Heterogeneity in respiratory electron transfer and adaptive iron utilization in a bacterial biofilm

In Bacillus subtilis, robust biofilm formation requires large quantities of ferric iron. Here we show that this process requires preferential production of a siderophore precursor, 2,3-dihydroxybenzoate, instead of the siderophore bacillibactin. A large proportion of iron is associated extracellularly with the biofilm matrix. The biofilms are conductive, with extracellular iron potentially acting as electron acceptor. A relatively small proportion of ferric iron is internalized and boosts production of iron-containing enzymes involved in respiratory electron transfer and establishing strong membrane potential, which is key to biofilm matrix production. Our study highlights metabolic diversity and versatile energy generation strategies within B. subtilis biofilms.

M ost bacteria are capable of forming surface-associated, architecturally complex communities, known as biofilms 1,2 . The Gram-positive bacterium Bacillus subtilis forms morphologically complex colony biofilms on solid surface and pellicle biofilms at the air/liquid interface, in specialized biofilm-inducing media 3,4 . Cell differentiation occurs in the B. subtilis biofilm both spatially and temporally, and is regulated by integrative signaling pathways and influenced by various environmental factors [4][5][6] . Cell differentiation generates phenotypically distinct cell types within the biofilm, such as matrix producers, swimmers, competent cells, antibiotic producers, and so on. Each cell type possesses unique features and functionality, yet different cell types complement with each other in that the entire community shows synergy and cooperation 6,7 . In summary, complexity and heterogeneity is a hallmark feature of the B. subtilis biofilm.
Iron is an essential nutrient element for growth and a cofactor for various enzymes and proteins involved in key biological processes in the bacteria, in particular cellular metabolism and energy generation 8 . In the natural environments, iron availability is very limited due to extremely low solubility of ferric iron (Fe 3+ ) under neutral pH (10 −18 M) 8 . Bacteria thus developed various strategies to uptake iron from the environment. When intracellular levels of iron are low, bacteria turn on multiple iron uptake systems for iron acquisition. In B. subtilis, those iron acquisition systems are composed of transporters for the import of elemental iron, ferric citrate, ferrioxamine, ferrichrome, and petrobactin 9 . A key regulator, Ferric uptake regulator (Fur), is responsible for the regulation and derepression of the above iron acquisition systems when iron is limited 10,11 . Fur also regulates genes involved in the biosynthesis of a B. subtilis siderophore, bacillibactin (DhbA-CEBF), and a cognate uptake system (FeuABC-YusV) 12,13 . Bacillibactin is a catechol siderophore and binds iron with an extremely high affinity, allowing B. subtilis cells to acquire very low amounts of iron from the environment. In most pathogens, siderophores are essential for the bacteria to survive in the host or environment [14][15][16] . On the other hand, excessive intracellular iron could be toxic due to its involvement in the Fenton's reaction, which generates non-selective free radicals that can damage various biological molecules such as protein, lipid, and DNA 17 . Thus, in bacteria, iron homeostasis plays an important role in maintaining an appropriate range of intracellular iron levels to satisfy the normal need for metabolism and growth 8,15 .
Recent studies indicate importance of iron not only in freeliving bacteria but also in biofilm formation [18][19][20] . In biofilmforming bacteria such as Pseudomonas aeruginosa and Staphylococcus aureus, high levels of iron were shown to be important for robust biofilm formation 18,19 , suggesting a common and important role of iron in biofilm development. In B. subtilis, robust biofilm formation also demands unusually high levels of ferric iron, hundreds fold higher than needed for normal growth 3 , yet the exact reason is less clear. In this study, we aim to address the above question. We provide mechanistic details of why excessive iron is needed for robust biofilm formation in B. subtilis and uncover an adaptive strategy for acquisition and utilization of large amounts of iron during B. subtilis biofilm development.

Results
2,3-Dihydroxybenzoate is essential for biofilm formation. In a previous study 21 , we investigated the role of various nonribosomal peptides (NRPs) in B. subtilis in biofilm formation. Many of those NRPs had dispensable roles in biofilm formation as assessed by the biofilm phenotypes of the mutants 21 . For instance, two mutants deficient in either bacillaene or fengycin biosynthesis showed little noticeable biofilm phenotypes (Δpks and Δpps, Fig. 1a, b). Surprisingly, a ΔdhbA mutant deficient in the siderophore bacillibactin production was very defective in biofilm formation (Fig. 1b). Bacillibactin is a catechol siderophore that binds to ferric iron with an extremely high affinity 22 . The role of bacillibactin in iron acquisition is well known, but its role in biofilm formation has not been studied in B. subtilis until very recently 23 .
Bacillibactin biosynthesis relies on the dhbA-F operon encoding enzymes that carry out four sequential reactions converting 3chorismate to bacillibactin (Fig. 1a, c) 24 . Since dhbF encodes the most important enzyme involved in the final reaction in bacillibactin biosynthesis (Fig. 1c), a deletion mutant of dhbF (ΔdhbF) was constructed and the biofilm phenotype of the mutant was examined. To our surprise, the mutant did not exhibit any noticeable biofilm defect (Fig. 1d). Puzzled by this observation, non-polar in-frame deletion mutations for each individual genes in the dhbA-F operon were constructed and biofilm phenotypes by those in-frame deletion mutants were examined. As shown in Fig. 1d, ΔdhbA, ΔdhbB, and ΔdhbC all had a very severe biofilm defect. In contrast, ΔdhbE formed robust pellicle biofilms, similar to that of ΔdhbF. Since DhbE and DhbF are known to be only involved in the last step of bacillibactin biosynthesis, which (together with DhbB) converts 2,3-dihydroxybenzoate (DHB) to bacillibactin, a trimeric ester of 2,3-dihydroxybenzoate-glycine-threonine (Fig. 1c), these results suggest that bacillibactin is not important; rather, the precursor DHB plays an essential role in biofilm formation. To further test this idea, a chemical complementation was performed by supplementing pure DHB (1 μg ml −1 ) to the biofilm media. This time, ΔdhbA, ΔdhbB, and ΔdhbC all formed robust, wild-type-like pellicle biofilms (MSgg + DHB, Fig. 1d). We also noticed that in order to completely rescue the biofilm defect of ΔdhbA, ΔdhbB, and ΔdhbC to the wild-type level, supplementation of about 60 μM (equals to about 1 μg ml −1 ) DHB was needed, implying that wild-type cells produced and secreted DHB sufficiently. This implication was also supported by a recent study, in which secreted DHB amounts were directly measured 23 . DHB is able to bind ferric ion, albeit at a lower binding affinity compared to that of bacillibactin 9 . A similar DHB, 3,4-dihydroxybenzoate, or protocatechuic acid, produced by Corynebacterium glutamicum, is a well-studied iron chelator 25,26 .
dhbA-F genes are differentially expressed in biofilm cells. To learn more about DHB biosynthesis, we decided to investigate the regulation of the dhbA-F operon under biofilm conditions by quantitative real-time PCR (RT-qPCR). To do so, a pair of primers were designed to probe transcription for each of the dhbA, dhbB, dhbC, and dhbE genes, while three pairs were designed to probe dhbF transcription since the size of dhbF (~7.13 kb) is almost 60% of the entire operon (Fig. 2a). Results from RT-qPCR showed that in cells collected from pellicle biofilms, expression of all genes in the dhbA-F operon was detected except for dhbF (black bars, Fig. 2b). In fact, all three pairs of primers failed to detect the transcription of dhbF (# indicates below detection limit, Fig. 2b). This result was quite interesting to us for two reasons. First, differential expression of the genes within the dhbA-F operon was observed, which could provide genetic basis for strong production of DHB (over bacillibactin) during biofilm formation. In a recent study, Rizzi et al. 23 quantified the production of DHB and bacillibactin during pellicle biofilm formation in B. subtilis and found that DHB was produced about 10fold higher than bacillibactin. Second, the dhbA-F operon was readily expressed in the iron-rich biofilm medium (50 μM FeCl 3 ). This seemed to contradict to the knowledge that the dhbA-F operon is repressed by Fur under iron-rich conditions in B.
subtilis 10,13 . The qPCR assay was repeated by using pellicle biofilm cells from modified MSgg that contained only 1% of the regularly added ferric iron (0.5 μM FeCl 3 ). Again, it failed to detect any expression of the dhbF gene (gray bars, Fig. 2b). Expression of all other genes in the operon was detected, but at a much lower level when compared to that of 50 μM FeCl 3 . The undetectable expression of dhbF under both iron conditions (50 and 0.5 μM of FeCl 3 ) seemed to match with the observed wildtype-like biofilm phenotypes of ΔdhbF, suggesting a largely dispensable role of bacillibactin in biofilm formation. Lastly, as a note, the expression of dhbF was detected when cells were grown under extreme iron-limiting conditions (MSgg without any added FeCl 3 ), and the ΔdhbF mutant grew much slower in minimal media without FeCl 3 supplementation (data not shown), consistent with previous studies 22 .
dhbA-F is regulated by AbrB in response to biofilm signals. The observation that the expression of the dhbA-F operon was much lower at the lower iron condition (0.5 μM) than the higher iron condition (50 μM) was somewhat surprising given that the dhbA-F operon is known to be repressed by Fur under iron-rich conditions in B. subtilis 13 . Therefore, increased expression of the dhb genes under the higher iron condition (50 μM FeCl 3 ) was unlikely due to the regulation by Fur. To search for additional regulators for the dhbF operon under biofilm conditions, the regulatory sequence of the operon was analyzed and a putative AbrB recognition motif was identified 27 . AbrB is a biofilm repressor and a transition state regulator known to control a number of operons involved in biofilm formation and secondary metabolites biosynthesis [28][29][30] . To test if AbrB regulates the operon, a promoter-lacZ fusion for the dhb operon (P dhbA -lacZ) was constructed and integrated to the chromosomal amyE locus of the wild-type strain, the ΔabrB mutant, and the Δfur mutant (as a positive control). Expression of the reporter fusion in the wildtype and the mutants was assayed. The results confirmed that in the Δfur mutant, the activity of the reporter fusion was very   , and ΔdhbA(YQ90, insertional deletion). Cells were incubated in MSgg at 30°C for 2 days before images were taken. Scale bar, 5 mm. c A 4-step enzymatic conversion from 3-chorismate to bacillibactin carried out by enzymes encoded in the dhbA-F operon 24 . d Formation of pellicle biofilms by non-polar in-frame deletion mutants of the dhbA-F operon and chemical complementation. Strains shown are WT(3610), ΔdhbA(YQ97, inframe deletion), ΔdhbB(YQ98), ΔdhbC(YQ99), ΔdhbE(YQ100), and ΔdhbF(YQ101). Cells were incubated in MSgg at 30°C for 2 days before images were taken. For chemical complementation, DHB was added at the final concentration of 1 μg ml −1 (~60 μM). Scale bar, 5 mm strongly induced (Fig. 2c), suggesting that in the wild-type strain, Fur-mediated strong repression was in place. Interestingly, in the ΔabrB mutant, the activity of the reporter fusion was also strongly induced (Fig. 2c). This indicates that under biofilm conditions, the dhb operon is co-repressed by both AbrB and Fur and derepression can be triggered by either Fur in response to iron limitation or by inactivation of AbrB via sensing of biofilm signals (Fig. 2a). Previous studies have shown that AbrB was gradually inactivated in the presence of biofilm signals through the known biofilm pathway and the master regulator Spo0A 30,31 . The observed modest derepression of the dhbA-F operon in wild-type cells in later time points (blue line, Fig. 2c) was consistent with the above idea and with the expression profile of other AbrBregulated genes such as the biofilm matrix operon tapA-sipW-tasA 30 .
Biofilm formation demands large amounts of ferric iron. Although ferric iron is very poorly soluble in regular media, large amounts of ferric iron were shown to be needed in the biofilm media for B. subtilis to form robust biofilms (Fig. 2d). A decrease of the concentration of ferric iron in the media from 50 to 5 μM already started to impact biofilm formation while a further decrease to 0.5 μM severely blocked biofilm formation (Fig. 2d).
The block in biofilm formation was not due to growth inhibition, since only a mild difference was seen in growth even if ferric iron was not at all added to the media (presumably trace amount of ferric iron was present in the water and other chemical ingredients during media preparation) ( Supplementary Fig. 1A).
Excess iron stimulates production of iron-binding enzymes.
Higher media iron concentrations might boost yet unknown activities critical for robust biofilm formation. Many of the ironresponsive genes are under the control of Fur 9,32 . We argued that those genes would be less likely to be directly involved since many of them are primarily involved in iron uptake and since those genes would be further repressed by Fur upon increasing iron concentrations 10 . Our focus was then shifted to genes encoding proteins that contain iron as a cofactor. Activities of those proteins depend on iron binding and thus cellular iron availability 8 . One category of those proteins consists of enzymes involved in cellular metabolism and energy generation, such as glycolysis, citric acid cycle, and electron transfer chain (ETC) (Fig. 3a, b). Previous studies have shown that the abundance of some of those enzymes increased significantly in response to elevated intracellular iron availability 33 . It was unclear whether increased abundance of those proteins was due to upregulated gene expression The dhbA-F operon is differentially expressed and regulated by AbrB. a A diagram demonstrating a putative co-repression of the dhbA-F operon by both Ferric uptake regulator (Fur) and the biofilm regulator AbrB. Repression by Fur is known to be iron-responsive, while AbrB-mediated repression is in response to biofilm signals 10,30 . Short lines below represent regions in each of the genes in the dhbA-F operon covered by pairs of primers in real-time PCR analyses. Three pairs of primers were designed to probe the dhbF gene expression due to its unusually large size. b Real-time PCR analyses to probe expression of the genes in the dhbA-F operon. Total RNAs were prepared from cells in the pellicle biofilm developed under two different ferric iron supplementations (50 vs. 0.5 μM FeCl 3 ). Expression of dhbA, dhbB, dhbC, and dhbD was found much higher in cells grown in media with 50 μM FeCl 3 than in media with 0.5 μM FeCl 3 . The relative transcriptional level in the latter media was set at 1; fold changes represent the fold differences in transcription under two different media conditions. "#" indicates that the transcription of dhbF was below detection limit using each of the three pairs of the probes under both media conditions. Data presented are the mean ± s.d. (n = 3). Error bars represent standard deviations. Statistical significance was assayed using unpaired t test via Prism 6. Stars indicate P values <0.01. c β-Galactosidase activities of cells bearing the P dhbA -lacZ transcriptional fusion. Cells were grown in MSgg supplemented with 50 μM FeCl 3 at 37°C with shaking. Cultures were collected periodically and assayed for β-galactosidase activities.
Reporter strains used here are YQ141(WT), YQ142(Δfur), and YQ255(ΔabrB upon increasing iron availability, and more importantly whether any of those proteins might be critical for biofilm formation. To test that, three different experiments were carried out. In the first experiment, a number of genes encoding metabolic enzymes containing iron as a cofactor were selected (highlighted in red, Fig. 3a, b). Deletion mutants for each of those genes were constructed and biofilm phenotypes of the mutants were examined. Some mutants had no noticeable biofilm phenotype, such as ΔlutABC 34 and ΔcitB (Supplementary Fig. 2A), while some others, although successfully constructed, were severely impaired in growth, such as ΔgltA, ΔnhdF, and ΔqcrA (data not shown). Those mutants were not further investigated. The remaining three mutants, ΔsdhC, ΔmenH, and ΔgpsA, were most interesting to us since they all showed a very severe biofilm defect (Fig. 3c), and no noticeable growth defect compared to the wild type ( Supplementary Fig. 1B). Interestingly, all three enzymes are known to be involved in respiratory electron transfer (Fig. 3b), while SdhC, part of the succinate dehydrogenase complex, is also involved in citric acid cycle (Fig. 3a) 35 . These results indicated that although impaired electron transfer had much less of an impact on growth under tested conditions, it was detrimental to biofilm formation in B. subtilis.
In the second experiment, expression of selected genes (sdhC, menH, gpsA, nhdF, and qcrA) encoding for enzymes involved in electron transfer 36 was investigated by RT-qPCR. Cells were collected from pellicle biofilms developed under two different media ferric iron conditions (0.5 vs. 50 μM FeCl 3 ). An elevated induction in expression of all five genes was observed under the 50 μM media ferric iron concentration, compared to that of 0.5 μM (Fig. 3e). When the qPCR assay was performed using cells collected under shaking conditions, an even stronger induction for all five genes was observed under the 50 μM media ferric iron concentration, compared to that of 0.5 μM (ranged from 10.5-to 28.3-fold increase, Supplementary Fig. 2B). These results suggested that the abundance of those proteins could increase significantly upon increasing concentrations of iron in the growth   Highlighted in red are enzymes whose activities depend on iron as a cofactor. NADH dehydrogenase, glycerol-3-phosphage (G-3-P) dehydrogenase, and succinate dehydrogenase are three primary substrate dehydrogenases denoting electrons to the electron transfer chain 67 . The electron transfer from succinate dehydrogenase to the electron carrier quinone can be blocked by a chemical inhibitor thenoyltrifluoroacetone (TTFA) 37 . c Pellicle and colony biofilm formation by the wild-type and three metabolic mutants of B. subtilis, ΔsdhC, ΔmenH, and ΔgpsA. Cells were incubated in MSgg at 30°C for 2 days before images were taken. Scale bar at the top panel, 5 mm; Scale bar at the bottom panel, 2.5 mm. d Inhibition of B. subtilis pellicle biofilm formation by TTFA. Cells were incubated in MSgg at 30°C for 2 days before images were taken. TTFA was added at the final concentration of 0, 50, or 100 μM at the beginning of inoculation. Scale bar, 5 mm. e Real-time PCR analyses to probe expression of the genes involved in electron transfer. Total RNAs were prepared from cells collected from the pellicle biofilm in MSgg but with two different ferric iron supplementations (50 vs. 0.5 μM FeCl 3 ). Expression of sdhC, ndhF, gpsA, menH, and qcrA was found much higher in cells in media with 50 μM FeCl 3 than in media with 0.5 μM FeCl 3 . The relative transcriptional level in the latter media was set at 1; fold changes represent the fold differences in transcription under two different media conditions. Data presented are the mean ± s.d. (n = 3 media. In the third experiment, we tested if blocking electron transfer by using small inhibitory chemicals could similarly impact biofilm formation in B. subtilis. 2-Thenoyltrifluoroacetone (TTFA), a small chemical inhibitor known to block the electron transfer from succinate dehydrogenase to the electron carrier quinone (Fig. 3b), was applied 37 . As shown in Fig. 3d, TTFA inhibited robust biofilm formation at concentrations (e.g., 50 μM) that did not significantly impact the growth of the cells ( Supplementary Fig. 1C). In summary, these results suggest that media iron concentrations have a strong effect on the expression of iron-containing ETC proteins, and that some of those proteins are essential for robust biofilm formation in B. subtilis.
Membrane potential is key to biofilm matrix production. One of the key activities of the ETC is to establish strong membrane potential. Because of the role of succinate dehydrogenase (complex II) in the ETC 38 , we speculated that the ΔsdhC mutant may have impaired membrane potential. To test that, cells were collected from the biofilm colony of ΔsdhC and membrane potential of the cells was measured by using the dye thioflavin T (ThT, Fig. 4a). According to previously published studies 39,40 , accumulation of this fluorescent dye inside the cells and thus the fluorescent density of the cells anti-correlate with the membrane potential of the cells. As shown in Fig. 4a, the majority of cells collected from the ΔsdhC colony biofilm demonstrated strong fluorescence (shown as cyan fluorescent protein (CFP)), indicating a low membrane potential of the cells. Cells were also collected from the wild-type colony biofilm (the outer region, Fig. 4b). In contrast, the majority of those wild-type cells demonstrated very weak fluorescence, indicating high membrane potentials of the cells (Fig. 4b). Interestingly, when cells were collected from the inner region of the wild-type colony biofilm, those cells showed intermediate fluorescence, suggesting that they had relatively lower membrane potentials compared to the cells in the outer region of the same colony biofilm (Fig. 4b).
Quantitative analyses of the fluorescence signals of individual cells were performed in parallel. The results further demonstrated the significant difference in membrane potential between the wild-type and the ΔsdhC cells, and even between cells in the outer and inner regions of the wild-type colony biofilm (Fig. 4d,  e). Lastly, a known membrane depolarizing agent, carbonyl cyanide m-chlorophenylhydrazine (CCCP) 41 , was applied to validate the ThT fluorescent dye-based analyses of membrane potential. The result showed that the addition of CCCP (1 mM) significantly impaired membrane potential of the wild-type cells as indicated by the strong fluorescence inside the cells (Fig. 4c), but slightly less severe than the impairment caused by ΔsdhC (Fig. 4d, e). Previous studies have shown that cells in the outer region of a colony biofilm tended to be robust matrix producers, while cells in the inner region were often old and weaker in matrix production 42,43 . Thus, a positive correlation between membrane potential and matrix production of the cells seemed to exist. To test if there is such a potential correlation, a dual-labeling technique was applied in that the ThT dye (shown as CFP) was used as an indicator for the membrane potential of the cells while cells also contained a P tapA -mKate2 fluorescent fusion to report the expression of the key biofilm matrix operon tapA-sipW-tasA 30 . Results from the assay showed that in the wild-type colony biofilm, cells from the outer region had both strong expression of the matrix gene reporter (mKate2) and strong membrane potential (which anticorrelates with CFP signals) (Fig. 4f). Cells from the ΔsdhC colony biofilm showed the opposite in that the majority of those cells were both weak in matrix gene expression and in membrane potential (Fig. 4g). Our results thus far supported the idea that membrane potential is closely linked to matrix gene expression and biofilm robustness in B. subtilis.
Large amounts of soluble iron are matrix associated. In all the experiments described above, when concerning iron concentrations, they were always referred to the amounts of ferric iron in the media. This was not satisfactory first because ferric iron was largely insoluble in the media, and second the intracellular concentrations of iron was unknown. Thus, we decided to investigate the intracellular concentrations of iron when cells were grown under two different media ferric iron concentrations (either 50 or 0.5 μM FeCl 3 ) by using inductively coupled plasma-mass spectrometry (ICP-MS). Surprisingly, results obtained from the assay showed that although iron concentrations in the media differed by 100-fold, the intracellular total iron concentrations differed by just less than one-fold (~70%, Fig. 5a). It was more surprising to recall that when 0.5 μM FeCl 3 was provided in the media, B. subtilis biofilm was severely impaired (Fig. 2d).
We had hypothesized on (i) either extracellular association of large amounts of soluble irons in the biofilm or (ii) heterogeneous distributions of iron within different regions of the cells in the biofilm. To test the above hypotheses, cells from either the inner or the outer region of the wild-type colony biofilm were picked. Both the intracellular iron concentration and the concentration of soluble iron associated extracellularly with the biofilm were measured similarly using ICP-MS (Fig. 5b). For the latter, a protocol was previously published by us to allow separation of cells from the extracellular matrix in a biofilm 44 . The most striking result obtained from those iron measurements was that large amounts of soluble irons (precipitated iron was removed by filtration) were found associated with the biofilm matrix for the cells collected from the outer region of the colony biofilm, more than 10-fold higher than the intracellular concentration of iron from the same cells (Fig. 5b). In contrast, the difference in the intracellular iron concentrations in the cells from the outer or center region was rather modest (~50%, Fig. 5b). These results depicted two different scenarios for possible roles of ferric iron in promoting biofilm formation, one being internalized and boosting the activities of iron-containing enzymes involved in electron transfer for establishment of strong membrane potential (as shown in Figs. 3e, 4a-e), and the other being solubilized (potentially by secreted DHB) and extracellularly associated with the biofilm matrix (as shown in Fig. 5b).

Extracellular iron could be involved in electron transfer.
Although it is possible that those extracellularly associated ferric iron function to facilitate matrix assembly, we speculated that those ferric irons were involved in respiratory electron transfer but extracellularly, either as terminal receptors or shuttles among different cells 45 . The rationale for the second possibility is that efficient electron transfer normally depends on molecular oxygen as the terminal receptor; however in the biofilm, internal layers of cells likely lack the access to molecular oxygen (which we show next). For those cells to perform electron transfer and establish membrane potential, they will need help of alternative electron receptors intracellularly and/or extracellularly.
To test this possibility, three experiments were performed. In the first experiment, accessibility to molecular oxygen by cells in different depth of a colony biofilm was tested by using an automated oxygen microelectrode piercing through the colony biofilm (see Methods). Both the outer and the center regions of the colony biofilm were chosen for oxygen measurement (Fig. 5b). In the outer region, there was a decline in oxygen levels from the top to the bottom of the colony biofilm (blue line, Fig. 5c) and the rate of oxygen depletion was initially slow but accelerated when reaching about 1/3 of the biofilm depth (−120 μm from the surface). In the center, the pace of oxygen depletion was more or less linear (red line, Fig. 5c). This result in general supported the idea that cells in the bottom of the colony biofilm may have difficulty in access to oxygen and performing oxidative respiration.
In the second experiment, the redox potential distributed along the vertical axis of the biofilm was measured. For technical reasons, this time we chose the pellicle biofilm formed in a beaker in order to increase the depth of the biofilm to suit the redox microelectrode used in the assay (Fig. 5d). Decreasing redox potential from the top to bottom along the vertical axis of a pellicle biofilm was observed (stronger reduction in the bottom, Fig. 5d). This result correlated with decreasing access of the cells to molecular oxygen (Fig. 5c). However, caution may be needed here since measurement of redox potential and that of molecular oxygen were performed in two different biofilm settings (pellicle and colony biofilms, respectively). Changing of redox potential implied that extracellular redox reaction (electron transfer) might happen in the bottom layers of the pellicle biofilm. We further probed the membrane potential status in parallel in the cells at the top and bottom layers of the pellicle biofilm. A published protocol 46 by us was modified to allow separation of cells at the top and bottom layers of a pellicle biofilm by applying a metal mesh in between (see Methods) (Fig. 5e). Cells at the top and bottom were separately collected and measured for membrane potential using the ThT fluorescence dye as described above. Cells at the top demonstrated strong membrane potential as indicated by very weak fluorescence (Fig. 5f). Interestingly, cells at the bottom also showed relatively strong membrane potential (Fig. 5f), implying active respiratory electron transfer in those cells even without sufficient access to molecular oxygen. We speculated that respiratory electron transfer and establishment of membrane potential in those cells relied on large amounts of matrix-associated extracellular irons as the electron receptors or shuttles. Lastly, we also looked at the viability of the bottom (and the top) layers of cells in the pellicle biofilm by using live/dead staining. The result showed that cells at the bottom of the pellicle biofilm had a similar live/dead ratio when compared to the cells at the top (Supplementary Fig. 3).
In the third experiment, in order to test if extracellular electron transfer (EET) does occur in the B. subtilis biofilm, a cyclic voltammetry (CV) measurement (Fig. 6a) was performed by coating the wild-type (WT) biofilm on a working reticulated vitreous carbon (RVC) foam electrode. As shown in Fig. 6b, two redox peaks appeared with potentials at around −40/−170 mV (pointed by arrows in the red curved lines). This indicates that the biofilm coated on the RVC foam electrode can reduce Fe 3+ to Fe 2+ in the aqueous solution. In the two control measurements, one with the biofilm medium MSgg only and the other with MSgg supplemented with 1 µg ml −1 of DHB, no clear redox peak was observed (Fig. 6b).

Discussion
Iron is an essential nutrient element for growth in almost all organisms. A unique question investigated in this study is why excessive amounts of ferric iron, hundreds fold higher than for normal growth, are needed to promote robust biofilm formation in the bacterium B. subtilis. Two important roles of iron in promoting B. subtilis biofilm formation are proposed in this study. For one, large amounts of iron were found associated extracellularly with the biofilm matrix. Results from electrochemical assays suggested that EET occurred in the B. subtilis biofilm and those matrix-associated extracellular irons could be involved in the EET by potentially acting as electron receptors and shuttles. EET may be especially important for cells located in the deep layers of the biofilm and thus blocked from access to molecular oxygen. For the other role of iron, our evidence supported the idea that a small proportion of iron was internalized and played a critical role in boosting the production of multiple iron-containing enzymes essential for intracellular electron transfer (IET) during oxidative respiration. Both the EET and IET could help to establish strong membrane potential, which we then showed, was linked to strong matrix gene expression during B. subtilis biofilm formation.
Biofilms are considered a very heterogenous environment 47 . Cells in different spatial locations in the biofilm may adopt different strategies for respiration and energy generation. Here we propose a model on respiratory electron transfer during B. subtilis biofilm development (Fig. 7). For cells in the upper layers where oxygen is sufficient, they perform respiratory IET using molecular oxygen as the terminal electron acceptor and electrons being donated to the ETC 48 by various substrate dehydrogenases. For cells in the bottom layers of the biofilm where oxygen becomes very limited or even absent, they perform EET using extracellular matrix-associated iron as the terminal electron acceptor. In light of the above model, we would like to point out several limitations in this study. First, evidence on the presence of large amounts of extracellular iron in the deep layers of the biofilm (instead of the entire biofilm) is still needed to support the above model. Second, for redox potential measurement, only pellicle biofilm (but not colony biofilm) was applied since the measurement of redox . Error bars represent standard deviations. Statistical significance was assayed using unpaired t test via Prism 6. Star indicates P value <0.05. b Measurement of intracellular and extracellular (matrix-associated) iron concentrations for cells collected from either the inner or the outer region (indicated by the red and green squares) of a wild-type biofilm colony. The biofilm colony was developed in regular MSgg agar media (50 μM FeCl 3 ) at 30°C for 2 days prior to collecting cells. Data presented are the mean ± s.d. (n = 5). Error bars represent standard deviations. Statistical significance was assayed using unpaired t test via Prism 6. The same letter on the bars of each column indicates no significant difference. c Measurement of molecular oxygen concentrations along the vertical depth of the colony biofilm. Both the inner (red line) and the outer (blue line) region of the colony biofilm were picked for analysis. d Measurement of redox potential along the vertical depth of the pellicle biofilm. e A cartoon demonstration of the setting to use a metal mesh (Nalgene) to separate top and bottom cells in a pellicle biofilm. f Cells from the top and bottom of a pellicle biofilm were collected and assayed for membrane potential status using the fluorescence dye ThT as described previously. Scale bar, 10 µm. Source data are provided as a Source data file needs aqueous environment to form a circuit between the redox probe and the reference probe and only pellicle biofilm can satisfy this requirement. As colony and pellicle biofilms are two different settings, caution needs to be taken when accessing the results obtained from either colony or pellicle biofilms alone.
In the biofilm environment, B. subtilis can be considered an electrochemically active microorganism (EAM and likely takes advantage of both IET and EET to perform respiration in order to establish membrane potential and generate energy. EAMs are microorganisms that can transfer electrons from cells to extracellular electron acceptors such as minerals, contaminants, and electrode 45,49,50 . So far, more than 100 EAMs have been isolated or identified 49 . Most of them are Gram-negative bacteria and belong to Proteobacteria phylum 51 . Most studies on the EET mechanism are based on Shewanella spp. and Geobacter spp. 52,53 . More recently, Gram-positive bacteria B. megaterium and Enterococcus faecalis also showed the capacity of EET 54,55 . Studies have shown that oxygen accessibility is clearly a key factor driving biofilm heterogeneity. Here we present another interesting example of metabolic heterogeneity in cells within the B. subtilis biofilm.   subtilis is an electrochemically active microorganism (EAM) and likely takes advantage of both IET and EET to perform respiration and establish membrane potential. During EET, ferric iron could get free electron from any of the components in the electron transfer chain, including various initial substrate dehydrogenases, quinone, or cytochromes. The precursor 2,3-dihydroxybenzoate (DHB), but not bacillibactin, is an iron-binding molecule for iron solubilization and acquisition under iron-rich conditions. DHB-iron complex will be taken into the cells via the FeuABC-YusV-mediated transport system 9 In this study, a link between membrane potential and strong matrix gene expression was demonstrated during B. subtilis biofilm formation. Traditionally, the importance of membrane potential has been discussed in the context of ATP generation, bacterial motility, chemotaxis, molecule transport, and cell division. In recent studies, membrane potential has been demonstrated to be important for the electrical communication within cells during the development of bacterial communities in B. subtilis 39,56 . However, why membrane potential is essential for biofilm formation in B. subtilis is still unclear. Previous studies have suggested the role of membrane potential in transport of key metabolites such as glutamate 40 . Glutamate is a charged amino acid, whose uptake has been shown to depend on ion gradient and membrane potential 57 . Glutamate is a key component in the biofilm-inducing media for B. subtilis 3 and its absence will result in structurally much weaker biofilms 58 .
We also discovered an adaptive strategy for iron acquisition during B. subtilis biofilm formation. Our evidence suggested that in order to utilize necessary amounts of ferric iron in the media, B. subtillis cells preferentially produced and relied on the siderophore precursor DHB (a monomeric intermediate of the trimeric bacillibactin) instead of the final siderophore bacillibactin. We argued that this could be an important adaptive strategy for the bacterium for several reasons. First, since the size of the last dhbF gene is very large (7.13 kb), expressing this gene and translating the protein may be very costly and possibly error prone. Second, although bacillibactin has an extremely high binding affinity toward ferric iron, uptake and then release of free iron into the cytosol of the cell is not straightforward. The release of free iron needs an extra hydrolysis step due to extremely high binding affinity of bacillibactin and the hydrolyzed bacillibactin is not reusable 59 . Therefore, alternating production of DHB and bacillibactin depending on iron availability in the environment seems to be an important adaptive strategy. In fact, in other species such as E. coli, the homologous gene of dhbF, entF, is separated from other ent genes, while clustered with the fes gene encoding an enterobactin hydrolase ( Supplementary Fig. 4) 60 . This provides a possible genetic basis for alternative production of the two iron-binding molecules in other bacteria 60 . The putative mechanism for differential expression of the dhbA-F operon is unknown. We suspect that a transcription attenuation-like regulation between dhbF and the gene upstream of dhbF may be responsible for substantially attenuated expression of dhbF under iron-rich conditions.
Strain construction. All the insertional deletion mutants used in this study (listed in Supplementary Table 1) in the B. subtilis 168 background were purchased from the Bacillus Genetic Stock Center (http://www.bgsc.org) and introduced into NCIB 3610 via transformation following a previously described protocol 63 . In order to make marker-less deletion in the bacillibactin biosynthetic genes, the dhbA, dhbB, dhbC, dhbE, and dhbF insertional deletion strains marked with an erm-HI resistance cassette in the B. subtilis 3610 background were transformed with the plasmid pDR244 (a temperature-sensitive suicide plasmid with a constitutively expressed Cre recombinase gene) 64 . Transformants were selected on LB agar plates supplemented with spectinomycin at 30°C (permissive temperature). Transformants were then streaked onto LB agar plates and plates were incubated at 42°C (non-permissive temperature). Cells from single colonies were spotted onto LB plates, LB plates supplemented with erythromycin and lincomycin, and LB plates with spectinomycin. A transformant that grew only on the LB agar plate but not on LB agar plates supplemented with any of the three antibiotics, likely both lost the erm antibiotic resistance cassette and was cured for the pDG244 suicide plasmid 64 . Those transformants were selected and verified for in-frame deletion using PCR. To construct reporter strains with the P dhbA -lacZ fusion, the promoter sequence of the dhbA-F operon was amplified by PCR by using the primers P dhb -F and P dhb -R (all primers are listed in Supplementary Table 2) and the genomic DNA of 3610. PCR products were then cloned into the vector pDG1728 to make a P dhbA -lacZ fusion. The integration of the P dhbA -lacZ fusion or the P tapA -mKate2 reporter fusion, which was from the strain TMN503 65 into the chromosomal amyE locus of virous B. subtilis strains and verification of such integrations were described in a previous publication 66 .
Microscopic imaging. For imaging of colony and pellicle biofilms, a Leica MSV269 dissecting microscope with a Leica DFC2900 camera and ×4 magnification was used. Same exposure and acquisition settings were applied to all colony and pellicle biofilm samples. For single-cell fluorescence imaging, cells were cultured as described above. After incubation for indicated times, colony biofilms were collected, resuspended in phosphate-buffered saline (PBS), and disrupted with vortexing. One milliliter of cell resuspension was mildly sonicated with the 5-s pulse at 1.5 output scale for three times (Branson, Model W185). Cells were centrifuged at 14,000 rpm for 1 min and washed briefly with PBS. For imaging, 1 μl of the resuspension was placed on a 1% (w/v) agarose pad and covered with a cover slip. Cells from three independent biological replicates were imaged using a Leica DFC3000 G camera on a Leica AF6000 microscope. Non-specific background fluorescence was determined by quantifying wild-type cells with no reporter fusion. Imaging of samples collected from different time points was conducted using the same exposure settings. For observation of the ThT fluorescence dye, the setting for observing CFPs was used, with the excitation wavelength at 426-450 nm and the emission wavelength at 502-538 nm. For observation of mKate2, the excitation wavelength was set at 540-580 nm and the emission wavelength at 610-680 nm.
Biofilm assays. For colony biofilm formation, cells were grown to exponential phase in LB broth and 2 μl of the culture was spotted onto MSgg media solidified with 1.5% (w/v) agar. The plates were incubated at 30°C for 2 days. For pellicle biofilm formation, cells were grown to exponential phase in LB broth, and 3 μl of the culture was inoculated into 3 ml of MSgg liquid media in a 12-well microtiter plate (VWR). For treatment with DHB, TTFA, or CCCP, the chemical compound was diluted and added to the medium at the indicated final concentrations. Images of colony and pellicle biofilms were taken as described above using a Leica MSV269 dissecting scope and a Leica DMC2900 camera. For separation of the top and bottom cells in pellicle biofilms, pellicle biofilms were developed as described above but with a piece of metal mesh (Nalgene) at the surface of the biofilm liquid media. After 2 days of incubation, the metal mesh was lifted, pellicle biofilms cells attached to the top and bottom of the metal mesh were collected, and cells were then treated as described for microscopic analyses.
Assays of β-galactosidase activities. Cells were cultured in MSgg medium at 30°C with shaking. One milliliter of culture was collected at each indicated time point and cells were centrifuged down at 5000 rpm for 10 min. Cell pellets were suspended in 1 ml Z buffer (40 mM NaH 2 PO 4 , 60 mM Na 2 HPO 4 , 1 mM MgSO 4 , 10 mM KCl, and 38 mM β-mercaptoethanol) supplemented with 200 μg ml −1 lysozyme. Resuspensions were incubated at 37°C for 15 min. Reactions were started by adding 200 μl of 4 mg ml −1 ONPG (2-nitrophenyl-β-D-galactopyranoside) and stopped by adding 500 μl of 1 M Na 2 CO 3 . Samples were then briefly centrifuged down at 5000 rpm for 1 min. The soluble fractions were transferred to cuvettes (VWR), and absorbance of the samples at 420 nm was recorded using a Bio-Rad spectrophotometer. The β-galactosidase-specific activity was calculated according to the equation (Abs 420 /time × OD 600 ) × dilution factor × 1000. Assays were conducted in triplicate.
Iron quantification by ICP-MS. Quantification of matrix and intracellular iron followed a published protocol with minor modifications 10 . For the quantification of matrix-associated iron, the colony or pellicle biofilms after incubation at 30°C for the indicated time, were collected and disrupted in 4 ml of 1× chelex-treated PBS buffer with pipetting. Samples were then mildly sonicated with 5-s pulse at the 1.5 output scale for three times (Branson, Model W185), normalized to OD 600 = 1.0 by using 1× chelex-treated PBS buffer. One milliliter of normalized samples was centrifuged at 14,000 rpm for 1 min, and the supernatant was ready for quantification of the matrix-associated iron. For measurement of the intercellular iron concentration, cell pellets after centrifugation were washed once with buffer 1 (1× PBS buffer, 0.1 M EDTA) and then twice with buffer 2 (1× chelex-treated PBS buffer). Cell pellets were resuspended in 400 μl of buffer 3 (1× chelex-treated PBS buffer, 75 mM NaN 3 , 1% Triton X-100) and incubated at 37°C for 90 min. Lysed cells were centrifuged and the total protein content was quantified using a Bradford assay. Samples were then mixed with 600 μl buffer 4 (5% HNO 3 , 0.1% (v/v) Triton X-100) and heated in a 95°C sand bath for 30 min. The debris was removed by centrifugation. The total iron in all samples was analyzed by Bruker Aurora M90 ICP-MS. The OD 600 values of the biofilm samples were later converted to colonyforming units using standard plating techniques. The total iron concentration was presented as molar per cell (mean ± SD; n = 10).
Electrochemical characterization. To characterize electron transfer between the B. subtilis biofilm and ferric ions in electrolyte, CV was performed in a threeelectrode cell by an electrochemical workstation (BioLogic). An RVC foam (3 cm × 3 cm × 6.5 mm) was used as the working electrode, while a platinum plate (1 cm × 1 cm) and Ag/AgCl (saturated, KCl) were used as the counter and reference electrode, respectively. The MSgg medium, which contained 50 μM of FeCl 3 as the unique electron-acceptable component in this system, was directly used as electrolyte. The pellicle biofilm of B. subtilis was coated on the surface of RVC foam electrode prior to the measurement. CV tests were operated in the range of −1.0+ 1.0 V at a scan rate of 20 mV s −1 .
Oxygen profiling of biofilms. For oxygen profiling, colony biofilms of B. subtilis 3610 were developed as described above. A Unisense Microprofiling System with a 50-μm-tip Clark-type oxygen microsensor (Unisense OX-50) was used to measure the oxygen concentrations in the colony biofilm. The oxygen microsensor was calibrated according to the manufacturer's instructions, and measurements were taken throughout the depth of the biofilm. During the measurement, the step size was 10 μm, the measurement time period was 5 s, and the wait time between different measurements was 5 s. Three different colonies were tested, and representative data were shown.
Redox profiling of biofilms. For redox profiling, 3610 cells were grown to exponential phase in LB broth at 37°C with shaking. One hundred microliters of the culture was then inoculated into 100 ml of MSgg liquid medium in a 200 ml beaker and incubated at 30°C for 3 days for pellicle biofilm development. For the assay, a 500-μm-tip redox microelectrode with a reference microelectrode (Unisense RD-500 and REF-RM) was used to measure the extracellular redox potential in the pellicle biofilm by using the same workstation used in the oxygen profiling. After calibrating the redox micro-electrode according to the manufacturer's instructions, redox measurements were taken throughout the depth of the pellicle biofilm. During the measurement, the step size was 50 μm, the measurement time period was 5 s, and the wait time between measurements was 5 s. The redox potential was set to zero at the surface of the pellicle biofilm, and relative values within the pellicle biofilm were plotted. Three different pellicle biofilms were used for measurement and representative data were shown.
Measurement of membrane potential. To measure the membrane potential of the cells in the biofilm, MSgg agar plates supplemented with 10 μM ThT, the fluorescent dye for membrane potential measurement was used. ThT is a cationic dye that acts as a Nernstian voltage indicator, which accumulates in cells as membrane potential decreases (accumulation thus negatively correlates to cell membrane potential) and the fluorescence intensity in the cells therefore increases 39 .
Quantitative real-time PCR. Total RNAs of the cells in different treatment groups were extracted by using TRIzol (Invitrogen) according to the manufacturer's instructions. Isolated RNAs were reverse transcribed into singlestranded complementary DNA (cDNA) using a High Capacity cDNA Reverse Transcription Kit (Applied Biosystems). Primers used for RT-qPCR are listed in Supplementary Table 2. qRT-PCR was performed by using Fast SYBR TM Green Master Mix (Applied Biosystems) and a Step One Plus Real-Time PCR system (Applied Biosystems). The 16S rRNA gene was used as an internal reference. The relative expression of specific genes was calculated by using the 2 −ΔΔCT method 36 .
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.
Data availability