McaA and McaB control the dynamic positioning of a bacterial magnetic 1 organelle 2

15 Magnetotactic bacteria (MTB) are a diverse group of microorganisms that use intracellular chains 16 of ferrimagnetic nanocrystals, produced within their magnetosome organelles, to align and 17 navigate along the geomagnetic field. The cell biological and biochemical properties of 18 magnetosomes make them a powerful model for studying the molecular mechanisms of 19 biomineralization and compartmentalization in bacteria. While several conserved magnetosome 20 formation genes have been described, the evolutionary strategies for their species-specific 21 diversification remain unknown. Here, we demonstrate that the fragmented nature of magnetosome 22 chains in Magnetospirillum magneticum AMB-1 is controlled by two genes named mcaA and 23 mcaB . McaA recognizes the positive curvature of the inner cell membrane while McaB localises 24 to magnetosomes. Along with the MamK actin-like cytoskeleton, they create space for addition of 25 new magnetosomes in between pre-existing magnetosomes. Phylogenetic analyses suggest that 26 McaAB homologs are widespread and may represent an ancient strategy for organelle positioning 27 in MTB. 28


Introduction 29
Cellular compartmentalization results in the formation of different organelles, which need to be 30 positioned correctly to fulfil their specific functions and ensure proper inheritance throughout cell 31 division 1 . Organelle positioning in eukaryotic cells mainly relies on cytoskeletal and motor 32 proteins 1 . Many bacteria also produce organelles 2 , and actively regulate their placement in the 33 cell. For example, the protein-bounded carbon-fixation organelle, the carboxysome, uses the 34 nucleoid as a scaffold with helper proteins that ensure equal distribution in the cell and proper 35 segregation into daughter cells 3 . Similarly, it has been proposed that the carbon storage 36 polyhydroxybutyrate (PHB) granules associate with nucleoids to mediate segregation during cell 37 division 4, 5 . A widely studied example of bacterial lipid-bounded organelles is the magnetosome 38 compartment of magnetotactic bacteria (MTB). Magnetosomes mineralize ferrimagnetic 39 nanoparticles composed of magnetite (Fe3O4) and/or greigite (Fe3S4) 2, 6 , which are used as a 40 compass needle for navigation along the geomagnetic field. Magnetic navigation is a common 41 behaviour in diverse organisms, including bacteria, insects, fish, birds, and mammals 7, 8 . MTB are 42 the simplest and most ancient organism capable of magnetic navigation 9 and fossilized 43 magnetosomes chains have been used as robust biosignatures 10, 11 . Thus, magnetosome production 44 in MTB is an ideal model system for studying mechanisms of organelle positioning, understanding 45 the evolution of magnetic navigation, and connecting the magnetofossil record to the history of 46 life on Earth. 47 To function as an efficient compass needle, individual magnetosomes need to be arranged into a 48 chain. Various and complex magnetosome chains (single-or multi-stranded, continuous or 49 fragmented) are found in diverse MTB groups 12,13,14 . The mechanisms leading to distinct chain 50 configurations remain unknown, but may reflect strategies for adaptations to specific biotopes 10 . 51 The most widely-studied model MTB strains Magnetospirillum magneticum AMB-1 (AMB-1) 52 and Magnetospirillum gryphiswaldense MSR-1 (MSR-1) are closely related Alphaproteobacteria 53 species sharing 96% identity in their 16S rRNA gene sequences 15 . However, their magnetosome 54 chain organisation strategies are distinct. In AMB-1, magnetosomes containing magnetic crystals 55 and empty magnetosomes are interspersed to form a chain that is fragmented in appearance, 56 extends from pole-to-pole in the cell, and remains stationary during the entire cell cycle 16 . In 57 contrast, in MSR-1, magnetic crystals are arranged as a continuous chain at the midcell and the 58 divided daughter chains rapidly move from the new poles to the centre of the daughter cells after 59 in other MTB. We hypothesize that the MIS is a remnant of an ancient duplication event which 91 paved the way for an alternative chain segregation strategy in AMB-1. This mode of chain 92 segregation may lower the energy requirements for separating magnetic particles at the division 93 septum and eliminating the need for rapidly centring the chain after cell division. 94

MIS genes control the location of and spacing between magnetosomes 96
To investigate its possible role in magnetosome formation and placement, we deleted the entirety 97 of the MIS from the AMB-1 genome. The deletion's effect on magnetosome production was 98 assessed by measuring the coefficient of magnetism (Cmag) using a differential 99 spectrophotometric assay that quantifies the ability of MTB to orientate in an external magnetic 100 field 27 . Unexpectedly, the Cmag values of ΔMIS cultures are much higher than Wild-type (WT) 101 cultures (Fig. 1a), indicating that, as a population, ΔMIS cells better align with the applied external 102 magnetic field. As expected, transmission electron microscopy (TEM) images of WT AMB-1 103 cultures show that the magnetic crystals are organised into a chain with gaps from cell pole to pole 104 (Fig. 1b). In contrast, the crystals in the ΔMIS strain are organised into a continuous chain at the 105 midcell ( Fig. 1b). Analysis of TEM images shows that the number and length of crystals are similar 106 ( Supplementary Fig. 1a, b), but the shape factor of crystals (width/length ratio) differs between 107 WT (0.82) and ΔMIS (0.92) strains ( Supplementary Fig. 1c, d). These data were collected from 108 strains grown under microaerobic conditions. To ensure that the observed phenotypes were not 109 generated by specific growth conditions, the experiment was repeated under anaerobic conditions 110 and yielded similar results ( Supplementary Fig. 2). 111 Magnetosome biogenesis in WT AMB-1 begins with the invagination of bacterial inner membrane 112 to form empty magnetosomes (EMs), followed by the crystallisation of ferrimagnetic minerals to 113 form crystal-containing magnetosomes (CMs) 2, 6 . To directly observe the organisation of 114 magnetosome membranes, we imaged WT and ΔMIS cells with whole-cell cryo-electron 115 tomography (cryo-ET) (Fig. 1c, d). Similar to WT 28 , the magnetosome membranes of ΔMIS 116 mutant are invaginations of the inner membrane (Fig. 1d, lower left corner). We then measured 117 the diameter of magnetosome membranes and the length of crystals. The size distribution of EMs 118 and CMs, as well as the linear relationship between the sizes of crystals and magnetosome 119 membrane diameters are similar between the WT and ΔMIS mutant ( Supplementary Fig. 3). 120 Additionally, in both strains, the EMs are significantly smaller than the CMs ( Supplementary Fig.  121 3a). One major phenotypic difference between the two strains is that in WT EMs are present at 122 multiple sites between CMs in the magnetosome chain (Fig. 1c) whereas EMs only localise at both 123 ends of the continuous chain of in the ΔMIS strain (Fig. 1d). Analysing the size and 124 biomineralization status of magnetosome membranes relative to their subcellular position shows 125 that the location of EMs is random along the magnetosome chain in WT but is at both ends of the 126 chain in ΔMIS cells ( Fig.1 e, f). These results together suggest that MIS genes control the location 127 of magnetosomes but not the size of magnetosome membranes. 128 Based on the different sizes and locations of magnetosomes, we hypothesised that the newly-made 129 magnetosomes are added at multiple internal sites of the chain in WT, but only added at the ends 130 of the chain in ∆MIS. To test this hypothesis, we designed a pulse-chase experiment to label and 131 follow a marker protein that incorporates into magnetosomes at the early steps of membrane 132 invagination (Fig. 2a). We examined the magnetosome marker proteins MamI and MmsF 29,30,31 . 133 Newly-synthesized MmsF proteins incorporate into both the new and old magnetosomes (see more 134 details in supplementary results and supplementary Fig. 4), indicating that it is not suitable for the 135 pulse-chase experiment. 136 The transmembrane protein MamI is needed for EM invagination from the inner membrane of 137 AMB-1 26, 31 . We first checked the localization of MamI-GFP in WT and ∆MIS. Structured 138 illumination fluorescent microscopy (SIM) imaging shows MamI-GFP localises as a continuous 139 line from cell pole to pole in WT AMB-1 (Fig. 2b), indicating localization to both the EMs and 140 CMs. In ∆MIS, MamI-GFP only localises in the middle of the cells in a pattern reminiscent of 141 magnetosome organisation as seen in cryo-ET images (Fig. 2b). We then performed pulse-chase 142 experiments using MamI-Halo. The Halo-ligand JF549 was used as the pulse to mark old 143 magnetosomes and the JF646 ligand was chased in to identify the newly-made magnetosomes (  Table 1). As expected, JF549-marked old 148 magnetosomes display gaps, which are filled with the JF646-marked newly-made magnetosomes 149 in WT AMB-1 ( Fig. 2d and Supplementary Fig. 5a). Conversely, the JF549-marked old 150 magnetosomes still mainly show a continuous chain at the midcell of ΔMIS, and the JF646-marked 151 newly-made magnetosomes localise at both ends of the chain ( Fig. 2d and Supplementary Fig. 5a). 152 Together, these results confirm our hypothesis that the varying chain phenotypes between WT and 153 ΔMIS strains are in part due to changes in the location where new magnetosomes are added. 154 Additionally, we found that the length ratio between the MamI-GFP marked magnetosome chain 155 and the cell body is significantly larger in WT than in ΔMIS (Fig. 2c). As mentioned above, the 156 number of crystals in WT and ΔMIS cells is similar, indicating the distance between the 157 magnetosomes might be different in these two strains. We therefore measured the edge-to-edge 158 magnetosomes distance and found that the distance between all magnetosomes in WT is about 159 twice as long as in the ΔMIS strain (Fig. 1g), while the distance between EMs in these two strains 160 is similar (Fig. 1h), indicating the difference is mainly due to the distance between CMs. 161 To summarise, MIS genes control the shape of crystals, the distance between CMs, and the location 162 for the addition of newly-made magnetosomes leading to the characteristic pattern of chain 163 organisation in AMB-1. 164

Comprehensive dissection of the MIS new chain organization factors 165
To identify the key genes that control magnetosome positioning, we conducted conventional 166 recombination mutagenesis to create unmarked deletions of selected segments in the MIS (Fig.  167 3a). We first deleted large domains (LD1 and LD2) to narrow down the region of interest in LD1 168 (Fig. 3b,c and Supplementary results). We then generated small islet region deletions (ΔiR1, ΔiR2, 169 ΔiR3, and ΔiR4) of LD1 to pinpoint specific genes involved in chain organisation. Genes in the 170 iR2 region control magnetosome positioning, while those in iR3 contribute to crystal shape control 171 (Fig. 3b,d and Supplementary results). iR2 contains a small putative operon with two hypothetical 172 genes (amb_RS23835 and amb_RS24750), which we have named magnetosome chain assembly 173 genes A and B (mcaA and mcaB) (Fig. 3a). The reference genome in NCBI shows the iR2 region 174 includes a third transposase gene (amb_RS23840) (Fig. 3a), which does not exist in our lab strain. 175 We then deleted these two genes individually. Both ΔmcaA and ΔmcaB strains have dramatically 176 higher Cmag compared to WT and are similar to ΔMIS (Fig. 3b). Similar to the ΔMIS mutant, 177 ΔmcaA and ΔmcaB strains contain continuous crystal chains in the midcell region when viewed 178 by TEM (Fig. 3d), indicating that both play essential roles in magnetosome positioning. 179

McaA localises to the positively-curved cytoplasmic membrane as a dashed line 180
We interrogated the localization of McaA and McaB in order to understand their role in controlling 181 magnetosome positioning. McaA is predicted to contain a signal peptide, followed by a 182 periplasmic von Willebrand factor type A (VWA) domain, a transmembrane (TM) domain, and a 183 cytoplasmic C-terminus ( Fig. 4a and Supplementary results). However, in some bioinformatic 184 predictions, the signal peptide region is predicted to be a TM domain with the N-terminus facing 185 the cytoplasm (Supplementary Table 2

McaB associates with crystal-containing magnetosomes 210
McaB is predicted to contain one TM domain that is close to the N-terminus, which is mostly 211 facing the periplasm ( Supplementary Fig. 9a). We confirmed the cytoplasmic location of C- organisation under low iron conditions. We then examined the dynamics of magnetosome chain 244 organisation as WT and cells missing mcaAB transitioned from low to high iron conditions to 245 trigger magnetite production in EMs (Fig. 5b). We performed pulse-chase experiments using 246 MamI-Halo with cells growing from iron starvation (pulse with JF549) to standard iron growth 247 conditions (chase with JF646). As expected, the pulse experiments show a continuous chain of 248 EMs in the middle of all WT and mcaAB deficient cells under iron starvation conditions (Fig. 5c  249 and Supplementary Fig. 5b). After iron addition to the growth medium, we observed the formation 250 of gaps between JF549-marked old magnetosomes in WT, but not in ∆MIS or ∆iR2 cells (Fig. 5c  251 and Supplementary Fig. 5b). In addition, the JF646-marked newly-made EMs filled the gaps 252 between older magnetosomes in WT but were only added at both ends of the chain in ∆MIS and 253 ∆iR2 ( Fig. 5c and Supplementary Fig. 5b). Accordingly, quantitative analysis shows a low 254 colocalization coefficient of the pulse and chase signals in WT, ∆MIS, and ∆iR2 cells 255 (Supplementary Table 1). Together, these results support the hypothesis that McaA serves as a 256 landmark on the positively-curved inner membrane and coordinates with McaB to control the 257 location and spacing between CMs, allowing the addition of newly-made EMs to multiple sites 258 between pre-existing magnetosomes in the chain of WT AMB-1, which forms the fragmented 259 crystal chain. 260 McaA contributes to the differences of mamJ and mamY deletions between AMB-1 and 261

MSR-1 262
As mentioned above, the phetotypes of mamJ and mamY deletion mutants in AMB-1 and MSR-1 263 are distinct. MamJ is proposed as a linker to attach MamK filaments to magnetosomes and its 264 deletion in MSR-1 causes the magnetosome chain to collapse and form an aggregate 20 . In contrast, 265 the deletion of mamJ and its homolog limJ in AMB-1 still shows a magnetosome chain with some 266 minor structural defects 21 . However, deletion of the entire MIS in a DmamJDlimJ strain causes a 267 dramatic chain collapse phentype resembling those of MSR-1DmamJ mutant 10, 20 . MIS contains a 268 second mamJ homolog called mamJ-like and our deletion analysis shows that it does not contribute 269 to chain maintenance (Fig. 3a, Fig. 6a MamY is a membrane protein that directs magnetosomes to the positively-curved inner membrane 275 in MSR-1, thus aligning the magnetosome chain to the motility axis within a helical cell 22 . When 276 mamY is deleted in MSR-1, the magnetosome chain is no longer restricted to the positively-curved 277 regions of the membrane and can also be found at the negatively-curved membrane leading to a 278 much lower Cmag compared to WT 22 . Surprisingly, when mamY is deleted in AMB-1, the Cmag 279 is similar to WT (Fig. 6b), and the magnetosome chain still localises to the positively-curved 280 membrane ( (Table 1 and Supplementary Fig. 11f, g). Together, 330 these results indicate that McaA, -B influence the dynamics of MamK filaments which, in turn, 331 leads to the AMB-1-specific pattern of magnetosome chain organization. 332 mcaA and mcaB genes are specific to MTB 333 To understand the evolutionary origins of the mcaAB system, we searched for homologs of these 334 two genes in diverse species of MTB. Distant homologs were found in 38 MTB species. All of 335 them belong to MTB strains either with characterized magnetosome chain phenotypes (Fig. 8a, b) 336 or metagenomes obtained from a magnetic enrichment (supplementary Fig. 16 a, Fig. 16 a, b) suggest a recent acceleration of the evolution of 365 Mca proteins which could be linked with their neofunctionalization in AMB-1. 366

Discussion 367
In this study, we uncovered the mechanisms of a novel magnetosome chain organisation strategy 368 that explains phenotypic differences between closely related MTB. We demonstrated that the 369 fragmented crystal chain organisation in WT AMB-1 strain is not growth condition dependent but shows that MamK cytoskeleton is composed of short filaments and located along the magnetosome 382 chain in both WT and ΔMIS cells (Fig. 1c, d). Based on FRAP experiments, MamK filaments in 383 WT display local recovery that can be caused by monomer turnover 384 (depolymerization/polymerization), filament sliding, or formation of new filaments. In many 385 McaAB deficient cells, MamK filaments recover and at the same time move across the cell, which 386 might help position the magnetosomes in the midcell (Fig. 8c). We propose that the McaAB system 387 localises the turnover of MamK filaments to allow for new magnetosome addition in between pre-388 existing magnetosomes in WT AMB-1 (Fig. 8c). It is notable that the action of two proteins is sufficient to fundamentally alter the assembly and 431 organisation of magnetosome chains in AMB-1 as compared to MSR-1, one of its closest relatives. 432 We propose that the alternative mode of chain organisation in AMB-1 may provide advantages 433 that have led to its selective maintenance. Since magnetic particles are arranged as sub-chains 434 along the length of AMB-1, daughter cells are ensured to inherit equal numbers of magnetic 435 particles that are centrally positioned. Additionally, the distribution and spacing of CMs and EMs 436 may reduce the forces needed to separate magnetic particles. In contrast, MTB such as MSR-1, 437 need to break the closely located continuous crystal chain in the middle and dynamically reposition 438 the entire chain after cell division, which could be more energy-demanding than the stationary 439 ones in WT AMB-1. However, as seen in our Cmag data, AMB-1 cells, as a population, align 440 better in magnetic fields in the absence of mcaAB. Given the specific biological interventions 441 required for their assembly, preserved magnetite or greigite chains are also considered an 442 important criterion for magnetofossil recognition and characterization 10, 11 . Thus, understanding 443 the selective pressures that dictate the species-specific mechanisms of chain organisation in 444 modern day organisms can provide much needed insights into the conditional functions of 445 magnetosomes across evolutionary time. 446 Methods 447

Bacterial growth 448
The 1.5-mL stock cultures of AMB-1 strains were prepared as described previously 40  The details about generation of plasmids and strains are described in the supplementary methods. 471 The strains, plasmids, and primers used in this study are described in Supplementary Tables 5 to  472 12. 473

Deletion mutagenesis 474
A two-step homologous recombination method was used to generate deletion mutants in AMB-1 475 strains as previously described 41 . Briefly, an approximately 800 to 1000 bp region upstream and 476 downstream of the deleted gene or genomic region were PCR amplified from the AMB-1 genomic 477 DNA using primer pairs (A, B) and (C, D), respectively (Supplementary Table 8). The two PCR 478 fragments were cloned into the SpeI restriction site of the pAK31 suicide plasmid using Gibson 479 assembly to generate the deletion plasmids (Supplementary Table 6). The deletion plasmid was 480 conjugated into AMB-1 strain using E.coli WM3064 donor strain. Colonies that had successfully 481 integrated the plasmid were selected on MG agar plates containing 15 μg/mL kanamycin. To 482 select for colonies that had undergone a second recombination event to lose the integrated plasmid, 483 a counter-selectable marker sacB, which is toxic in the presence of sucrose, was used. Colonies 484 were then passed in 10 mL of growth media without kanamycin and plated on MG agar plates 485 containing 2% sucrose. The resulting sucrose-resistant colonies were checked for the successful 486 deletions at their native locus by colony PCR with primers listed in Supplementary Table 9. 487

Cellular magnetic response 488
The optical density at 400 nm (OD400) of AMB-1 cultures in the green-caped tubes was measured 489 at 24 hours and 48 hours using a spectrophotometer. A large magnet bar was placed parallel or 490 perpendicular to the sample holder outside the spectrophotometer, the maximum and minimum 491 OD400 were recorded. The ratio of the maximum to the minimum was designated as AMB-1 cells' 492

Transmission electron microscopy (TEM) 494
For imaging the whole AMB-1 cells by TEM, 1-mL AMB-1 cells were taken from the 10-mL 495 cultures that grew under different conditions. The 1-mL cells were pelleted and resuspended into 496 5-10 µL of MG medium. The resuspended cells were applied on a 400-mesh copper grid coated 497 with Formvar and carbon films (Electron Microscopy Sciences). The grids were glow-discharged 498 just before use. Then the air-dried cells were imaged on an FEI Tecnai 12 transmission electron 499 microscope equipped with a 2k x 2k charge-coupled device (CCD) camera (The Model 994 500 UltraScan®1000XP) at an accelerating voltage of 120 kV. Crystal size quantification and 501 statistical tests were performed as described previously 42 . 502

Cryo-electron tomography (cryo-ET) 503
Cryo-ET sample preparation and data collection were performed as described previously 42 . The 504 two-dimensional (2D) images of WT and ΔMIS cells were recorded using JEOL JEM-3100 FFC 505 FEG TEM (JEOL Ltd.) equipped with a field emission gun electron source operating at 300 kV, 506 an Omega energy filter (JEOL), and a K2 Summit counting electron detector camera (Gatan). 507 Single-axis tilt series were collected using SerialEM software 43 from −60° to +60° with 1.5° 508 increments, at a final magnification of 6,000x corresponding to a pixel size of 0.56 nm at the 509 specimen, and a defocus set to −15 μm under low dose conditions (a cumulative electron dose of 510 ~120 e/A 2 ). Tomogram reconstructions were visualized using the IMOD software package 44 . 511 Amira was used for the 3D model segmentation (Thermo Fisher Scientific). 512

Size and location analysis of magnetosome membranes 513
The reconstructed tomograms were visualized using a 3dmod software package 44 . To evaluate the 514 relative size and location of magnetosome membranes, the diameter of each magnetosome 515 membrane was measured as described previously 40 . The size of magnetosome membranes in each 516 cell was sorted from largest to smallest and then was nominated from 1 (largest) to 0 (smallest) 517 accordingly. The first magnetosome membrane on the left of the chain was numbered as 0, the 518 middle one was numbered as 1, and the last magnetosome membrane on the right of the chain was 519 numbered as 2. The location of other magnetosome membranes was nominated accordingly. 520 For measuring the distance between individual magnetosome membranes, the shortest distances 521 between neighbouring magnetosome membranes were found by working through the tomogram 522 slices and manually measured by 3dmod. 523

Structured illumination fluorescent microscopy (SIM) 524
To stain the genomic DNA, AMB-1 cells growing in 10-mL MG medium of the green-capped 525 tubes were collected by 16,800 x g for 3 min. The cell pellets were resuspended in a 1-mL fresh 526 MG medium and stained with 1.4 µM 4′,6-Diamidino-2-Phenylindole (DAPI) in dark at room 527 temperature for 15 min, and then washed 3 times with fresh MG medium. After washing, the pellet 528 cells were resuspended with 30-50 µL of MG medium and were immediately imaged by Carl Zeiss 529 Elyra PS.1 structured illumination microscopy with objective lens Plan-APOCHROMAT 530 100 ×/1.46. DAPI, GFP, JF549, and JF646 were excited by 405 nm, 488 nm, 561 nm, and 642 nm 531 lasers, respectively, and fluorescence from each fluorophore was acquired through 420-480 nm, 532 495-550 nm, 570-620 nm, and LP655 nm bandpass filters, respectively. Raw images were acquired 533 and processed using ZEN software (Zeiss). The processed images were then visualized using 534 Imaris (Bitplane). 535 MamI-GFP localization patterns in WT and ΔMIS cells were manually measured using the Fiji 536 software package 45 . The magnetosome chain showed a linear line across the cellular axis, so the 537 end-to-end distance of the GFP fluorescence line was considered as the length of the magnetosome 538 chain. A line that parallels the magnetosome chain was drawn from cell pole to pole, and this line 539 was considered as the length of the whole AMB-1 cell. 540

Pulse-Chase analysis 541
To study the addition of newly-made empty magnetosomes into magnetosome chains over time, images were captured through the LSM880 Zen software (Zeiss) and analysed using Fiji 45 . 580

Time-lapse imaging using HILO microscopy 581
For sample preparation, round coverslips (Matsunami, 25-mm diameter, 0.12-0.17 mm thick) 582 were used as the imaging support. The coverslip was coated with poly-L-lysine and 500 µL of 583 culture was added to an Attofluor cell chamber (Thermo Fisher Scientific). Then, a 5-mm thick 584 gellan gum pad (containing 0.55% gellan gum and 0.08 mM MgCl2 in MG liquid medium) on the 585 top of the coverslip to sandwich the cells against the bottom coverslip during time-lapse imaging. 586 After removing excess culture by a pipette, the chamber was filled with fresh MG liquid medium, 587 and the top of the chamber was covered with another coverslip to allow adequate microaerobic 588 conditions to support the growth of AMB-1 cells. The sample was set up under about 10% oxygen 589 atmosphere. Bacteria imaging and processing were then performed as previously described 16 . 590

Cellular fractionation 599
WT AMB-1 cells expressing pAK1255 were first grown in 50-mL MG medium in conical tubes 600 with a 1:100 dilution from stock cultures at 30 o C for two days and then grown in 2-L MG medium 601 in a microaerobic glove box (10% oxygen) for two days. These 2-L cells were pelleted by 602 centrifugation at 8,000 × g for 15 min and kept at -80 o C freezer for future use. Cell pellets were 603 thawed on ice and resuspended in 5-mL ice-cold 25 mM Tris buffer (pH 7.0). Pepstatin A and 604 leupeptin were added to a final concentration of 1 µg/mL, and PMSF was added to a final 605 concentration of 1 mM. The resuspension was passed through a French press two times at 1000 606 psi. From this step, all samples were kept on ice or at 4 o C. 20 µg/mL DNase I and 2 mM MgCl2 607 were added to the homogenate and incubated at 4 o C for 30 min. To separate the magnetosome 608 fraction, the cell lysates were passed through a magnetized MACS LS column (Miltenyi Biotec 609 Inc.) that was surrounded by magnets. After washing the column 3 times with 25 mM Tris buffer 610 (pH 7.0), the magnets were removed and the magnetosome fraction was eluted in 5-mL of 25 mM 611 Tris buffer (pH 7.0). To separate soluble and insoluble non-magnetic fractions, the column flow-612 through was centrifuged at 160,000 x g for 2 hours. The sedimented membrane fraction was 613 resuspended with 100 mM Tris buffer (pH 7.0) and both fractions were centrifuged a second time 614 at 160,000 x g for 2 hours. The resulting supernatant contained the non-magnetic soluble fraction 615 and the resuspended pellet contained the non-magnetic insoluble fraction. 616 Cellular fractions were analysed by SDS-PAGE as described previously 42 . In brief, different 617 fractions were mixed with 2x Laemmli Sample Buffer (Bio-Rad) and heated for 15 min at 95°C. 618 Proteins were resolved by Bio-rad stain-free any KDs gels before transfer to nitrocellulose 619 membrane by electroblotting. Immunological detection was performed with primary antibodies, 620 including anti-GFP polyclonal antibodies (Abcam), anti-Mms6 polyclonal antibodies (Produced 621 by ProSci Inc), or anti-HaloTag monoclonal antibody (Promega), and HRP-Conjugated secondary 622 antibodies (Bio-Rad). 623

Comparative genomics and molecular phylogenetics 638
McaA and McaB homologs were searched in other bacterial genomes available in public databases. 639 Protein sequences were aligned against reference proteins and non-redundant protein sequences of 640 the refseq_protein and nr NCBI databases respectively in October 2021 using the BLASTP 641 algorithm, a word size of 6 and default scoring parameters. A similar task was performed using 642 public genomic assemblies of MTB annotated with the Microscope platform 49 . BLAST hits with 643 an expectation value below 5 ´ 10 -2 were further analysed. First, pairwise sequence comparisons 644 were performed using BLASTP (BLAST+ version 2.10.0). Sequence clustering was then 645 performed with the Mmseqs2 50 clustering algorithm version 13.45111 to define groups of distant 646 homologs using the default parameters, a sequence identity threshold of 30% and an alignment 647 coverage of 80% for the longer sequence and for the shorter sequence. Guided by these first phylogenies, a second set of trees was built following the same approach 661 using genome sequences of strains for which transmission electron microscopy images of the 662 magnetosome chains are available and for which organisation could be compared with that of 663 Magnetospirillum magneticum AMB-1. Relationships between sequences and chain features were 664 then inferred after collecting all metadata including transmission electron microscopy images 665 published previously (Table S1). The synteny analyses were further explored using the tools 666 Tables 891 Table 1. The interaction results of BACTH. 892