There were consistent trends of white inclusions observed in STEM images and PHA concentrations analyzed through chemical analysis under different oxic and anoxic regimes, i.e., more white inclusions and higher PHA concentrations under oxic conditions and vice versa under anoxic conditions (Figures 2, 4, 5). Combined with EDX analysis, the results indicated that the carbon containing inclusions were highly PHA-rich. Intracellular PHA granules in bacteria occur mostly as different types of homo-and heteropolymers (i.e., PHB), and/or poly (3-hydroxybutyric-co-3-hydroxyvaleric acid) [P(3HB-co-3 HV)] (Khanna and Srivastava, 2005; García et al., 2013; Hierro-Iglesias et al., 2021). During oxic conditions in our different incubations, analysis of extracted PHA revealed the presence of PHB copolymerized with PHV in AMB-1, with the averaged ratio of PHB/PHV at (98 ± 1.9):(2.1 ± 1.7) (mol%; n = 160) (Figure 4). PHA aggregates have previously been reported in different MTB, such as Magnetospirillum gryphiswaldense strain MSR-1, Magnetococcus marinus strain MC-1, and Candidatus Magnetoglobus multicellularis, using chemical analytical and staining methods (e.g., Nile red staining; Schultheiss et al., 2005; Silva et al., 2008; Lefèvre et al., 2009; Fernández-Castané et al., 2017; Hierro-Iglesias et al., 2021). The calculated PHB/PHV ratio in our study was close to the value of 99:1 reported for MSR-1 in the exponential phase, which was grown under similar cultivation conditions (Hierro-Iglesias et al., 2021).

Referring to the carbon utilization by other PHA-accumulating microbes (Lemos et al., 1998; Lütke-Eversloh and Steinbüchel, 1999; Steinbüchel and Lütke-Eversloh, 2003), the potential carbon source of PHA synthesis for AMB-1 could be succinate, tartrate and acetate in medium (Supplementary Figure S2). These organics are converted to common intermediates, like acetyl-CoA that can be used as the substrate of PHA synthesis (Steinbüchel and Lütke-Eversloh, 2003). PHA yield was calculated as the unit amount of PHA produced (mmolC PHA) divided with unit amount of substrate consumed (mmolC carbon). For AMB-1 in our study, the estimated PHA yield was 0.036 ± 0.030 (maximum value of 0.11) under oxic conditions in all incubations, indicating that organic carbon added to medium was primarily used for cell growth (Figure 1; Supplementary Figure S2). The PHA yield was in the low range of values reported for other PHA-accumulating microbes (Lemos et al., 1998; Serafim et al., 2004; Albuquerque et al., 2007; Fradinho et al., 2014; Queirós et al., 2014; Sciarria et al., 2018). For instance, using an inoculum from a laboratory-scale EBPR reactor fed with acetate, Serafim and coworksers obtained a PHA storage yield of 0.68 mmolC PHA/mmolC carbon (Serafim et al., 2004). Bacteria enriched from activated sludge were reported to use acetate and butyrate as carbon sources for PHA synthesis, resulting in a maximum PHA yield of 0.77 ± 0.18 mmolC PHA/mmolC carbon (Sciarria et al., 2018). Since PHA granules occupied up to 88% of the cell area under oxic conditions (Figures 2, 5), the PHA yield of AMB-1 seemed to be underestimated, probably due to low recovery efficiency of the applied extraction method (Madkour et al., 2013). Further comparison between different PHA extraction methods using cultured AMB-1 cells would be helpful to achieve a more precise quantification of PHA storage capability. After depletion of organic carbon after day 10 in O7A5O8 and O3A3O3A3O8, and after approximately day 13 in O20, PHA continued to accumulate under oxic conditions (Figure 4; Supplementary Figure S2). The additional carbon source might be from the cell decay during anoxic periods (days 8–12 in O7A5O8, days 10–12 in O3A3O3A3O8), where cell growth stagnates and even decreases (Figure 1). In the oxic incubation, additional carbon source could originate during the stationary/death phase (days 14–20 for O20) (Figure 1). More investigations are needed in order to demonstrate the carbon source for PHA synthesis under different growth phases.

Oxygen had a profound effect on PHA storage in AMB-1 cells. Overall, bigger PHA inclusions and higher content of intracellular PHA inclusion were observed in cells during oxic periods, compared to anoxic conditions (Figures 2, 5). During continuous oxic conditions in O20, the number and area of PHA storage granules per cell increased remarkably with time, occupying 47 ± 23% of the cytoplasmic space on day 17 (Figures 2, 5; Supplementary Figures S7–S9). Conversely, PHA inclusions were not found during anoxic periods in A12O8, where they appeared after introducing O₂ to the bottle. Here, PHA inclusions accounted for 30 ± 10% of the cytoplasmic space on day 17 (Figure 5). Similar trends were also observed during short-term oxic-anoxic transitions in O7A5O8 and O3A3O3A3O8, where PHA inclusions accumulated in the presence of O₂, and decreased in both size and number in the absence of O₂ (Figures 2, 5; Supplementary Figures S7–S9). For instance, in O7A5O8, the area percentage of PHA inclusions decreased from 19 ± 5.2% on day 6 (oxic) to 3.0 ± 4.0% on day 12 (anoxic), and subsequently increased again to 41 ± 17% on day 17 (oxic) (Figure 5). The aforementioned microscopic observations were consistent with PHB and PHV results (Figures 2, 4, 5). During anoxic conditions, PHB and PHV accounted for 0.59 ± 0.66% and 0.0033 ± 0.0088% of dry cell weight, respectively, while values increased up to 9.9% (averaged 4.3 ± 2.5%) and 0.33% (averaged 0.14 ± 0.089%) of dry cell weight under oxic conditions.

The presence of PHA inclusions has been reported in many bacteria (Steinbüchel et al., 1992; Pötter and Steinbüchel, 2006). PHA granules can act as storage compounds of carbon and energy, which are required for the maintenance of metabolism and synthesis of cellular metabolites during starvation, in particular after growth resumes, as well as an electron sink into which excess of reducing power can be channeled (Wältermann and Steinbüchel, 2005; Pötter and Steinbüchel, 2006). Besides the primary storage function, PHA also can enhance robustness and survival of bacterial cells against environmental stress conditions, like under high or low temperature, freezing, oxidative and osmotic pressure, which is likely associated with their extraordinary architecture and biophysical properties (Obruca et al., 2020). It has been previously reported that the biosynthesis of PHA was promoted under imbalanced nutrient conditions, such as an excess of carbon source and electron donor and lack of another nutrient (e.g., nitrogen or sulfur) (Kessler and Witholt, 2001; García-Torreiro et al., 2016; Fernández-Castané et al., 2017). In our incubations, sodium succinate anhydrous, sodium tartrate and NO₃⁻ became limited from day 3–12 (Supplementary Figures S2, S3). Similar PHA storage and release patterns were observed under different oxic-anoxic transition regimes throughout the whole incubation period, revealing that O₂ had the strongest influence on PHA accumulation. The role of O₂ on PHA synthesis differs among different microorganisms (Mino et al., 1998; Borah et al., 2002). Similar to AMB-1, the presence of O₂ has been observed to enhance PHA accumulations by Bacillus mycoides (Borah et al., 2002), while PHA formations only took place after exhaustion of O₂ in PAOs and glycogen-accumulating organisms (GAOs) in EBPR systems (Mino et al., 1998).

The third type of granule identified inside AMB-1 cell was the phosphorus-rich granule, which was smaller and more electron dense than PHA granule, and less electron dense than the magnetosome. Each AMB-1 cell contained up to 12 spherical phosphorus-rich granules, with the diameter in the range of 0.00022–0.19 μm, while some cells contained no granules right in phosphorus (Figures 2, 3; Supplementary Figures S10, S11). Phosphorus-rich inclusions in single cells were located adjacent to the PHA inclusions. Elemental EDX analysis showed that these phosphorus-rich granules besides phosphorus contained oxygen and magnesium as major elements, and small amount of potassium and calcium (Figure 3). Dark red color was observed when cells were stained with toluidine blue (data now shown). Based on the dark red colour and the presense of metals in phosphorus-rich granules, we assumed that phosphorus-rich granules in the AMB-1 cells were consisted of polyP (Brock and Schulz-Vogt, 2011; Schulz-Vogt et al., 2019). The presence of similar phosphorus-rich inclusions has previously been observed in MTB with transmission electron imaging, especially in uncultured magnetotactic cocci (Lins and Farina, 1999; Cox et al., 2002; Keim et al., 2005; Lefèvre et al., 2009; Rivas-Lamelo et al., 2017; Schulz-Vogt et al., 2019; Li et al., 2021; Goswami et al., 2022). These phosphorus-rich inclusions were also classified as polyP. For example, the volume of polyP granules made up most of the cell volume of magnetotactic cocci in Lake Pavin, France (Rivas-Lamelo et al., 2017). Besides of phosphorus with a relative abundance of 65.9 ± 3.2%, the authors found magnesium, potassium and calcium elements associated with polyP granules in relative abundances of 22.8 ± 4%, 5.1 ± 3.1% and 6 ± 5.4%, respectively (Rivas-Lamelo et al., 2017). PolyP inclusions were also observed in MTB of the genus Magnetococcus in the suboxic zone of Black Sea (Schulz-Vogt et al., 2019). In the Black Sea MTB, the inclusions contained 26–34% phosphorus and 1–5% metals (e.g., iron and manganese) (Schulz-Vogt et al., 2019). PolyP is generelly composed of linear polymers of orthophosphate linked through high energy phosphoanhydride bonds (Kornberg, 1995; Cox et al., 2002). Each orthophosphate unit carries a monovalent negative charge at physiological pH, resulting in a large cation exchange capacity of polyP. The binding energy facilitates polyP to sequester Mg²⁺, Ca²⁺, K⁺ etc., (Reusch, 2000; Cox et al., 2002), consistent with the signals detected in EDX spectra of phosphorus-rich granules in AMB-1 cells (Figure 3). The binding of these metals leads to the high electron density of these polyP granules.

Similar to PHA inclusions, the presence of O₂ strongly enhanced the intracellular storage of polyP in AMB-1 cells, with more and bigger polyP granules under oxic conditions (Figures 2, 6; Supplementary Figures S10–S12). During continuous oxic conditions in O20, the number and area of polyP granules gradually increased from 3.4 ± 1.6 and 0.0032 ± 0.0014 μm² on day 3 to 6.5 ± 1.7 and 0.011 ± 0.0048 μm² on day 17, respectively, accounting for 0.69 ± 0.21% (day 3) and 5.1 ± 1.7% (day 17) of the cytoplasmic space. Short O₂ exposure time led to less polyP, with the area percentage of 1.9 ± 0.55% in A12O8, 2.4 ± 0.56% in O7A5O8, 2.8 ± 1.2% in O3A3O3A3O8 on day 17 (Figures 2, 6). PolyP granules were barely observed under anoxic conditions (Figures 2, 6). Similar to PHA synthesis, upon the depletion of PO₄³⁻ in medium, any additional phosphorus source for polyP synthesis might originate from cell decay during anoxic periods where cell growth stagnates and even decreases (day 8–12 in O7A5O8, day 10–12 in O3A3O3A3O8), and during the stationary/death phase in the oxic incubation (day 14–20 for O20) (Figure 1; Supplementary Figure S5). We did not observe any release of PO₄³⁻ after the transition from oxic to anoxic conditions. The amount of phosphorus released into medium might be too low to detect with the technique used in our study (detection limit of PO₄³⁻ was ~0.058 mM). The phosphorus utilization for polyP synthesis under different growth phases and nutrient availability remains to be investigated in future studies.

The physiological function of polyP inclusions in MTB remains unclear. According to other non-MTB polyP-accumulating bacteria, such as PAO in EBPR reactors, these polyP inclusions might serve as a source of ATP, or a response to oxidative stress (Seviour et al., 2003). Rivas-Lamelo et al. (2017) speculated that the massive accumulation of polyP in magnetotactic cocci in Lake Pavin was due to the effect of oxic/anoxic fluctuations, either by travelling vertically over short distances from the anoxic to oxic zone (and vice versa), or due to seasonal variation in the position of the oxic-anoxic interface. It has been hypothesized that MTB accumulated high polyP contents under oxic conditions, and released phosphate under anoxic conditions triggered by sulfide (Rivas-Lamelo et al., 2017; Schulz-Vogt et al., 2019).

The pathway of PHA synthesis in many bacteria involves enzymes of β-ketothiolase (PhaA), acetoacetyl CoA reductase (PhaB), and PHA polymerase (PhaC; Kessler and Witholt, 2001; García-Torreiro et al., 2016). Firstly, PhaA converts two molecules of acetyl-CoA to a molecule of acetoacetyl-CoA. The formed acetoacetyl-CoA is then stereoselectively reduced to form (R)-3-hydroxybutyryl-CoA by PhaB using NADH as the electron donor in most species. PhaC is responsible of PHA polymerization. The degradation of PHA is catalyzed by PHA depolymerase (PhaZ), which is able to hydrolyze amorphous native PHA granules yielding PHA monomers as final products (Kessler and Witholt, 2001; Liu et al., 2008; García-Torreiro et al., 2016). Similar pha genes have been detected in the genome of MSR-1 (Liu et al., 2008), which shares high genomic similarity with AMB-1 (Raschdorf et al., 2014). In a recent work, genomic excision of the phbCAB operon in MSR-1 was shown to eliminate the production of PHA granules (Raschdorf et al., 2014). In addition, presumptive magnetosome protein Mms16, associated with isolated magnetosomes from two Magnetospirillum strains (AMB-1 and MSR-1), is a PHB granule-bound protein (phasin) and acts in vitro as an activator of PHB hydrolysis (Schultheiss et al., 2005). There is however limited knowledge of the regulation of PHA metabolism in MTB at enzymatic and transcriptional level. Based on the metabolic response of other bacterial strains (e.g., Halomonas, Pseudomonas and Azotobacter) under nutrient deprivation, regulation of PHA metabolism might occur at different levels: activation of pha gene expression (i) by specific environmental signals, like nutrient starvation or (ii) by specific cell components or metabolic intermediates; (iii) inhibition of metabolic enzymes of competing pathways and consequently enrichment of required intermediates for PHA synthesis; or (iv) a combination of those (Kessler and Witholt, 2001; Castillo et al., 2013; García-Torreiro et al., 2016). Considering the higher AMB-1 growth rates and carbon consumption rates under oxic conditions (Figure 1; Supplementary Figure S2), enhanced carbon metabolism with high conversion of carbon to acetyl-CoA might result in more carbon fluxes spilled over into PHA synthesis (Castillo et al., 2013; García-Torreiro et al., 2016). Alternatively, the transition of anoxic or oxic condition might trigger the PHB synthesis by activating PHA gene expression or increasing the NADH/NAD⁺ ratio, where high NADH concentrations would inhibit citrate synthase and isocitrate dehydrogenase leading to the accumulation of acetyl-CoA for PHA synthesis (Senior et al., 1972; Senior and Dawes, 1973; Ling et al., 2018; Li et al., 2022). The exact mechanisms of O₂ mediated PHA synthesis in MTB call for further investigations.

Previous full genome studies indicated the presence of polyP kinases (Ppk1 and Ppk2), exophosphatases (Ppx) or phosphate regulon (Pho) in MTB, including AMB-1, MSR-1, Magnetospirillum magnetotacticum MS-1, MC-1 and Magnetofaba australis strain IT-1, resulting in the potential ability of MTB to synthesis and degrade polyP granules (Zhang et al., 2002; Rao et al., 2009; Schübbe et al., 2009; Araujo et al., 2016; Zhou et al., 2017; Koziaeva et al., 2019). Pho controls phosphate uptake, and Ppk1 reversibly catalyzes the formation of polyP, whereas Ppk2 and Ppx degrade polyP to produce ATP or phosphate (Rao et al., 2009; Schübbe et al., 2009). Besides, PolyP:AMP phosphotransferase (Pap) is a class II Ppk2, which can transfer the terminal phosphate residue from poly-P to AMP, producing ADP. For example, bacterial polyP inclusions in MTB affiliated with the genus Magnetococcus were found to contribute substantially to the phosphorus peak observed at the lower boundary of the suboxic zone of the Black Sea (Schulz-Vogt et al., 2019). The phosphorus maximum correlated with an increase in gene expression of ppk by several groups of bacteria including those of the family Magnetococcaceae, suggesting active bacterial polyP degradation (Schulz-Vogt et al., 2019). MTB were therefore proposed to shuttle up and down within the suboxic zone, scavenging phosphate at the upper of the suboxic zone and releasing it at the lower boundary. This is consistent with the results in our batch tests: when the growth of AMB-1 resumed under oxic conditions, the biosynthesis of PHA and polyP was promoted, which could act as storage compounds for energy and carbon needed for maintenance of metabolism and synthesis of cellular metabolites under anoxic conditions.

As homopolymers with unique molecular characteristics, both polyP and PHA (especially PHB) can assist to regulate internal ion concentrations by serving as vehicles for selective transport of ions across membranes (Reusch, 2000). PolyP is often correlated with PHB, where PHB solvates (dissolves) cations, and associates with polyP to form selective ion channels across plasma membranes (Reusch, 2000). In our work, we observed remarkable accumulations of carbon-and phosphorus-rich granules together with fewer magnetosomes (per cell) under oxic incubation periods (Figures 5, 6; Supplementary Figure S6). Since PHA formation diverts cellular resources from growth, high levels of PHA might hinder magnetosome preparation and synthesis (Fernández-Castané et al., 2017). Our results are in agreement with recent works regarding the energy competition between the formation of PHA and magnetosomes, where magnetosome production was negatively correlated with PHA formation (Raschdorf et al., 2014; Fernández-Castané et al., 2017, 2018). Furthermore, phosphate metabolism may be associated with magnetosome biosynthesis. It has been proposed that magnetite formation in MTB proceeds from the storage of iron in the form of phosphate-rich ferric hydroxide (FeP), which supposedly transforms to a transient and short-lived ferrihydrite (Fe₂O₃·nH₂O) followed by the reduction to form the final magnetite mineral (Baumgartner et al., 2013). However, considering the relatively small size of magnetosomes [(1.7 ± 0.84) × 10⁻³ μm²], it appears difficult to follow the release of PO₄³⁻ into medium after the separation from FeP.