Proteomic response of the marine ammonia-oxidising archaeon Nitrosopumilus maritimus to iron limitation reveals strategies to compensate for nutrient scarcity

Summary Dissolved iron (Fe) is vanishingly low in the oceans, with ecological success conferred to microorganisms that can restructure their biochemistry to maintain high growth rates during Fe scarcity. Chemolithoautotrophic ammonia-oxidising archaea (AOA) are highly abundant in the oceans, constitut-ing ~30% of cells below the photic zone. Here we examine the proteomic response of the AOA isolate Nitrosopumilus maritimus to growth-limiting Fe concentrations. Under Fe limitation, we observed a signi ﬁ cant reduction in the intensity of Fe-dense ferredoxins associated with respiratory complex I whilst complex III and IV proteins with more central roles in the electron transport chain remain unchanged. We concomitantly observed an increase in the intensity of Fe-free functional alternatives such as ﬂ avodoxin and plastocyanin, thioredoxin and alkyl hydroperoxide which are known to mediate electron transport and reactive oxygen species detoxi ﬁ cation, respectively. Under Fe limitation, we found a marked increase in the intensity of the ABC phosphonate transport system (Phn), highlighting an intriguing link between Fe and P cycling in N . maritimus . We hypothesise that an elevated uptake of exogenous phosphonates under Fe limitation may either supplement N . maritimus ’ endogenous methylphosphonate biosynthesis pathway - which requires Fe - or enhance the production of phosphonate-containing exopolysaccharides known to ef ﬁ ciently


Introduction
Ammonia-oxidising archaea (AOA) occupy a diverse array of ecosystems from hot springs to the oligotrophic open ocean. All cultivated AOA, recently placed into the updated class of Nitrososphaeria within the phylum Thermoproteota (Parks et al., 2020;Rinke et al., 2020) perform the first step of nitrification, the central component of the biogeochemical nitrogen (N) cycle, by oxidising ammonium (NH 4 + ) to nitrite (NO 2 À ). In the oceans, surface water nitrification can account for up to 30% of phytoplankton nitrogen nutrition (Yool et al., 2007) and provides an oxidised nitrogen niche for the large microphytoplankton (such as diatoms) that drive greater carbon export to the deep ocean relative to picophytoplankton such as cyanobacteria (Berg et al., 2003;Fawcett et al., 2011;Glibert et al., 2016). Marine AOA constitute up to 30% of total prokaryotic cells below the photic zone (Karner et al., 2001) and as such contribute a major fraction of gene copies and transcripts in marine metagenomes and metatranscriptomes (Hollibaugh et al., 2011). Marine AOA also play a role in major biogeochemical cycles through their contribution to the production of nitrous oxide (Freing et al., 2012), methane (Metcalf et al., 2012) and cobalamin (Doxey et al., 2015;Heal et al., 2017).
The success of marine AOA is owed to adaptations which make these microorganisms well suited to the inhospitable open ocean wherein nutrient fluxes are low. Large surface-to-volume ratios (Könneke et al., 2005), a high affinity for NH 4 + (Martens-Habbena et al., 2009), an energy-efficient C-fixation pathway (Könneke et al., 2014), and streamlined genomes with a low proportion of non-coding DNA (Walker et al., 2010;Santoro et al., 2015) make marine AOA oligotrophic-specialists or 'oligotrophs'. In~30% of surface oceans, low concentrations of dissolved iron (Fe) limit the growth of marine phytoplankton (Moore et al., 2013), but the role of Fe in modulating the growth of non-photosynthetic but environmentally important microbial groups has traditionally received much less attention. Recently we demonstrated that the model AOA isolate Nitrosopumilus maritimus strain SCM1, despite possessing a unique Cu-dominated centric biochemistry compared to its ammonia-oxidising bacterial counterparts (Walker et al., 2010;Amin et al., 2013;Shafiee et al. ,2021), requires a greater concentration of unchelated Fe (Fe 0 ) for growth than many cosmopolitan marine microorganisms (Shafiee et al., 2019). A greater Fe requirement for growth relative to other marine microorganisms suggests that AOA must occupy an Ferich niche and indeed aligns with observations of AOA maxima at the base of the photic zone (Santoro et al., 2013;Shiozaki et al., 2016). At this depth, Fe supply is great due to the combined effect of a decrease in competition for Fe (as phytoplankton become light limited) and Fe remineralisation from sinking organic matter. Fe fluxes from remineralisation can be episodic, depending heavily on levels of surface productivity and as such AOA may undergo prolonged periods of Fe starvation between nutrient pulses. Marine microorganisms have evolved a number of strategies to survive Fe scarcity including a reduction in Fe demand by downregulating the expression of Fe-binding proteins (retrenchment); replacing Fe-dense proteins with Fefree substitutes (La Roche et al., 1996;Erdner et al., 1999;Snow et al., 2015); producing Fe-chelating compounds to enhance access to extracellular Fe (Boiteau et al., 2016) or increasing the transcription or affinity of membrane transporter proteins (Kustka et al., 2007;Allen et al., 2008;Cohen et al., 2018;Maldonado et al., 2006). The biochemical strategies implemented under Fe scarcity can be gleaned from the change in protein expression (proteome) which acts as the basis of developing biomolecular tools to detect Fe limitation in vast expanses of mixed microbial communities in situ. For example, the phytoplankton Fe deficiency-induced protein A (IdiA) involved in cellular Fe scavenging has been used as a marker for Fe stress in open ocean field populations (Webb et al., 2001;Saito et al., 2014;Chappell et al., 2015). The application of proteomic biomarkers provides a powerful alternative to traditional nutrient amendment experiments -which often alter community composition and do not necessarily capture the multicomplexed nature of stressors -which can be applied to wide geographic regions.
Here we examine the proteomic response of the model marine AOA isolate, Nitrosopumilus maritimus strain SCM1 (herein referred to as N. maritimus) to Fe limitation. We present the label-free quantitative intensity profiles of N. maritimus under Fe-limited and Fereplete growth, thereby gaining important insight into the potential mechanisms adopted by AOA to navigate Fe scarcity.

Results and discussion
Growth rates, cellular Fe quota and protein detection Low concentrations of unchelated iron ([Fe 0 ]) led to a 59% reduction in specific growth rate (μ) and a decline in final nitrite (NO 2 À ) concentration from 913 to 185 μmol L À1 ( Fig. 1; Supplementary Fig. 1) between treatments. In Fe-replete N. maritimus we measured an intracellular-Fe quota of 1.1 Â 10 À10 μmol Fe cell À1 (±4.2 Â 10 À11 ) which reduced to 2.25 Â 10 À12 μmol Fe cell À1 (±3.9 Â 10 À13 ) in Fe-limited cells (Fig. 1). Of the 1795 predicted protein-encoding genes in the N. maritimus genome (www.jgi.doe.gov), 1346 discrete proteins were detected in the Fe-replete proteome (74.9%) and 1337 proteins in the Fe-limited proteome (74.5%) (at a 1% FDR). Biological replicates showed strong reproducibility with r 2 values exceeding 0.8 in all comparisons ( Supplementary Fig. 2). The detected proteome coverage was greater or consistent with previous proteomic recovery of N. maritimus (68% in Bayer et al., 2019;74% in Qin et al., 2018) and Candidatus Nitrosopelagicus brevis (~70% in Santoro et al., 2015) and supports the notion that AOA have streamlined genomes with a high proportion of predicted protein expression. Sixty-nine proteins showed a significant difference in label-free quantitation (LFQ) intensity between Fe treatments (FDR adjusted pvalue <0.05). All differential protein intensities discussed in this study are statistically significant at an FDRadjusted p-value <0.05. Relative to the Fe-replete treatment, 34 proteins increased significantly in LFQ intensity and 35 decreased in intensity in the Fe-limited treatment. Three proteins that were detected in the Fe-limited proteome were absent in all replicates (or below L.O.D.) of the Fe-replete proteome, whereas 12 proteins were detected in the Fe-replete proteome that was absent from all replicates of the Fe-limited proteome (Supplementary  Table 1). Proteins with predicted functions in the S-layer (Nmar_1547), ammonia oxidation (Nmar_1102), carbon fixation (Nmar_0273) and genetic information processing (Nmar_1034) showed the highest LFQ intensities in both Fe-limited and Fe-replete proteomes (Table 1).

Ammonia monooxygenase
Ammonia-monooxygenase (AMO), the hallmark enzyme of ammonia-oxidising microorganisms, is comprised of three subunits: AmoA, B and C (Nmar_1500, Nmar_1503, Nmar_1502). Under Fe limitation, AMO subunit LFQ intensity increased as a proportion of total signal from 1.75% to 2.37% with AmoB among the top 10 most intense proteins in the Fe-limited proteome (Table 1). Additionally, we found that AmoA (Nmar_1500) intensity increased with statistical significance in Fe-limited treatments -suggesting that N. maritimus' core ammonia oxidation machinery remains intact even when Fe is scarce. This highlights that the availability of the AMO complex may not be the rate-limiting factor which leads to the observed decline in ammonia oxidation rate when Fe 0 is low ( Fig. 1, Supplementary Fig. 1). In contrast, energy and Cu limitation reduces the transcription of all AMO subunits (Qin et al., 2018;Reyes et al., 2020), apart from amoC which is preferentially retained under NH 4 + starvation (Qin et al., 2018). The ostensible greater impact of Cu and NH 4 + limitation on the AMO complex may be expected given the larger predicted Cu requirement of AMO (based on AMO homology to methane monooxygenase) and the fact NH 4 + is required as its substrate: AmoC and AmoA putatively bind 7 Cu atoms collectively compared to AmoB which only requires 3 Fe atoms. However, it cannot be ruled out that transcriptional changes under Fe limitation are not apparent at the proteomic level due to differential rates of posttranslational protein degradation. This potential is highlighted by the divergence in AMO subunit abundance between metatranscriptomic and metaproteomic analyses: marine amoC and amoA transcripts typically outnumber amoB transcripts in metatranscriptomes (Feike et al., 2012;Shi et al., 2012) but initial metaproteomic surveys show that Nitrosopumilus-like AmoA and AmoB proteins are~three-fold more abundant and widespread compared with AmoC (Saunders et al., unpubl;Saito et al., 2020;Supplementary Fig. 3). This disparity highlights the importance of complementary proteomic and transcriptomic characterisation of AMO stoichiometry in future research.

Ammonia oxidation and the electron transport chain
In Fe-replete treatments, Complex I proteins were among the most intensely detected proteins in the N. maritimus proteome (Table 1) and are also predicted to be the most Fe dense based on our analysis of Fe-binding proteins in N. maritimus' genome-predicted proteome (Supplementary Table 1; See Biological function groups and Fe containing proteins). Complex I and putatively associated ferredoxins require a predicted 16 atoms of Fe compared with AmoC, complex III and complex IV which we predict in total require only 11 Fe atoms (Supplementary Table 1). Complex I is responsible for the generation of A B Fig. 1. A. Nitrosopumilus maritimus strain SCM1 growth rate and cellular quota (Q Fe ) under Fe-replete and Fe-limited growth, corresponding to 250 and 1050 pmol L À1 Fe 0 . n = 9. B. Relative growth rate (μ/μ max ) as a function of intracellular iron quota (Q Fe ) in N. maritimus and other marine microorganisms (Shafiee et al., 2021;Granger & Price, 1999;Fourquez et al., 2014;Fu et al., 2008;Cunningham & John, 2017;Shire & Kustka, 2015;Wilheim, Maxwell & Trick, 1996;Shi et al., 2012;Sunda & Huntsman, 1995) Nitrosopumilus maritimus n = 3. NAD(P)H -a reducing agent involved in numerous cellular processes including nutrient assimilation, biosynthesis, carbon fixation and metabolism. In Fe-limited treatments, we observed a significant reduction in the intensity of complex I proteins and ferredoxins (Nmar_0279; Nmar_0284; Nmar_1765; Nmar_1537, Nmar_0239) but no change in intensity of Fe-containing complexes III and IV or AmoC (Fig. 2). Complex III and amoC transcripts were also retained in N. maritimus under energy and nutrient limitation (Qin et al., 2018), pointing to their crucial roles in generating a proton motive force and stabilising the AMO holoenzyme under resource limitation. Qin and colleagues also found ferredoxins (specifically Nmar_1765 and Nmar_0239) to be among the most downregulated genes under Cu and NH 4 + limitation ( Supplementary Fig. 4). This suggests that the eschewal of Fe-containing ferredoxins may be a general, rather than an Fe-specific, limitation response (Supplementary Table 1). This assumes that the reduced intensity of ferredoxins in our study is primarily underpinned by a downregulation of ferredoxin genes as opposed to differential protein degradation rates between Fe treatments. In Fe-limited treatments we also observed a significant increase in the LFQ intensity of riboflavin synthase (Nmar_1617), flavodoxin (Nmar_1356) and a putative Cu-binding plastocyanin-like protein (Nmar_0815) suggesting that when starved for Fe, N. maritimus may functionally replace ferredoxins for Fe-free electron carriers as has been observed in numerous marine microorganisms under Fe limitation (La Roche et al., 1996;Peers and Price, 2006;Snow et al., 2015;Sepúlveda Cisternas et al., 2018). Both flavodoxin (Nmar_1356) and the Cu-binding plastocyanin-like protein (Nmar_0815) are downregulated in response to Cu limitation in N. maritimus (Qin et al., 2018) thereby suggesting that these proteins may potentially form part of a Fe-specific limitation response (Supplementary Fig. 4). Taken together, we posit that a reduction in Complex I ferredoxins and the potential substitution with Fe-free electron carriers, combined with the retention of Complex III and IV proteins, and AmoC presents a compensatory strategy designed to most efficiently reduce Fe demand while keeping energy production as intact as possible. A comparable compensatory strategy has been observed in cosmopolitan marine phytoplankton such as Trichodesmium (Shi et al., 2007;Snow et al., 2015). Yet despite such functional replacements, there is still a significant decline in growth rate under low Fe (Fig. 1A) suggesting that such potential functional substitution is not perfect and/or that the rate of Fe uptake is still insufficient under low Fe 0 to meet the lowered Fe demandsignalling N. maritimus' integral need for greater extracellular [Fe 0 ] to maintain a high growth rate.

Carbon fixation, TCA cycle and biosynthesis
The HP/HB cycle adopted by N. maritimus to fix carbon contains only one predicted Fe-binding protein (Nmar_1110), which we did not find to change significantly in intensity between Fe treatments. In contrast, we found that several predicted Fe-containing and Fe-free proteins involved in the TCA cycle and biosynthesis processes changed significantly in intensity between Fe treatments (expanded discussion below). Taken together, this supports the previous notion that Feretrenchment targets the processes which are densest in their Fe-use allowing N. maritimus' carbon-fixing capacity to remain intact.
In Fe-limited treatments, we observed significant changes to the intensity of proteins involved in the TCA cycle and biosynthesis indicating the presence of potential strategies to mitigate the effects of Fe depletion (Fig. 3). The TCA cycle is central to metabolism -releasing stored energy through the oxidation of acetyl-CoA and providing essential intermediates for biosynthesis. In marine phytoplankton, TCA intermediates are estimated to account for one-third of the total Fe requirement (Raven, 1988) and must be substituted for Fe-free alternatives under low Fe (Cohen et al., 2018). Enzymes with Fe-containing and Fe-free variants include fumarate hydratases and fructose biphosphate aldolases  (Horecker et al., 1972;Picaud et al., 2011). N. maritimus only has the genetic capability to produce the Fe-free variants of these enzymes indicating that the TCA cycle already has a parsimonious Fe demand irrespective of extracellular Fe. In Fe-limited treatments we observed a significant drop in intensity of a ferredoxin (Nmar_0413) predicted to mediate the breakdown of pyruvate to acetyl-CoA -a key step in acetyl-CoA supply to the TCA cycle (Fig. 2). Concurrently 4-hydroxy-2-oxovalerate aldolase (Nmar_0149), which supplies pyruvate to Nmar_0413 and Nmar_0414 from the breakdown of pentanoic acid, increased in intensity (Fig. 2). Considered together, we speculate that in order to compensate for the decline in ferredoxins under Fe limitation, N. maritimus may instead increase substrate supply (pyruvate) to the ferredoxins -thus saturating these enzymes with pyruvate and in doing so maintaining crucial acetyl-CoA supply to the TCA cycle under Fe limitation.

Detoxification of reactive oxygen species
In addition to their role in ammonia oxidation and the TCA cycle, ferredoxins participate in cellular defence against oxidative stress (Krapp et al., 1997), signifying the potential for a reduced capacity of Fe-limited N. maritimus to deal with oxidative stress. Additionally, the breakdown of Fe-S proteins under Fe limitation has been shown to directly generate excess oxidative stress (Liochev and Fridovich, 1994;Keyer and Imlay, 1996;Jang and Imlay, 2010). Therefore in our study, the breakdown of Fe-S clusters is implied by a significant reduction in the intensity of both Fe-containing proteins throughout the cell and the Suf protein family which are responsible for Fe-S cluster biosynthesis and repair (Supplementary Fig. 5: Nmar_0516;Nmar_0496;Nmar_0495). The potential for increased oxidative stress in Fe-limited cells is suggested by a significant increase in Nmar_1386 intensity, a protein annotated as RadA ( Supplementary  Fig. 5), shown to be responsible for DNA damage repair across the archaeal domain (Komori et al., 2000;Seitz and Kowalczykowski, 2000). RadA was also found to increase significantly in the proteomes of AOA exposed to H 2 O 2 (Bayer et al., 2019). We observed a concomitant increase in the intensity of proteins showing homology to Fe-free detoxification proteins thioredoxin reductase (Nmar_0672) and alkyl hydroperoxide (Nmar_1252). This suggests that N. maritimus may be reconfiguring its detoxification mechanisms to Fe-free alternatives -a common strategy employed by Fe-limited microorganisms to mitigate enhanced oxidative stress (Holmes et al., 2005;Yingping et al., 2014). We also found that a protein belonging to the universal stress family of proteins, UspA (Nmar_1659) showed a significant 1.7 log 2 fold increase in intensity in Fe-limited treatments ( Supplementary Fig. 5). UspA proteins in E. coli play a role in defence against superoxide-generating agents and in cell growth retardation during the stationary phase (Nyström and Neidhardt, 1994;Nachin et al., 2005). However, Bayer et al (2018) found that UspA family proteins were almost completely absent in cultures exposed to H 2 O 2 in N. maritimus, increasing in abundance only when grown in co-culture with a heterotrophic bacterium or supplemented with catalase -suggesting that UspA is regulated by something other than general, cellular oxidative stress. As uspA (Nmar_1659) was found to be downregulated in response to Cu and NH 4 + limitation by Qin et al. (2018), we postulate that this UspA family protein may function as an Fe-specific response protein although its exact role in protecting against Fe-induced stress remains to be explored.

S-layer proteins
In both Fe treatments, Nmar_1547 shows the greatest LFQ intensity in the N. maritimus proteome, contributing 18.9%-21.7% of all intensities (Table 1) The key role played by the S-layer in nutrient and energy acquisition sheds clarity on why there is a significant proteomic investment in S-layer biosynthesis. However, we did not observe any significant change in intensity of detected S-layer proteins, signalling that the S-layer does not respond to changing Fe.

Iron reductase activity and ABC transport
Spatially, following the S-layer, transmembrane proteins are the next selective barrier between the N. maritimus cell interior and external environment. In response to Fe limitation, microorganisms may implement numerous strategies to enhance access to extracellular Fe including altering the number or composition of transmembrane proteins, secreting compounds which solubilise inaccessible Fe, or increasing the activity of ferrireductases. The genome of N. maritimus has little diversity in metal transport proteins, consistent with characterised oligotrophs (Hogle et al., 2016) and further supports genome streamlining in AOA (Santoro et al., 2015). The majority of AOA isolates, including N. maritimus, only encode general cation transporters (Walker et al., 2010;Amin et al., 2013;Shafiee et al., 2019) which were not found to change in intensity under Fe limitation (Fig. 4). Physiological experiments indicate that N. maritimus reduces Fe 3+ to Fe 2+ prior to uptake (Shafiee et al., 2019), but the ferric reductase protein responsible has yet to be identified. One of the only Fe-containing proteins to increase in intensity under Fe limitation was Nmar_0679, currently annotated as a haem and Fe-S binding sulphite reductase. Sulphite reductases have been shown to reduce a broad range of Fe-chelated substrates via a similar mechanism to NAD(P)H:flavin oxidoreductase (Coves et al., 1993). If Nmar_0679 is indeed responsible for the Fe-reductase activity in N. maritimus, its elevated intensity in response to Fe limitation may signal a potential strategy to increase the bioavailability of dissolved Fe to the cell transporters.
In Fe-limited treatments, several components of the ABC phosphonate transporter (PhnCDE) increased significantly in LFQ intensity (Nmar_0250, Nmar_0874, Nmar_0875) or were only detected in Fe-limited treatments (Nmar_0876 - Fig. 4). Furthermore, a putative transcriptional regulator in the same operon as the phosphate PstSCAB transport system (Nmar_0478) shows the greatest log 2 fold increase in intensity out of all analysed proteins by an order of magnitude (Fig. 4). Out of these proteins, PhnD (Nmar_0875) and the putative transcriptional regulator of the phosphate PstSCAB transport system (Nmar_0478) show the greatest LFQ intensity in Fe-limited cultures, individually representing 5% and 1.5% of all signal respectively (Table 1). Previous work has shown that phnD is downregulated under Cu limitation and NH 4 + starvation in N. maritimus (Qin et al., 2018), and that Phn proteins decrease in abundance in numerous Nitrosopumilus-strains upon exposure to H 2 O 2 (Bayer et al., 2019). Therefore, we suggest that the significant increase in PhnD intensity observed in our study may be a Fe-specific limitation response (Supplementary Fig. 4; Supplementary Table 1). Phosphonates can be a significant source of phosphorus (P) to microbial communities starved of inorganic P (Cook et al., 1978;Clark et al., 1999;Dyhrman et al., 2006). As N. maritimus does not possess proteins with known C-P lyase activity and inorganic P limitation is not alleviated by supplementation with phosphonates (Metcalf et al., 2012;Walker et al., 2010), a link between Fe limitation and P nutritional status is unlikely. An analysis of the spent media from N. maritimus cultures shows that N. maritimus produces weak ligands as by-products of natural growth -biomaterials hypothesised to be similar to exopolysaccharides (EPS) (Amin et al., 2013). EPS are effective scavengers of cations such as Fe 2+ /Fe 3+ (Hassler et al., 2011;Hasslet et al., 2015;Gutierrez et al., 2012) owing in part to the negative charge carried by P-bearing groups of the EPS (Ozdemir et al., 2003). P-bearing functional groups have been thermodynamically characterised in Cu 2+ complexation at the N. maritimus cell surface (Gorman-Lewis et al., 2019) and are proposed to be due to the presence of EPSassociated methylphosphonate, which is produced endogenously by N. maritimus (Metcalf et al., 2012). The hallmark enzyme of methylphosphonate (Mpn) biosynthesis in N. maritimus, methylphosphonic acid synthase (MpnS), binds two Fe atoms (Metcalf et al., 2012) and the final step in the pathway requires free Fe 2+ as an electron donor ( Supplementary Fig. 6). We speculate that the observed increase in PhnCDE intensity under Fe limitation may represent a strategy to enhance the uptake of exogenous phosphonates thereby compensating for a reduction in the cell's own methylphosphonate biosynthesis capacity which relies on Fe. While we did not observe any significant change in intensity to proteins of the N. maritimus' putative Mpn biosynthesis pathway (Supplementary Fig. 6A), recent analyses of AOA mpnS gene biogeography (Qin et al., 2020) highlights the potential for Fe to play a role in regulating Mpn production by acting as a selective pressure for the presence of mpnS genes in AOA populations which warrants further investigation. The analysis showed that mpnS genes are enriched in open-ocean AOA populations in the mesopelagic (>200 m) relative to surface waters (<200 m), thereby mirroring profiles of dissolved Fe which show an increase in concentration with depth ( Supplementary Fig. 7). Based on the results presented in this study, it may be expected for phnD gene abundance to follow an inverse trend to mpnS abundance, showing an enrichment in shallow AOA populations wherein Fe is scarce relative to deeper populations which have excess to more plentiful Fe. To test this hypothesis, we performed a similar analysis to Qin et al. (2020) Fig. 8). A greater variance in phnD: amoA in shallow water AOA communities is consistent with the high variability in dissolved Fe concentrations in the surface waters compared with depth ( Supplementary  Fig. 7), driven by more intensive biogeochemical cycling of Fe in the surface water. Additional analyses would be required to rule out that mpnS and phnD gene biogeography is not driven by other physio-chemical factors or ecological factors, such as the dominance of deep-water ecotypes (WCB) at depth (Santoro et al., 2019). An alternative hypothesis is that the marked increase of PhnCDE intensity under Fe limitation is to drive greater phosphonate uptake as a means of enhancing (not compensating for) N. maritimus' methylphosphonate production -potentially explaining why methylphosphonate Fig. 4. Proteins involved in membrane transport encoded by N. maritimus genome, with proteins showing significant log 2 -fold changes in LFQ intensity between Fe-replete and Felimited treatments shown in colour. Protein colour on schematic corresponds to the bars on the plot below. biosynthesis proteins remain unchanged in response to Fe limitation. In doing so, more P-bearing groups may be provided to the EPS making the cell surface more chemically attractive to extracellular Fe ( Supplementary Fig. 6C and D). We speculate that increased Fe binding to Pbearing functional groups on the N. maritimus cell surface could provide more Fe substrate to N. maritimus' ferric reductase, consistent with an increase in the intensity of the aforementioned ferric reductase candidate under Fe limitation (Nmar_0679). In both instances these conjectures highlight that a more detailed characterisation of EPS of N. maritimus and its relationship with Fe cycling and phosphonates is required to robustly explain the increase in PhnCDE under Fe limitation observed in our study.

Conclusion
When growth is limited by Fe, N. maritimus primarily minimises Fe demand by reducing Fe-dense ferredoxins which putatively interact with Complex I to generate the reduced co-factor NAD(P)H, potentially functionally substituting these ferredoxins for Fe-free alternatives. In doing so, Fe-binding proteins which play a more central role in electron transfer and ammonia oxidation can be preferentially retained, allowing N. maritimus to continue in supporting key energy-yielding pathways under Fe limitation. We observed additional compensatory strategies within the TCA cycle, biosynthesis and detoxification demonstrating efforts to maintain the activity of these pathways whilst significantly reducing Fe requirements. Furthermore, in response to Fe limitation we observed a significant increase in the intensity of proteins with a putative role in the uptake of exogenous phosphonates. We speculate this reflects a potential strategy to enhance or compensate for a reduction in N. maritimus' endogenous methylphosphonate biosynthesis pathway which requires Fe-binding proteins.

Nitrosopumilus maritimus cultures and growth media
Polycarbonate culture vessels and culturing apparatus were acid cleaned (10% HCl vol./vol., 24 h) and UV sterilised prior to use. Triplicate cultures of N. maritimus strain SCM1 (herein referred to as N. maritimus) were grown in modified SCM medium (Könneke et al., 2005) buffered with HEPES at pH 7.5 and maintained at 28 C in the dark. Basal salt medium and macronutrients were treated with Chelex® 100 chelating ion exchange resin (BioRad, Watford, UK) to remove trace metal contaminants (Price et al., 1989) and passed through an acidwashed (10% HCl as before) 0.1 μm polycarbonate filter to sterilise in a metal-free clean laboratory. Nitrosopumilus maritimus was cultured at inorganic Fe concentrations ([Fe 0 ]) previously shown to be limiting (250 pmol L À1 Fe 0 ) and replete (1050 pmol L À1 Fe 0 ) to growth (Shafiee et al., 2019) but without promoting excess Fe storage. Ethylenediamine tetraacetic acid (Merck, Darmstadt, Germany) (final concentration 12 μmol L À1 ) was used to buffer Fe and other metals in cultures; [Fe 0 ] was controlled by varying the addition of FeCl 3 from filter-sterilised concentrated stocks made using >99.999% trace metal basis salts (Sigma-Aldrich, UK). [Fe 0 ] in cultures was calculated using Visual Minteq software (Gustafsson, 2010), including background Fe contamination of 50 nmol L À1 from basal SCM medium with included macronutrients and vitamins but without added trace metals -this was determined using inductively coupled plasma mass spectrometry (ICP-MS) as described by Zhang et al. (2018). Cultures were acclimated to limiting and replete [Fe 0 ] by transferring cultures consecutively into new media until specific growth rates (μ) did not vary with statistical significance (ANOVA, p < 0.05). Growth was determined by measuring concentrations of nitrite ([NO 2 À ]) spectrophotometrically over time (Griess, 1879), which correlates with cell counts even at low [Fe 0 ] (Shafiee et al., 2019). Specific growth rate (d À1 ) was calculated over the linear phase of semilog plots of nitrite concentration over time. Differences in specific growth rate (μ) between 250 and 1050 pmol L À1 Fe 0 treatments were tested for statistical significance using a two-tailed student's T-test in Minitab software (version 19.2020.2.0).

Cellular iron and phosphorus quota
All samples and reagents were processed using tracemetal clean techniques and in acid-washed plastic wear (10% HCl v/v, 24 h) in a clean laboratory prior to trace element analysis. N. maritimus cells cultured in 2L triplicates at 250 and 1050 pmol L À1 Fe 0 were harvested in mid-exponential phase by centrifugation at 5000g for 40 min at 4 C. Cell pellets of each replicate were divided into sub-samples for further trace metal analysis and proteomic analysis (cells for protein extraction were stored at À80 C prior to protein extraction). For trace metal analysis, cell pellets were washed three times with chelexed-SCM basal salt media, which has shown to be effective in removing weakly bound surface (Zhang et al., 2018;Shafiee et al., 2019). Cells were digested in acid-cleaned Teflon as follows: refluxing at 100 C with 16 N quartz distilled (q.d.) HNO 3 (produced in-house) and H 2 O 2 (ROMIL UpA™, Cambridge, UK) in a ratio of 3:2 overnight. The liquid was then allowed to evaporate to dryness and 2% q.d. HNO 3 was added to reflux for an hour to resuspend the dried samples. 56 Fe and 31 P concentrations were determined using a Quadrupole ICP-MS Perkin Elmer NexION 350 with Elemental Scientific Flow Injection Auto-sampler (FIAS) in a method optimised for high salt matrixes (Zhang et al., 2018). Per cell Fe quotas were calculated based on the relationship between NO 2 À and cell counts delineated in Shafiee et al. (2019).

Protein extraction
Prior to protein analysis frozen cell pellets were thawed on ice, then lysed by sonication in 10 Â 10s bursts (Hielscher UP200S ultrasonic processor, 70% amplitude) in 1 ml of 0.5% SDS/0.5% Triton extraction buffer (on ice). A 0.5% SDS/0.5% Triton extraction buffer (in a base of 20 mM Tris-Cl, pH 8) was predetermined to be the most effective detergent ratio for protein extraction from a range of buffer combinations in preliminary extraction experiments using N. maritimus cells ( Supplementary  Fig. 9). Pierce BCA protein assay kit (Thermo Fisher Scientific, UK) was used to determine protein concentration following extraction, using the same matrix as the protein extraction buffer. Protein concentration in extracts ranged between 0.6 μg μl À1 in Fe-limited samples to 1.42 μg μl À1 in Fe-replete cells. Forty micrograms of protein digest from each biological replicate was used for subsequent proteolytic digestion and detergent removal.

SP3 protocol
Proteolytic digestion and detergent removal from the cell lysate was performed using a single-pot, solid-phaseenhanced sample preparation protocol (Hughes et al., 2019). Proteins in extraction butter were denatured in 8 M urea (prepared in 100 mM ammonium bicarbonate buffer) at room temperature (RT) for 30 min. Disulphide bridges were reduced with Bond-Breaker™ TCEP Solution (Thermo Scientific) to a final concentration of 10 mM at RT for 30 min, followed by the addition of freshly prepared 2-chloroacetamide (in 100 mM ammonium bicarbonate buffer -final concentration 50 mM) at RT for 30 min in the dark. Protein sample was added to 100 μg of prepared Sera-Mag™ SpeedBeads™ Carboxylate-Modified Magnetic Particles (Sigma-Aldrich). In brief, magnetic particles were prepared by forming a slurry with HPLC grade MilliQ water (beads:MilliQ ratio of 1:4), allowing particles to settle on a magnetic rack and removing supernatant. Rinsing was repeated three times. Acetonitrile (ACN, Sigma-Aldrich) was added to the protein sample and prepared particles to 77% final ACN concentration and incubated for 30 min at RT. Sample tubes were placed on a magnetic rack and incubated for a further 10 min at RT until supernatant ran clear. To remove the detergent from the magnetic particles, a 2-step 70% ethanol wash followed by 1-step ACN wash was performed until bubbles were no longer visible on the magnetic particles. Magnetic particles were reconstituted in trypsin:total protein digest ratio of 1:50 (from 0.2 μg μl À1 trypsin protease in acetic acid stock) overnight. The following day, sample tubes were added to the magnetic rack and peptide-containing supernatant was removed. Remaining peptides were removed from magnetic particles with 2% Dimethyl sulfoxide (Sigma-Aldrich) in MilliQ and incubated for 2 min at RT off the rack. Both supernatants were combined and acidified with 1 μl 10% formic acid (FA -reagent grade) prior to analysis on mass spectrometry.

Proteomics methods components
Desalted samples were separated on an UltiMate 3000 UHPLC system (Thermo Fischer Scientific, UK) and electro-sprayed directly into a Q-Exactive Mass Spectrometer (Thermo Fischer Scientific) through an EASY-Spray nano-electrospray ion source (Thermo Fischer Scientific). The peptides were trapped on a C18 Acclaim™ PepMap™ 100 pre-column (300 μm i.d. Â 5 mm, 100 Å, Thermo Fisher Scientific) using solvent A (0.1% FA in MilliQ) at a pressure of 500 bar. The peptides were separated on an in-house packed analytical column (75 μm i. d. Â 50 cm packed with ReproSil-Pur 120 C18-AQ, 1.9 μm, 120 Å, Dr. Maisch GmbH, Germany) using a linear gradient (length: 65 min, 15%-35% solvent B (0.1% FA in ACN), flow rate: 200 nl min À1 ). Data were acquired in a data-dependent mode. Full scan MS spectra were acquired in the Orbitrap (scan range 350-1500 m/z, resolution 70 000, AGC target 3e6, maximum injection time 50 ms À1 ). The 3 or 20 most intense peaks were selected for HCD fragmentation at 30% of normalised collision energy (resolution 17 500, AGC target 5e4, maximum injection time 120 ms À1 ) with first fixed mass at 180 m/z. A Top 3 method (three MS/MS collected per MS scan) was used for the label-free quantitative comparisons while a separate run was performed using a Top 20 method for qualitative analysis.

Data analysis pipeline
Identification of peptides was performed using MaxQuant computational proteomics platform (version 1.5.3.30) with the implemented Andromeda algorithm (Cox and Mann, 2008;Wi sniewski et al., 2013) by searching MS/MS spectra against a database consisting of the protein translation of the N. maritimus genome obtained from UniProt (https://www.uniprot.org/proteomes/UP000000792). Protein matches were filtered to an FDR of 1% at the peptide to spectrum match, peptide and protein levels. Measured intensities were converted to LFQ intensities using MaxLFQ algorithm (Cox et al., 2014) integrated within MaxQuant (using default parameters), using both unique and razor peptides for protein quantification. The MaxLFQ algorithm uses a non-linear optimisation model to normalise the peptide intensities thereby ensuring that profiles of LFQ intensities across samples accurately reflect the relative amounts of the proteins -taking into account all the peptide ratios for each protein measured in all pairwise comparisons of the different quantified samples and removing outliers (Cox et al., 2014). Protein LFQ intensities do not equate to protein abundance, and therefore intensities of the same protein across treatments are only robustly comparable -intensities of different proteins within treatments cannot be accurately compared. Mean LFQ intensities for Fe-limited and Fe-replete treatments were calculated from replicates and tested for significant differences using a student's t-test at an FDR-adjusted p-value of 0.05 in Perseus software (version 1.5.1.6). All differentially intense proteins are listed in the Supplementary Table 1. No statistical analysis was performed for proteins which were absent or below the limit of detection, and instead a presence-absence response was assigned.

Biological function groups and Fe containing proteins
Biological function categories (Supplementary Table 1) were assigned to proteins using the Gene Ontology (GO database). Where GO terms were not available, we used the Kyoto Encyclopaedia of Genes and Genomes (Kanehisa et al., 2007) reference pathways or InterPro to assign putative biological functions. In order to gain an insight into the effect of Fe limitation specifically on the Fe-containing proteome, we identified the Fe-containing proteins within the N. maritimus proteome. A full list of proteins encoded by the N. maritimus genome was downloaded from UniProt (proteome ID UP000000792) which was subsequently submitted to the PHYRE2 protein fold recognition server to identify homologues for known protein structures (Kelley et al., 2009). Significant matches (>95% confidence) were submitted to metalPDB (Putignano et al., 2018;Andreini et al., 2012) which models the 3D minimal functional sites surrounding metals. Metals were assigned if the search returned significant sequence similarity to metal-binding sites in other proteins (e-value <10 À5 ). This identification is underpinned by the assumption that the genomepredicted proteome of N. maritimus has been accurately annotated and that any archaeal proteins which are homologous to characterised bacterial Fe proteins also have similar Fe stoichiometry. By identifying homologues based on predicted 3D protein structures which are highly conserved, we have greater confidence that Fe stoichiometry will also be conserved.