Ferrous Iron Binding Key to Mms6 Magnetite Biomineralisation: A Mechanistic Study to Understand Magnetite Formation Using pH Titration and NMR Spectroscopy

Abstract Formation of magnetite nanocrystals by magnetotactic bacteria is controlled by specific proteins which regulate the particles’ nucleation and growth. One such protein is Mms6. This small, amphiphilic protein can self‐assemble and bind ferric ions to aid in magnetite formation. To understand the role of Mms6 during in vitro iron oxide precipitation we have performed in situ pH titrations. We find Mms6 has little effect during ferric salt precipitation, but exerts greatest influence during the incorporation of ferrous ions and conversion of this salt to mixed‐valence iron minerals, suggesting Mms6 has a hitherto unrecorded ferrous iron interacting property which promotes the formation of magnetite in ferrous‐rich solutions. We show ferrous binding to the DEEVE motif within the C‐terminal region of Mms6 by NMR spectroscopy, and model these binding events using molecular simulations. We conclude that Mms6 functions as a magnetite nucleating protein under conditions where ferrous ions predominate.


Introduction
Essential to many organisms, iron is an important component of many biological processes. [1] Due to the inherent redoxa ctivity and pH sensitivity of this transitionm etal its presence in cells must be carefullyc ontrolledt op reventp otentially harmful effects from reactive oxygens pecies [1c] or iron precipitation. [2] Many proteins have therefore evolved to utilise, control and harness the useful capabilities of iron whilst minimizing any potentially damagings ide effects. [1a, b] In the case of magnetotactic bacteria (MTB), they have evolved to take advantage of the magnetic characteristics of certain iron oxidesb yp roducing biogenic magnetic nanoparticles [3] withini nternal lipid vesicles termedm agnetosomes. [4] These vesiclesa re in effect an anoreactor for the precise synthesis of, most commonly,t he iron oxide magnetite (Fe 3 O 4 ). [3c, 5] The formation of nanocrystalline magnetite is tightly controlled by as uite of biomineralisa-tion proteins which are present within the lipid membrane of the magnetosome. [6] The nucleation,crystal growth and regulation of the final size and shape of the particle are strictly regulated by these proteins. As ingle strain of MTB harbours ah ighly uniform population of nanoparticles;h omogeneous in termso fs ize, shape,a nd chemical composition. Research is currently focusingo ni dentifying andc haracterising these bio-mineralisationM ms (magnetosome membrane specific) proteins in order to elucidate ad etailed understanding of iron oxide biomineralisation.
One key protein found tightly bound to the magnetite particles of Magnetospirillum magneticum AMB-1 [7] is a6kDa protein, Mms6. Mms6 comprisesahydrophobic N-terminal region and ah ydrophilic C-terminalr egion (KSRDIESAQSDEEVELRDA- Figure 1. Mms6 in magnetotactic bacterial magnetosome.I nb lue is the hydrophilic acid-rich region and in pink is the hydrophobic membraneregion with the sequence below.Experimental schematic showing pH recording duringa ddition of base to ferric/ferrous solution with or without Mms6 micelles is also illustrated. LA) whichc ontains ah igh number of residues with acidic sidechains that have been implicated in the ferric iron binding capabilityo ft he protein (Figure 1). [7,8] If the mms6 gene is removed from MTB (so Mms6 protein is not produced) the resulting nanoparticles which form are both small and poorly formed compared to the wild-type. [9] Mms6 has therefore been classifieda sa ni mportant member of the magnetite biomineralisationm echanism. Due to its amphiphilic sequence, purified Mms6 self-assemblesi nto micellar structures. [8a] These large aggregatesa re able to both bind and accumulate ferric ions from solution, [7,8,10] and purified assemblies of Mms6 on abiomimetic magnetosome interior surface demonstrate magnetite formation properties,i ndicating that Mms6 can act as ap otential iron oxide nucleation site for subsequent crystal formation. [11] In addition, the acid rich C-terminal part of Mms6 has been studied and was found to exhibit some similar characteristics to the full-length protein such as iron binding and al imited ability to affect magnetite crystal growth. [8a, 10, 12] Magnetic nanoparticles have received much attention due to their potential use in aw ide range of different applications spanning various scientific disciplines and technologies. [13] These range from the magnetic components of ferrofluids [14] and data-storage devices, to precision applications in medical diagnostics and therapies. [15] For this latter use, magnetite is one of the most desirable materials due to its magnetic properties and general low toxicity. [15a] There are an umber of synthetic routes for magnetite nanoparticles including thermald ecomposition, and high temperature methods in the presence of organic surfactants. [16] An alternative and simple method of producing magnetite nanoparticles is by room temperature co-precipitation (RTCP). This results in particles with ah igh size polydispersity making them unsuitable for critical biomedical applications. Taking inspiration from nature where MTB synthesize magnetite nanoparticles with ah igh degree of control over size and shape, purified Mms6 has been introduced into RTCP and other reactions. [7,17] Ther esulting products demonstrate improved homogeneity in both size and mineral type, suggesting that Mms6 is ablet oc ontrolt he formation of magnetite nanoparticles in vitro. [7,17,18] However,t he exact mechanism by which Mms6 achievest his type of control both in vivo and in vitro remains poorly understood.
In this study we analyse the effecto fM ms6 during as imple RTCP of magnetite by monitoringd ifferences in the pH profile during the reaction process. This was performed under ar ange of different reactionc onditions to build up ar igorousa nd detailed picture of Mms6 activity at different ratios of ferric and ferrous iron. Ap revious study of iron oxide formationu sing this approach [19] has provided am ethodology by whicht oi nvestigatet he effect of these additives on the crystallisation process.W ef ind that Mms6e xerts its greatest influence over the process in the pH range where ferrous ions are precipitated out of solution (> pH 7). Previous studies have shown that Mms6 interacts with ferric iron butt here is, to the best of our knowledge,c urrently no analysiso fa ny potential interaction with ferrous iron. Therefore, to complemento ur study we have used 2D NMR spectroscopy to investigate the ferrous-binding activity of the C-terminal peptider egion of Mms6, whichr e-veals that the peptide has specific interactions with ferrous iron centred around the DEEVE acidic residue cluster. Further insight is gained throughm olecular dynamics simulations of this clustera nd itsi nteractions with both ferric and ferrous iron. This study therefore provides an ew perspective on the activity of the Mms6 protein by suggesting it is interacting with both ferric and ferrous iron during the magnetite crystallisation process, and that it is the predominanta nd specifici nteraction with ferrous iron that is the key to driving the reaction towards magnetite. This finding has implications for our understanding of magnetite biomineralisation and how we can controlm agnetite formation in biomimetic magneticn anoparticle synthesis for wider nanotechnology applications.

Results
Room temperature co-precipitation of magnetite (RTCP) The precipitationo fi ron oxidesf rom solution is complex [20] and proceeds through an umber of intermediate iron oxides. A comprehensive study of the co-precipitationo ff erric and ferrous salts by Ruby et al. [19,21] employed careful pH titration measurements to identify processes occurring during precipitation reactions. The quantities of Mms6 protein availablem ean we must use much lower concentrationsa nd reaction volumes than those used by Ruby et al. which makes the experiment very sensitive.I nt his study we use as imilart itration method to monitor the progress of the reaction and the nature of the intermediates formed both with and without the addition of Mms6. This allowsu st ou nderstand if, and how,the protein affects the co-precipitation process and to explain how Mms6 is able to exert control over the size and the mineral species formed, specificallym agnetite, when added to aR TCP reaction. [12] Here, as olutiono fm ixed valence iron salts is precipitated by slowly raising the pH with the addition of base. There are an umber of reactions present in this system [19b] that can lead to several iron oxide and iron oxyhydroxide products or intermediates,n amely the mixed ferric/ferrous minerals magnetite ( Figure 2a shows titration curves for the co-precipitation of ferric and ferrous iron sulfates with NaOH in the absence of Mms6. The molar ratio of Fe 3 + to total iron is represented by X (e.g.,F e 3 + /Fe 2 + ratio of 1:2g ives X = 0.33) and ranges from 0 to 1. R is the molar ratio between the concentration of NaOH and total iron. For these experiments an effective rate of 0.05 R min À1 was used for the addition of base, meaning that in our experiments R increases linearly with time.
The step titration curvess hown in Figure 2h ave three plateaus separated by two steps defined by the equivalence points E 1 (step between plateaus1and 2, shown by circles in Figure 2) and E 2 (step between plateaus 2a nd 3, shown by crossesi nF igure 2). E*i sd efined as the central point of the second plateau (shownbya na sterisk in Figure 2a).
At the lowest pH conditions the first plateau describes OH À consumption through precipitation of ferric basic salts [e.g., the ferric oxyhydroxide, schwertmannite;E q. (5)] due to the lower solubility of ferric ions in the presence of base. The middle plateau corresponds to the conversion to, and formation of, ar ange of possible iron minerals: ferrous hydroxide, green rust, magnetite and goethite [Eqs.
(1)-(4)] in variousp roportions, increasing in amountso ft he latter minerals as the ratio of ferric to ferrous ions increases ( Figure 2). Interestingly, the ferric oxyhydroxide begins to convert into green rust and magnetite by incorporationo ff errous ions and through elec-tron transfer.O nce all the ferric solids have been converted, the excessf errous ions precipitate as ferrous hydroxide. [19b] These mixed oxidesa re retained up to the endo ft he titration at pH 12.5. At this point the reaction mixture has ah ighly negative redox potential( approximately À750 mV;s ee Supporting Information Figure S1) demonstrating its high propensity to oxidise.I fl eft to age, overnight, with as mall amounto fo xidation, mineral dehydration can occur,c onverting other iron oxidess uch as green rust (whichc an be considered as intermediates)t om agnetite. The dark green solution becomes black.
Using the methodology and nomenclatureo fR uby et al. [19a] we plotted the positions of equivalence points derived from the pH titration data on am ass-balance diagram of R versus X (Figure 2b). The first equivalence pointsa tE 1 lie along the line R 1 = 2.75 X.T his represents the theoretical relationship for the formation of the sulfated ferric oxyhydroxide schwertmannite, in the first plateau [see Eq. (5), balanced for z = 1/8, thus OH/Fe ratio = 2.75:1]. [19a] The relative quantity of base (R)r equired is linearly dependent on the ratio of ferric/ferrous species( increasinga sF e 3 + increases;F igure 2a and b) due to the fact that there are increasingq uantities of ferric iron to precipitate. The second equivalence point, E 2 ,d escribes the nature and proportions of insoluble iron oxides that are formed in the second plateau (depending on the initial X ratio). As discussed above,t his could be am ixture of green rusts, magnetite and other ferrous/ferric hydroxide nanoparticles (Supporting Information Figure S2 shows products of the reactiona tX = 0.3 and 0.5 showingacombination of green rust and magnetite at lower X ratio, and more pure magnetite at the higher). Stoichiometric magnetite would be formed at X = 0.67 (i.e.,2Fe 3 + /1 Fe 2 + ). However,i ts hould be noted that extracting av ertical line from the initial X ratio to the products is as implistic representation of the reaction. This reaction has am aturation step after the titration where as mall amount of oxidation occurs slowly,o vernight,t of orm the most stable products,r esulting in an on-vertical conversion to the most stable products after this E 2 point. In practice, many magnetite nanoparticle synthesis schemesr eport highers tarting ratios of Fe 2 + with 0.33 X 0.6667, in recognition of partial oxidationi nt he final stages of the process. [7,17] Generally,i ncreasing the amount of Fe 2 + in solution results in lower quantities of magnetite and higherq uantities of ferrous (oxy)hydroxide particles, while too high aq uantity of Fe 3 + will result in ferric (oxy)hydroxide such as goethite. In summary,t he titrations shown in Figure2demonstrate the expected pH-dependent formationo ff errous hydroxide, green rusts and magnetite, as well as the information to understand how and when these are forminga nd in what quantities.

Influence of Mms6 on ferrous iron components during the co-precipitation of magnetite
Previous reports support the finding thatm ixed iron oxide populations are often obtainedf or RTCP.T his is seen for example in the control sample TEM images presentedi nA memiya et al. [17a] However,i nt he presence of Mms6, Amemiya et al. Figure 2. a) pH measurements during roomtemperature co-precipitationfor total iron concentrations of 50 mm at arate of 0.05 R min À1 with equivalence points highlightedbyc ircles (E 1 )and crosses( E 2 ). The central equivalence point is marked by an asterisk (E*). Thesed ata are similar to results obtained at much higherc oncentrations. [19a] b) Massb alanced iagramfor the formation of various iron oxides showing the equivalence points from (a). The error was too small to be shown( E 1 = zero (to 3d ecimal points) and E 2 = AE 1.3 %). GR = green rust and SWM = schwertmannite. showedt hat the homogeneity of the sample was greatly enhanced with respect to both the iron oxide phase (controlled for magnetite, with removal of undesired iron oxide by-products) and narrowing of the size distribution of magnetite nanoparticles. It therefore appears that Mms6 can controlt he formation of the specific mineral magnetite, as well as its size.
To study Mms6 further we added purified protein to the reaction at ac oncentration of 10 mgmL À1 of reaction solution. The pH titration profile was recorded at variousv alues of X, shown in Figure 3a.T he R positions of equivalence points E 1 and E 2 were plottedo nt he mass-balanced iagram to allow comparison between the control titrationa nd the corresponding Mms6 experiment (Figure 3c).
From thesep Hd ata we can see that there is negligible differenceb etween the titrations at low pH, and thus the R values of E 1 do not alter between the control and Mms6 experiments, with all falling very close to this ideal line for the formation of schwertmannite, in agreement with previous control studies. [19a] Results in the literature have shown that Mms6 is able to bind ferric ions. [7,8] However,t his is not noticeably the case in this study and as such we propose that Mms6 may only bind to ferric ions at higherp H, or if there is no competing ferrous ions, or that the binding to soluble ferric ions is at concentrations too low to detect in this titration. It should be noted that the methodology used in this study is likely to only detect the formation and precipitation of solids from solution and is unlikely to detect the binding of iron ions by Mms6 alone.T he consistency between control and Mms6-mediated RTCP reactions at E 1 indicatesM ms6 does not alter the balance of iron speciespresent at low pH.
With Mms6 present, the formation of the second plateau occurse arlier and at lower pH for X 0.4 (Figure 3a and b). This shows that Mms6-containing reactions begin to form new precipitates at al ower pH after the same quantity of base has been added compared to the control reactions. This indicates that in the presence of Mms6, Fe 2 + and hydroxide ions combine more readily with the ferric oxyhydroxidet og enerate the mixed valence iron mineralsm agnetite and green rust. [19] Here it shouldb en oted that magnetite precipitates at al ower pH than green rust [see Eqs.
(2) and (3)],t hus comparably more magnetite is being precipitateda tt his second plateau in the Mms6 mediated reactions than the controls for X 0.4, demonstratingt hat Mms6 is able to direct the nucleation and formation of magnetite preferentially.T he differenceb ecomes negligible when the reaction is performed under conditions where the ferric and ferrous iron are equallyb alanced or ferric ions predominate (Figure 3a and b).
For each value of X,from the E 2 positioni nt he mass-balance diagram (Figure 3c,S upporting Information Ta ble S1) the relative amounts of each of the possible minerals (ferrous hydroxide, green rust and magnetite)w ere calculated (Figure 4a; note 0.2 X 0.6 as the higher X point is outsidethis mass balance mineral regime and shows negligibled ifference between control and Mms6 samples). These relative amounts are shown in Figure 4b (values in Supporting Information Ta bleS 2). At X = 0.2 (the most extreme ferrous-rich ratio tested) the control reaction products are dominated by green rust and ferrous hydroxide species (as the excess ferrous ions are precipitated) with negligibly small quantities of magnetite being produced. However, at this same X value with the addition of Mms6, approximately 20 %o ft he mineral species formed is magnetite, as imilar quantity to the green rust, consuming the available ferric ions. Our data show that Mms6 is acting to promote the formation of magnetite at lower X values. This magnetite increasei sl arger in the ferrous iron rich reactions (i.e.,w hen X is lower), with the quantity of magnetite coinciding with the protein free reactions at around X = 0.4. Then at higherv alues of X there is less effect on the quantity of magnetite formed (as this is able to occur more readily anyway). However,m ore green rust is formed at the expense of ferrous hydroxide (compared to the control), and this is more able to matureand convert to magnetite under these conditions (Figure 4b). Furthermore,t he magnetic data follow this trend (Supporting Infor-mationF igure S3 and Ta ble S3).
Ta ken together,o ur analysis indicates that the main role of Mms6 is to sequester and bind ferrous ions, particularly at the later,h igher( post E 1 )p Hs tages where they would normally start to precipitate.T he result of this activity is that Mms6p romotes the formation of magnetite under unfavourable ferrousrich conditions when it is less able to yield magnetite as aproduct in the controlr eaction( Figure 4b). Mms6c ould thus be considered as acting as a" mineral/ferrous ion buffer" seemingly enhancing the propensity to nucleate magnetite, even when this does not normally occur, as in the case where X = 0.2, and directingi ts precipitation towards magnetite rather than other mixed valence or ferrous salts.

Ferrousand ferric iron interactions with Mms6-derived peptides
Despite extensive study there is currently no structurali nformation on Mms6 or its C-terminal iron-binding domain. We have attempted protein crystallization and NMR spectroscopy on the full-length protein but this has provedi ntractable due to protein aggregation and disordered structures. However,i n order to obtain structural informationr elatingt om etal binding, the whole protein (particularly the proposed hydrophobic self-assembly part) is most likely not required. The C-terminal region is the mosta cidic and thus proposed to be the iron binding site. There is also some evidence that the C-terminal region alone can show some activity in magnetite formation reactions, suggesting it probably retains the ability to bind iron ions even when the remaining two thirds of the full length protein is removed. [12] In order to address the questions of the metal-bindingp roperties of the protein we have employed a2 DN MR spectroscopica pproacht oi nvestigate possible structure and iron binding characteristics (particularly and uniquelyc onsidering ferrous iron binding) of a2 0a mino acid peptides equence consisting of the acidic regiono fM ms6 under physiological pH conditions. We designate the Ac-KSRDIESAQSDEEVELRDAL-Am peptide as C20Mms6 (N and Cterminus werem odified by acetylation and amidation, respectively,t on egatep ossible interactions with the termini which would not appear in the full-length protein at these residue positions). Crucially the C-terminal region does not have the same micellar-forming properties as full-length Mms6, which simplifies its study by NMR spectroscopy.
We performed TOCSY and ROESY NMR experiments on the C20Mms6p eptidei ns odium phosphate buffer at pH 7( pH of the species formed at the second plateau where we see Mms6 activity) in order to obtain ac omplete assignment of the peptide for the backbone plus sidechains as far as Hb or Hg.Anassignment can be found in Supporting Information Table S4 and ap ortion of the spectrum can be seen in Figure 5. The 2D NMR studies of the C20Mms6 peptides equence do not show any evidence for ad efined structural conformationi nt he absence of iron. We did not observe any significant chemical shift differencesf rom random coil values [22] (largestH a difference from expected random coil values is 0.092 ppm, mean Ha difference is 0.021(AE 0.028) ppm;c omplete detailsc an be found in Supporting Information Figure S4). This indicates that the peptidea dopts an unstructured conformation in solution. In addition we did not observe any non-sequentialN OE peaks due to through-space interactions between side chains.
The presenceo fi ron ions in NMR experiments severely suppresses peak intensities through paramagnetic relaxation enhancement (PRE). This mechanism is ar esult of interaction of  the unpaired electrons of iron with the NMR active nuclei. PRE can be au seful tool in structural biology for identifying residues of am etalloprotein which are involved in bindingp aramagnetic metals, [23] but is undesirable here because it leads to signals becoming broad and unobservable. Hence we used low peptidec oncentrationsw hich allow the addition of extremely smallq uantities of iron to keep PREs to am inimum. This strikesagood balance between maintenance of the signal level and longera cquisition times. Furthermore,a ddition of iron to peptides amples acts to lower the pH whichc an show as hift. These can mask the chemical shift changes due to metal binding. We therefore included in excessanon-chelating buffer,M ES, to strictly regulate the pH during the experiment and we checked the pH carefully before and after addition of metals and adjusted it where necessary,t oe nsure that any differences we observed were in fact due to metal bindingr ather than pH changes.
Previously reported studies have shown that Mms6 can bind Fe 3 + as well as other metals such as calcium and magnesium, but not copper,m anganese or zinc. [7] Based on this evidence we screenedapanel of metals consisting of Fe 2 + ,F e 3 + ,C a 2 + , and Zn 2 + .T his allowed us to comparem etals which C20Mms6 either should bind (Fe 3 + ,C a 2 + )o rs hould not bind (Zn 2 + )a nd to test the binding of the unknown Fe 2 + .C onsidering that magnetite is composed of an orderedm ixture of both ferrous and ferric ions, and from the insight we have obtained from the pH titrations, this experiment was designed particularly to probe Fe 2 + interaction and binding as there has been no previous analysiso fM ms6 binding to ferrous iron. Ferrousa nd ferric iron chloride were tested in separateN MR experiments by titrating iron in smalli ncrements into the NMR tube until the signal levels were too small to be detected, indicating the maximum quantity that could be added. To interpret the C20Mms6 chemical shifts upon addition of metal ions we analysed both the shift of the amide proton peaks, Figure 6a,a s well as the mean shift value for the side chain signals, Figure 6b.A sw ee xpected, the addition of zinc to the C20Mms6 peptideg ave rise to consistently small shifts along the length of the peptidec hain, indicating no specific or significant binding to any residues. Addition of either ferric or calcium ions, both predicted to bind, did result in an umber of significant chemical shift changes.T he most shifted residues are located at the acid-rich cluster D11-E13 with the largest average sidechain shift recorded for ferric binding at the sidechain of glutamic acid E13. However,t he changesw ere small. Furthermore, losses in signal intensity due to PREs from Fe 3 + were widely distributed along the peptides equence, furtheri mplying nonspecific binding of Fe 3 + .
The most significant shifts and intensity changes were observed upon the addition of ferrous iron. The magnitudes of the shifts in this case are up to fivefold greater than those observed for ferric addition. Residues D11, E12, E13, and E15 are all implicated in binding as well as (more weakly) E6 located furthera way on the polypeptide chain. This clearly indicates that C20Mms6 is able to specifically coordinate Fe 2 + ,t hrough the carboxylate side chains of the five key residues mentioned above,though presumably not all at the same time. Theamide groups of thep eptide main chain (Figure 6a)a lso show significant shifts, potentially indicating that the backbone amide of these residues may be involved in binding the metal ions directly or that the chemical environments of the amide backbones have changeda sar esult of bindingt hrough the side chain groups. The latter reason seems more likely as it implies some conformational changeo ft he peptidea si tc oordinates ferrous iron with several sidechains simultaneously.

Simulation studies
To complement the experimental study,a nd gain further insight at the molecular level, am odel peptide (DEEV) was subjected to molecular dynamics (MD) simulations using DL_POLY classic [24] to investigate ferrous iron binding. DEEV is the region of Mms6 that displays significant metal binding in our NMR experiments. Interactions were explored with 2nsM Ds imulations between DEEV and as ingle ferrous ion placed initially at one of 12 potential binding sites (Supporting Information Figure S5). The position of the ferrous ion during each simulation was trackedr elative to the oxygen atomso ft he peptide to determinet he strength of the interaction and the most likely binding sites. We found that 16 %o ft he sampled configurations resulted in ferrous binding, defineda st he metal ion being positioned within 3 of at least one oxygen atom of the peptide. Am ore detailed pictureo fb inding was obtained by considering which of the oxygen atoms within the peptide were most likely to be involved in ferrous binding. These results are summarised in Figure 7a.
We find that one of the most favoureds ites is the carboxylic acid sidechain of the second glutamic acid (E13). However,t he most significant binding from the simulations is with the main chain carbonyl, located between the two glutamate( E12-13) residues,w ith 56 %o ft he simulations that resultedi nf errous binding displaying interactions with this carbonyl. Similarly,t he NMR data report large chemical shifts of the amide protons for both of these flanking glutamate residues,w hichc ould be due to Fe II interaction with the peptide backbone of this region. However,i tc ould also be due to multiple bonds between the oxygen atoms in the peptide and the ferrous ion. Further analysis of the MD trajectoriesw hen the ferrous ion is bound to the backbone in the (E12-13) region reveals that in 90 %o f these occurrences it is also bound to one or other of the oxygen atomsi nE 13 carboxyl group, compared to just 6% of the configurations where it is bound to the oxygen in the backbone only.W ed on ot observe significant binding to the aspartate residue (D11) in our simulations. However, in our simulation,t he peptidea dopts am ore compacta rrangement upon ferrous binding, with hydrogen bonds present between the amide protons of the glutamate residues and the amide carbonyl of the N-terminal capping group (Figure 7b). This rearrangement is likely due to the short length of the peptide, but could indicate that, whilst not participating in ferrous iron binding directly,t hese residuesm ay undergo significant movement upon addition of metal ions, leadingt ot he chemical shift changes seen.
We also observed that when the ferrous ion is replaced with ferric ion in the simulations,t he additional positive charge makes binding to the peptide more favourable, buta lso means iron binding is seen indiscriminately at all the oxygen sites present. We believe it is likely that the additional selectivity shown by ferrous ions maye xplain why largera nd more specificc hemical shift changes are detected experimentally in the ferrous system.
The ferrous-oxygen radial distribution function (RDF;F igure 8a)e ffectively provides as ummary of the time spent by the ferrous ion at ap articulard istance from an oxygen atom within the peptide and gives us an indication of the stability and strength of binding (Figure 8b).
The RDF represents the combination of an umber of simulations, each searching ad ifferent part of configurational space. Therefore, it is reasonable to interpret the intensity of the RDF as the probability of aferrous ion being at aparticulardistance from an oxygen atom, making it ag ood approximation for the partitionf unctiono ft he system.T he RDF can be transformed into ap lot of free energy versusi ron-oxygens eparation using the statisticalt hermodynamic relationship: where g(r)i st he radial distribution function ( Figure 8a). The RDF will converge to av alue of unity at large separations, meaningt he calculated free energy will be zero when the separation is large and that the calculated values represent the free energy change in bringingt he ferrous ion towardst he peptidetoaseparation, r (Figure 8b).  Inspection of Figure 8a shows that the binding to the peptide occurso ver an arrow range of separations between 2.55 and 3.25 (optimum bond distance 2.85 ). The intensity of this peak, and the narrow range of bond distances it covers, suggestst hat when iron is bound it is held relatively strongly. The fact that the RDF drops close to zero between the first and second peak illustrates that once af errous ion binds to aparticular site, it will remain bound to the peptidefor asignificant time in most cases.
These observations are clearer when the data are plotteda s an energy profile ( Figure 8b). Ad eep energy well is evident close to the surfacew hich corresponds to ab inding energy of À6.7 kJ mol À1 (À2.7 RT). This is sufficiently large that it could not be overcomeb yt he kinetic energy possessed by the atoms. Therefore, once af errous iron has enteredt he well it is unlikely to leave during the simulation. To leave the adsorption layer,t he ion must also cross the energy barrier at 3.5 separation, giving at otal barrier to desorption of 10.2 kJ mol À1 (4.0 RT).
Whilst indicative of the nature of binding, it must be noted that Figure 8i sa na verage of the behaviour of af errous ion in multiple simulations and does not give any indication of how the energy profile differs at sites where iron binding is known to be favoured and where it is not observed.
Such insight can be gained by plotting the RDF (and associated free-energyp rofile) for the individual interactions between ap articular oxygen and the ferrous ion (SupportingI n-formationF igures S7 and S8). At residueE 13 and the bridging carbonyl oxygen between E12 and E13 sites, the binding energy is significantly higher,b etween À9a nd À11 kJ mol À1 (or 3.5 to 4.5 RT), with the additional energy required for the ferrous ion to leave the adsorption layer approximately 4kJmol À1 ( % 14 kJ mol À1 ). This suggests that once an ion enters these binding sites it will not leave for the duration of the MD simulation and that binding shouldb eo bservable on an experimental timescale. The sites in D11a nd E12, where limited iron binding is seen in the MD simulations, display ab inding energy of approximately À4kJmol À1 . This is close to the kinetic energyp ossessed by the atoms in ambient conditions. An additional energy barrier must still be crossed for the ion to completely leave the binding site but is still of an order that such desorptioni sp ossible and hence why only limited binding is observed in the MD simulations. When extended to larger time scales, this would represent as ystem where continuous exchange between binding and desorption would be possible, especially when concentration effects, not present in our MD simulations, are taken into account.
Similar calculations were performed using Fe 3 + instead of Fe 2 + .T he additional chargeo nt he iron made the overall binding energy to this negatively charged peptidem ore favourable. However,i ti ss ignificant that binding was less specific to E12/E13.

Discussion and Conclusions
Analysing the effect of Mms6 in situ through pH measurements throughout the magnetite-formation process has en-abled us to obtain an insight into the protein's functiona nd in particular, at what stage in the reactioni te xerts its effect. We see no influence of the protein in the early stages of the reaction at lowp Ha tt he point where ferric oxyhydroxidem inerals are formed,c onsistent with the theoretical isoelectric point of 4.2 of the acidic iron binding region of the protein.I ti sf rom pH 5a nd upwards that we see Mms6a ffecting the formation of the minerals, at the point that ferrous ions begin to precipitate out of solution and mixed valencei ron minerals are formed. From our data it appears that Mms6p romotes the formation of magnetite under ferrous-rich conditions, even in conditions in which magnetite does not readily form. All previous studies on the iron binding activity of Mms6 have focused on ferric-ion binding of the acidic C-terminal region [7,8] and often at low pH. From our study it seems Mms6 is more likely to interact predominantly with ferrous ions at pH > 5. In this context we performed an NMR metal-binding experiment under these conditions and we saw that Mms6 does indeed show specific ands ignificant interactions with ferrous ions compared to both positive and negative control metals, including ferric ions. The residues responsible for binding metals vary for ferrous and ferric ions, but the DEEVE clusterw as found to have the biggest shift for both iron ions, with the double EE site having the prominents hift for ferric ions and all the acidic residues of the whole DEEVEr egion significantly shifted (up to 5 > Fe 3 + )i nt he presence of ferrous ions.
The interaction of this smaller peptideo fi nterestw ith ferrous and ferric ions was thus modelled by atomistic simulation in order to furtheru nderstand the binding event.T his indicated that the acidic residues may bind the ferrous iron not only through their carboxylate sidechainb ut also through the peptide carbonyls. The MD calculations also show that multidentate binding is preferred, especially for ferrous ions, and that bidentate binding of Fe 2 + to E12 carbonyl and E13 carboxylate is particularly frequent.F erric ions show binding, but unspecifically across the acidic amino acids.
As the pH is raised,a nd particularly at high ratios of ferrous to ferric ions, the initially formed ferric oxyhydroxidep recipitate must re-dissolve and reform into ad ifferent crystalline structure in order to incorporate ferrous ions andf ormm agnetite. Our datas uggestt hat Mms6 is facilitating this process. We therefore propose an in vitro mechanism in which Mms6 displays little activity below pH 4-5. Above this, Mms6 specifically sequesters andb inds ferrous ions. We also propose that the Mms6's known affinity for ferric ions promotes the dissolution of the unstable ferric precursors to sequester ferric ions as well. We have shown heret hat the role of Mms6i st op romote the formation of magnetite at high ferrous ratios, which are conditions where magnetite is formedp oorly if at all in the absence of Mms6. We therefore propose that Mms6 not only helps to dissolveu ndesired precipitates,s uch as ferric oxyhydroxide, but helps the formation of magnetite by acting as an ucleation protein.
Mms6 is knownt os elf-aggregate [10] and it is already proposed that Mms6a ssembles to form an acidic, charged, C-terminal surface to bind multiple iron ions. [8,25] The self-assembly of Mms6 could space these acidic residues to bind both fer-rous and ferric ions in am agnetite crystallographic geometry to template and promote the initial formation of magnetite over other iron mineralsbylowering the free energy for formation of magnetite.
Relating the reactions occurring in solution in this experiment to those found in vivo is challenging. Mms6 is associated with the magnetosome membrane, so the C-terminal acidic surfaceo fa nM ms6 assembly will differ in its in vitro micelle structurec ompared to the arrangementi nt he internal lumen of the magnetosome.I nthe magnetosome there are transporters that transport iron ions into the magnetosome. [26] All current knowledge points to the transport of ferrous ions only, with partial oxidation of some of the ferrous ions to generate the ratio required for magnetite formation. [26] Furthermore, ferrous transporters are antiporters meaning they bring ferrous iron in as they take protons out. This will help to regulate the internal pH, but the precise pH of this environment is also not currently known. Additionally,d ue to the iron transport system it is highly unlikely that there will be SO 4 2À counterions, ruling out the formation of schwertmannite and greenrust phases.
However,s imilaritiesb etween the two systemsc an be drawn to predict the action of Mms6i nt he magnetosome. While the pH of the interior of the magnetosomeh as not yet been determined, it must at some point be high enough to enable magnetite to crystallise (! pH 7), and we have established in this study that Mms6 is active in these conditions.I n the natural system we assume some of the ferrous ions are converted to ferric (through the action of oxidase enzymes), and thus aferric mineral will precipitate (due to its chemical insolubilitya tp H! 2.5). There have been severalr eports of such ferric precursors, [27] and while our suggested mechanism relies on dissolution of this precursor,i ts exact form is immaterial, be it schwertmannite (in vitro) or ferrihydrite (in vivo). Interestingly,M ms6 activity is greatest in ferrous-rich conditions, which is likely to be the predominant state in magnetosomes if it is confirmed that iron transport is of ferrous iron.I ti st herefore likely that in vivoc onditions in the magnetosome are similar to our in vitro conditions with respectt op ossible pH, ferric precursors in af errous-ion-rich solution,a nd the protein's ability to form as elf-assembled chargeds urfacec apable of sequestering ferrous and ferric iron in the exact proportions to promote the crystallization of magnetite.
Therefore we propose, both in vitro and in vivo, that Mms6 is am agnetite nucleation protein, able to bind both ferric and specifically ferrous iron to initiate the process of templating the magnetite crystal.

Experimental Section
Ferrous and ferric sulfate were purchased from Sigma-Aldrich and used without further purification. ICP-AES was used to check the hydration level of both compounds prior to each set of experiments. Peptide C20Mms6 KSRDIESAQSDEEVELRDAL, featuring an acetylated Nterminus and an amidated Cterminus, was purchased from Genscript (USA). The peptide was dissolved in and dialysed, over-night, against ac hosen buffer in ad ialysis cassette with aM WCO of 1kDa, to remove trace salts. MES and NaH 3 PO 4 were purchased from Sigma-Aldrich and used without further purification.

Biochemical methods
The mms6 sequence from Magnetospirillum magneticum AMB-1w as introduced into ap TTQ18-based expression vector by cohesive-end cloning with the resulting plasmid, pHis8mms6, encoding N-terminally octahistidine tagged Mms6. The protein was produced in E. coli BL21 star (DE3) cells (Invitrogen) harbouring ap RARE (Merck) plasmid to compensate for codon bias in the mms6 sequence. Cells were cultured in autoinducing Superbroth (Formedium) at 37 8Cw ith shaking for 24 hi nt he presence of carbenicillin and chloramphenicol to select for the pHis8mms6 and pRARE plasmids, respectively.C ells were lysed by sonication in 25 mm Tris pH 7.4, 100 mm NaCl. The insoluble material, containing the His8-Mms6 inclusion bodies, was collected by centrifugation at 16 000 g and resuspended in 6 m guanidine hydrochloride, 25 mm Tris pH 7.4 to solubilise the proteins. Further centrifugation at 16 000 g was performed to remove any material not solubilised by the guanidine treatment. The supernatant was mixed with nickel charged nitrilotriacetic acid (NTA) resin (Amintra resin, Expedeon) to allow binding of the histidine tagged Mms6. The resin was subsequently packed into ag ravity flow column and washed extensively with wash buffer (6 m guanidine hydrochloride, 25 mm Tris pH 7.4, 10 mm imidazole) before elution of the bound protein in 300 mm imidazole-supplemented wash buffer.T he eluted protein was refolded by rapidly diluting into al arge volume of refolding buffer (500 mm NaCl, 25 mm tris pH 7.4) before being concentrated using a1 0kDa molecular weight cut-off centrifugal concentrator (Sartorius). The concentrated material was subjected to centrifugation to remove any small amounts of precipitated protein before dialysis against 500 mm NaCl using a3 .5 kDa molecular weight cut-off slide-a-lyser (Thermo Scientific). The refolded His8-Mms6 was quantified by absorbance at 280 nm and stored at 193 K.

NMR spectroscopy experiments and analysis
For NMR studies all samples were prepared in a9 0% H 2 O/10 % D 2 Os olvent mixture. NMR spectroscopy was performed on aB ruker AvanceD RX instrument at 600 MHz, equipped with at riple resonance ( 1 H-13 C/ 15 N-2 H) cryoprobe with z-gradient, at 25 8C. TOCSY experiments were carried out with as pin lock power of 8.3 kHz for 45 ms, and ROESY experiments with as pin-lock power of 2.27 kHz with a2 00 ms mixing time. Solvent suppression was achieved with a3 -9-19 pulse sequence with presaturation during the relaxation delay of 1.5 sf or the TOCSYexperiment, and with excitation sculpting with gradients for the ROESY experiment.
For sequential assignment, 1mm C20Mms6 was prepared in 20 mm NaH 3 PO 4 buffer at pH 7. 3-(Trimethylsilyl)-2,2',3,3'-[D 4 ]-propionate (TSP) was added as as tandard (0.0 ppm). For metal binding studies, 50 mm C20Mms6 was dialysed extensively against MES buffer (50 mm)a tp H5.66 AE 0.02. Metals were dissolved in the same buffer,a nd added directly to the NMR tube stepwise until the metal concentration was 150 mm.A te ach addition, the pH was checked and adjusted to 5.66 AE 0.02 if necessary.

Co-precipitation of magnetite
Ultrapure (MilliQ) water was degassed under vacuum and sparged with nitrogen for at least 1h to remove dissolved oxygen. An aliquot (20 mL) of this water was added to at hree-necked glass flask. The solution is isolated from the air by continuous sparging with nitrogen. This is also used to mix the solution. Fe salts were measured to 0.1 mg accuracy and stored in asmall plastic tube. Initially Fe 3 + was added to the solution by removing 1mLo ft he solution and dissolving the Fe salt in the tube. This was then added to the solution with ag lass pipette. Fe 2 + was then added in the same way.E ach set of experiments varies the X parameter from 0.2 to 0.5. One batch of nitrogen sparged 1 m NaOH solution was used per set of experiments to ensure consistency. AM ettler To ledo 7M ulti-pH meter with Micro-Pro probe was immersed in the solution. Data were logged on ac omputer via serial port communication at three second intervals for the duration of the experiment. The NaOH injection needle was pre-primed before the start of the experiment. This solution was added continuously to the reaction flask at ar ate of 50 mLmin À1 via as yringe pump driver.

Simulation studies
The molecular dynamics simulations were based on am olecular mechanics approach where the intra-and intermolecular interactions within the peptide were described using the generalised Amber force field (GAFF) [28] which also incorporates av ersion of the TIP3P water molecule. [29] Partial charges on the peptide were determined using semi-empirical calculations at the AM1 level of theory. [30] At otal of 24 production calculations were performed (2 different ions with 12 starting configurations) where the peptide was placed at the centre of a4 040 40 box which contained approximately 2000 water molecules. In all cases the 3D boundary conditions were applied within the isobaric isothermal NPT ensemble which allows the sizes of the simulation cell to change during simulation. The temperature and pressure were maintained at 300 Ka nd 10 5 Pa using aN ose-Hoover thermostat and barostat with ap eriod of 0.1 and 0.5 ps. [31] The trajectories were generated using the Verlet leapfrog algorithm with at ime step of 1f s. [32]