactors that perform better compared to those which perform less well.
In an attempt to control variability, in the third experimental round, reactors that showed high performance were used to inoculate further fresh sterile reactors, with lower performing reactors as a control. The rationale behind this was that it would be possible to transfer a known good community of bacteria into further reactors. Often, previous studies have adopted a form of this technique, inoculating with an already enriched culture from previous biofilms (37). The reactors given the good inoculum were observed to initiate current quicker, and generally have higher peak currents and coulombs recovered compared to those given the bad inoculum. However, for both groups, but especially the good reactors, variability in performance remained very high. The same acclimated inocula produced very different performance when redistributed into new reactors. This suggests that there may also be a spatial element to the distribution of bacteria which is important, as well as their composition (46), and/ or that the key electrogenic bacteria exist in flocs, which remain clumped together even within the mixing process, and therefore are not distributed evenly into the new reactors. Additionally, in the third inoculation, good reactors had a very different community composition than the bad reactors, dominated by Gammaproteobacteria rather than Actinobacteria. This round of inoculation also showed the largest difference in performance between both good and bad reactors, in terms of both peak current and total coulombs recovered.
It is also seen, in contrast to the first experiment (where the addition of fresh active wastewater had little impact on the established biofilm), that in the final run, the addition of fresh wastewater inocula to the preacclimatized inocula caused the differences between the good and the bad reactors to diminish significantly. This was highlighted by the 16S sequencing, which showed a large change in the relative abundance of the reactors from the third inoculation to the fourth, resulting in both the good and the bad reactors becoming genetically similar. In this case, unlike in the first experiment, all bacteria were in suspension and had to establish on the clean anode, and the "new" electrogens in the wastewater were clearly able to compete with the poor electrogenic community to establish reasonable performance. If reseeding is to be used as a technique to transfer biofilms, then implanting patches of biofilm is likely to be a better solution than putting the biofilm back into solution (47).
Variability in the performance of microbial fuel cells is a highly observable phenomena, as metabolism is recorded in real time as the current output. Although some variability in performance will be caused by minor differences in reactor set-ups even with identical reactors, the experiments conducted here show that some of this variability is biological and difficult to control. Differences in the microbial input into reactors are likely to cause some of this variability. Even within the same inocula, there may be clusters of viable electrogens, meaning the same inocula amount may have highly variable electrogen amounts. This may also be the case for competing and syntrophic bacteria. The use of pure cultures, or preacclimatized cultures, may avoid some of this variability, as it is seen that the performance of the MFC is carried forward into subsequent batches even when, and more so when, fresh inocula is added. However, there is also likely to be a spatial element to this. When a preacclimatized inocula is removed from a biofilm and mixed up, it offers only a limited advantage over non acclimatized inocula. This will be important as the technology moves into real applications, which are likely to involve a continuous flow of live wastewater. Pilot studies have shown that with multiple cassettes within the same reactor, some perform better than others in terms of current generation, and this behavior can be sustained over long time periods under continuous flow of wastewater (24)(25)(26).
The variability shown in these experiments also has significant consequences for laboratory-scale testing on MFCs. Hypotheses could be shown to be correct or incorrect purely from the random effect of the inoculum. For example, using the data from our study, if we hypothesized that a reactor could recover more charge if fed pure acetate than if it were fed complex substrate, by comparing R19 in run 4 with R24 in run 6, then acetate recovers more charge than wastewater (80 AÁs versus 119 AÁs). However, the opposite of this is seen when comparing R11 in run 4 and R11 in run 6, where wastewater-fed reactors recovered more charge than acetate ones (158 AÁs versus 97 AÁs). Even if the reactors were in triplicates, by selecting certain reactors, each scenario is still possible within the data generated (see Fig. S2 and S3 in the supplemental material).
This study shows that replication within microbial fuel cell experiments is very important. Even with duplicate or triplicate reactors, natural variability in performance could confound whatever parameter was being tested. It also highlights the need for more research into the causes of this variability, and strategies to reduce it, including how scale may impact upon this. The good biofilms in this study produced up to 7 times more current at their peak than the bad ones and recovered or removed substantially more substrate. This could easily make the difference between economic viability versus nonviability, and compliance versus noncompliance.

MATERIALS AND METHODS
Reactor configuration. Two reactor designs have been used during this study. The type-A design was used initially to investigate the variability in a large number of identical reactors. However, it became clear that modification to the MFC design would be needed to enable easier access to the biofilm. Therefore, some of the results in this study will refer to the initial design, termed type-A. The following design, which was used for the reseeding experiments, is termed type-B.
Type-A reactor design. The experiment demanded an MFC design that enabled high replicability, high throughput, and the ease to analyze the substrates simply and quickly. Therefore, the 28 identical air-cathode MFCs were cylindrical chambers with three sample ports, sealed at the opposite end with a plastic bung and each had a 60 mL working volume. The anode was 40 mm in diameter, 3 mm width carbon felt (SGL Carbon, Wiesbaden, Germany) connected to 0.6 mm 2 stainless steel wire (Clarke Tools, Chronos Ltd., Dunstable, UK), fed out of a sample port through a rubber bung. The air cathode was 0.2 mg/cm 2 20% platinum on a Vulcan carbon cloth electrode (Fuel Cell Store, TX, USA), platinum side facing the air. A 300 X resistor connected the anode wire and the carbon cloth cathode, secured by crocodile clips. All wire connections were soldered to maintain high connection. The carbon cloth cathode was 50 mm Â 50 mm, with an extra 10 mm Â 20-mm section cut at the top and folded at 90 degrees to allow crocodile clips to connect. This was glued to a 70 mm Â 70-mm rubber square with a 40-mm diameter circle cut out and glued onto the end of the tube using Gorilla epoxy resin (Gorilla Glue Ltd., Chorley, UK) (Fig. 13).
Type-B reactor design. The new design, termed type-B, consisted of 18 identical MFCs. These reactors were made using inexpensive plastic containers designed for food storage (LocknLock Ltd., Seoul, South Korea). They have the advantage of being inexpensive, and therefore enabling a high number of replicates. They are also simple to open, resulting in easy access to the biofilm, while maintaining water tightness when closed.
The reactors were cylindrical chambers with two sample ports, resulting in a 100-mL working volume (Fig. 14). The anode was 20 mm Â 20 mm Â 3 mm carbon felt (SGL Carbon, Wiesbaden, Germany) connected to 0.6 mm 2 stainless steel MIG welding wire (Clarke Tools, Chronos Ltd., Dunstable, UK) fed out of a sample port through a rubber bung. The air cathode was 0.2 mg/cm 2 20% platinum on a Vulcan carbon cloth electrode (Fuel Cell Store, TX, USA), platinum side facing the air. A 300 X resistor connected the anode wire and the carbon cloth cathode, secured by crocodile clips. All wire connections were soldered to maintain high connection. The carbon cloth cathode was 50 mm Â 50 mm, with an extra 10 mm Â 20-mm section cut at the top, which was folded at 90 degrees to allow crocodile clips to connect. This was glued to a rubber circle (outer diameter 60 mm) to aid its seal to the reactor and limit damage to the air cathode. A 40-mm diameter circle was cut out of the rubber to fit around the 40-mm diameter hole cut out of the reactor lid, with the cathode glued on top. Epoxy resin (Gorilla Glue Ltd., Chorley, UK) was used to glue both the cathode and rubber together, and the rubber to the reactor lid, to provide a watertight seal (Fig. 14).
Electrochemical analysis. Potentiostatic electrochemical impedance spectroscopy (EIS) was performed on the reactors after construction to determine the variability of internal resistance between reactors prior to their use. Whole-cell impedance measurements were completed in two-electrode configuration as the MFC setup prevented the use of a reference electrode to study impedances on individual electrodes (31). The anode acted as the working electrode, while the air-cathode served as both the reference and counter electrode. EIS was performed using an Autolab PGSTAT302N potentiostat and NOVA software (Metrohm, Switzerland). Working electrode potential was set at the average open-circuit potential of reactors filled with blank acetate solution to determine the internal resistance of the reactors prior to use. AC voltage amplitude was 10 mV and frequencies between 100 kHz to 50 mHz were applied as this range is sufficient to capture the key processes during MFC operation (32,33).  Internal resistance values were estimated from the high frequency intercept of the real impedance axis of Nyquist plots. EIS spectra showed that the internal resistance between reactors was minimal. EIS measurements were repeated during the biofilm growth phase to investigate if there was a relationship between internal resistance and rate of current production for each reactor. Charge transfer resistances could not be determined as data points at lower frequencies (,1 Hz) became chaotic, therefore, no valid estimates were yielded. This could be due to the system not being at steady state during EIS measurements, as cell populations and reactant concentrations are changing during the biofilm growth phase, which could influence the current response and ultimately the resulting impedance spectra (32,33).
Experimental runs. We performed inoculation experiments using the type-A design, and reseeding experiments using the type-B design. All reactors were operated within a laboratory at ambient temperatures around 22°C. Inoculation refers to clean sterile reactors with a fresh anode, being filled with an acetate mix combined with either raw wastewater or return sludge liquor (RSL). Raw wastewater was collected from Birtley Wastewater Treatment Works (Northumbrian Water Ltd. [NWL]) and RSL was collected from Howdon Wastewater Treatment Works (NWL). Standardizing runs refers to having a new fresh substrate added without change to the anode.
The inoculation experiments used 28 replicas of the type-A design for the air-cathode MFC. The reactors were assembled identically and inoculated with the same mixture of 50% fresh raw wastewater and 50% acetate solution, all at the same time. The reactors were then left for 12 days, allowing the biofilm to form, develop, and consume the substrate in the reactor. The current produced was measured using high resolution multichannel data loggers from PicoTech (34), and the COD consumed was measured as described below. Following this, the reactors were subject to three standardizing runs using a raw wastewater-acetate mix. These standardizing runs were when the reactors were all fed the same mixture, in an attempt to normalize the performance between then all. The reactors were then cleaned, UV sterilized, and rebuilt with fresh new anode and the experiment repeated; the standardizing runs were then run with a pure acetate feed solution.
The reseeding experiments were performed using the type-B design to enable easier biofilm access. The 18 reactors were inoculated with an RSL-acetate mix, and the current was allowed to develop. We then chose the two highest performing reactors and the two lowest performing reactors in terms of current production. The anodes from these reactors containing the biofilm were mixed thoroughly with a mixture of sterile wastewater and acetate solution. This mix was then used to inoculate 9 reactors each, i.e., half with the "good" inocula and half with the "bad" inocula. This exercise was repeated twice, on each occasion taking the best and worst two biofilms to reseed clean sterile reactors in a sterile substrate. A final run was made of this process using nonsterile wastewater mixed with acetate as the substrate.
Analytical methods. Following the end of each run, all reactors were sampled and measured for COD. COD removal was measured using Merck COD cuvette tests (25 to 1,500 mg/L) in duplicates according to standard methods. Voltage was measured using Pico 6 software across a 300 X resistor, with ADC-20 and ADC-24 PicoLogers for continuous measurement. Biofilms were removed from the reactors following runs 9, 10, and 11. Careful handling ensured no damage to the biofilm, which was placed in the effluent from the reactor it came from. This was then mixed using a magnetic stirrer, before being used as inoculum for the following run.
Substrate sterilization. Sterilization of the wastewater was accomplished by filtration and UV light. The RSL was first allowed to settle before the soluble fraction was passed through a 0.2-mm filter. This was then left under a UV light for 30 min prior to being used as a substrate. Acetate mixes were sterilized using an autoclave.
DNA extraction and sequencing of 16S rRNA genes. Samples were extracted either at the point of inoculation or following the end of each trial. They were then stored in a freezer at 220°C. Genomic DNA was extracted using the DNeasy PowerLyzer PowerSoil kit (Qiagen, UK) as per the manufacturer's instructions. Extracted DNA underwent sequencing at NU-OMICS (Northumbria University, UK). Sequencing targeted the V4 region of the 16s rRNA following the Earth Microbiome 16S Illumina amplicon protocol (35) using primers 515F (36) to 806R (37). Sequencing was carried out on using 250 Â 2 chemistry on a MiSeq V2 500-bp reagent cartridge (Illumina, UK). Sequences were processed using mothur v.1.48.0 software (38). Processing was performed using the standard MiSeq procedure accessed in July 2022 with OTU matching performed at 97% similarity against the Greengenes reference database. Data were analyzed using Agile Toolkit for Incisive Microbial Analysis (ATIMA) developed by the Center for Metagenomics and Microbiome Research at the Baylor College of Medicine.
Calculations. Current density, COD removal, volumetric treatment rate (VTR), coulombic efficiency (CE), peak current density, and total coulombs were calculated as follows.
Voltage was measured using high resolution multichannel data loggers from PicoTech (34). This was converted to current density based on projected anode surface area. According to Ohms law, the current can be calculated as: where I is the current (amps), V is the voltage (volts), and R is the resistance (Ohms). This was then divided by the surface area to get current density: where A is the projected anode surface area (m 2 ), and J is the current density (A/m 2 ). VTR has been calculated using the COD removed and the retention time of the substrate, where S is COD removed (mg/L), HRT is the retention time (days), and VTR is the volumetric treatment rate (kgCOD/m 3 Áday).
VTR ¼ S HRT Â 1; 000 Total coulombs (Cp) were calculated by the sum of the total current. As the PicoLog software recorded the volts over a 300 X resistor every minute, this was multiplied by 60 to convert to seconds, and then summed to give the total charge produced during the batch cycle.
CE was calculated by: where CE is the coulombic efficiency, C p is the total coulombs (current over time), and C n is the theoretical coulombs that could be recovered from the COD removed. Theoretical coulombs were calculated based on Logan et al. (61), and therefore CE was calculated by: where F is Faraday's constant (96,485 C/mol of electrons), 8 is a constant used for COD (39), V An is the liquid volume in the anode chamber (L), and DCOD is the change in the chemical oxygen demand (COD) (g/L) that has occurred over the batch cycle.
There is a tacit assumption that the increase in current represents growth rates of electrogens. Therefore, the average increase in amps (A) per second has been taken from each reactor for the period of a rise in current, and the rise in current is reported in mA.s. Additionally, the point at which the current started to increase above a threshold of 0.02 A/m 2 , was labeled the time of initiation, and is recorded in days.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. SUPPLEMENTAL FILE 1, DOCX file, 0.3 MB.