Nitrogen and sulfur for phosphorus: Lipidome adaptation of anaerobic sulfate-reducing bacteria in phosphorus-deprived conditions

D. alkenivorans has previously been shown to synthesize diacylglycerols (DAGs), dietherglycerols (DEGs), and “mixed” acyl/ether glycerols [AEGs, containing one ester-bound fatty acid and one ether-bond alkyl chain (19, 39)]. This unusual lipid synthesis capability makes D. alkenivorans an excellent model to obtain a holistic picture of lipidome plasticity under variations in temperature, pH, salinity, and nutrient concentration (${[\mathrm{P}\mathrm{O}}_4^{3-}]$ and ${[\mathrm{N}\mathrm{H}}_4^{+}]$). We employed a previously developed analytical and computational workflow (31, 41) to analyze compositional changes in the lipidome by performing dedicated cultivation experiments in which one of the parameters was varied (Fig. 1; all performed in triplicate to account for natural biological variability). This approach allowed us to systematically collect and analyze high-resolution data-independent MS/MS spectra generated by ultra-high-pressure liquid chromatography coupled with high-resolution tandem mass spectrometry (UHPLC-HRMSⁿ) analysis of lipid extracts, enabling the thorough deconvolution of the lipidome. By using this untargeted approach (31), 390 individual lipids, belonging to six major classes, were found and putatively identified (Figs. 1B and 2A). Among these, isoprenoid quinones displayed the highest diversity, with a total of 47 individual species identified. Molecular network analysis (SI Appendix, Fig. S1) revealed that only a few menaquinones (MK) were consistently abundant across all culturing conditions, such as MK (8:7), while others were present in trace amounts and showed no significant variation. The remaining five lipid classes were represented by intact polar lipids (IPLs) all composed of 12 to 33 individual lipids (Fig. 2A). These latter classes included phosphatidylethanolamines (PE) and phosphatidylglycerols (PG), cardiolipins (CL), and glycolipids with a glucuronosyl (GlcA) headgroup. The core lipids of PE and PG consisted of DAGs, AEGs, and DEGs. GlcA, on the other hand, only contained AEGs (Fig. 3A) and DEGs (SI Appendix, Fig. S2) as its core lipids. The CL contained four alkyl/acyl chains: tetraetherglycerols (TetraEGs), triether/monoacyl glycerols (TriEGs), diether/diacyl glycerols (DiEGs), monoether/triacyl glycerols (MonoEGs), and tetra-acylglycerols (TetraAGs) (40). In addition, a class of sulfur-containing amino IPLs with a headgroup characterized by a m/z 205.064 (C₇H₁₃O₃N₂S⁺) fragment was observed specifically in the cultures with low ${[\mathrm{P}\mathrm{O}}_4^{3-}]$. Molecular network analysis revealed that these unknown sulfur-containing amino lipids differed from each other by the summed alkyl chain carbon atoms and number of double bonds (Fig. 3B). Based on their MS² spectra (see for details SI Appendix, Figs. S3 and S4 and Text), the unknown IPLs are composed of the common core lipids seen in the other IPLs (DAGs, AEGs, or DEGs) bound through glycerol to a polar headgroup (Fig. 3B and SI Appendix, Figs. S5 and S6) consisting of a butyramide linked to a cysteine moiety. Consequently, we tentatively assigned these sulfur-containing aminolipids as N-butyramide cysteine glycerols (BACys).

To investigate the basis of the lipid adaptation, we performed a detailed analysis of the molecular composition of the lipidome, focusing separately on polar headgroups, alkyl/acyl chains, and ether/ester combinations. The relative abundances of major IPL classes remained relatively stable in response to changes in ${[\mathrm{N}\mathrm{H}}_4^{+}]$, pH and salinity (Fig. 4 A–E). However, notable changes were observed for the phospholipids when the growth temperature increased from 30 (optimal) to 40 °C. The relative abundances of PG-DEGs and PE-DEGs decreased (P < 0.01, Student’s t tests), while those of PE-AEGs and PE-DAGs slightly increased (P < 0.01). Interestingly, the relative abundance of CL significantly increased from 2 to 19% with increasing growth temperature from 30 to 40 °C (Figs. 4C and 5A). Moreover, significant changes were also observed with the reduction in ${[\mathrm{P}\mathrm{O}}_4^{3-}]$. Under phosphorous starvation, the relative abundance of phospholipids (PG, PE, and CL, Fig. 4 A–C) decreased significantly from ca. 90% of total lipids to <10%, while that of the nonphospholipids GlcA and BACys showed a prominent increase (Figs. 4 D and E and 5A). Notably, the relative abundance of GlcA rose from <1 to ca. 20% of total lipids, and the newly formed sulfur- and nitrogen-containing lipids, BACys, increased from <1 to ca. 65% of the total lipids as ${[\mathrm{P}\mathrm{O}}_4^{3-}]$ was reduced (Fig. 5A).

In addition to the changing relative abundance of major lipids with increasing temperature and phosphorus starvation, the number of lipid species also showed marked changes. For example, the number of lipids with a relative abundance >0.5% rose from 18 to 39 when temperature increased from 30 to 40 °C (Fig. 5B). This increase is primarily attributed to the significant rise in the number of CL, especially CL-DiEGs and CL-TetraEGs. Additionally, under phosphorus limitation, 21 BACys and 9 GlcA lipid species emerged, each with an abundance >0.5% (Fig. 5B). Additionally, at high ${[\mathrm{P}\mathrm{O}}_4^{3-}]$, the proportion of P-containing lipids: N-containing lipids: S-containing lipids was approximately 84:15:1. However, under phosphorus limitation, this ratio shifted to 5:63:32. This aligns well with previous studies that showed cyanobacteria can produce 1 to 130-fold increases in sulfoquinovosyldiacylglycerols (SQDG) to substitute PG under phosphorus limitation (14, 42).

To gain a comprehensive view of the overall trend of total diglyceride phospholipids, CL, and non-phospholipids under the culturing conditions, we combined the different types of diglyceride phospholipids, CL, and nonphospholipids, respectively. A significant decrease was observed in the ratio of summed diglyceride phospholipids to summed CL with the elevation in temperature and phosphorous stress (Fig. 6A, P < 0.001). A significant decrease was observed in the ratios of summed phospholipids (including both diglyceride and CL) against summed nonphospholipids in response to phosphorous stress (P < 0.001, Fig. 6B).

Next, we examined the impact of the different culturing conditions on the number of carbon atoms and double bond equivalents (DBEs) of the alkyl/acyl chains in the cores of all IPLs (Fig. 6 D and E and SI Appendix, Fig. S7). IPLs with 0 DBEs were by far the most abundant, comprising 92 $\pm$ 3% of the total IPLs. The average DBEs of all IPLs were strongly affected by phosphate limitation (P < 0.001), and to a lesser extent by temperature (P = 0.004), pH (P = 0.009), and salinity (P = 0.002). In contrast, [${\mathrm{N}\mathrm{H}}_4^{+}]$ did not influence DBEs (Fig. 6D). The majority of alkyl/acyl chains contained between 28 and 32 carbon atoms (79 $\pm$ 7%) (Fig. 6E). Similar to DBEs, the average number of carbon atoms of the alkyl/acyl chains was highly conserved under varying pH and ${[\mathrm{N}\mathrm{H}}_4^{+}]$. However, it exhibited significant variability in response to changes in ${[\mathrm{P}\mathrm{O}}_4^{3-}]$ (P < 0.001) and, to a lesser extent, salinity (P = 0.002), and temperature (P = 0.015) (Fig. 6E).

To determine the role of ether/ester bond changes in lipidome adaptation, we further investigated the major lipid classes with varying ether/ester combinations (Figs. 4 and 6C). Within PG, significant changes were observed in those with DEGs as core lipids under ${[\mathrm{P}\mathrm{O}}_4^{3-}]$ limitation and temperature elevation but not with AEGs and DAGs (Fig. 4A). For PE, a significant decrease was observed in DEGs under decreased $[{\mathrm{P}\mathrm{O}}_4^{3-}]$ and temperature elevation, while the relative abundance of AEGs and DAGs increased when temperature was elevated (Fig. 4B). Similar trends were also observed in CL with different combinations of ether/ester bonds when facing a temperature increase (Fig. 4C). For the non-phospholipids GlcA and BACys, DEGs and AEGs were found to significantly increase at phosphate limitation, but not under other conditions. When we examined the ratio of ether/ester chains among all lipid combined, we found that it was less significant compared to the impact of polar headgroups when temperature was elevated or under $[{\mathrm{P}\mathrm{O}}_4^{3-}]$ limitation (Fig. 6C, P < 0.05). This observation aligns with previous research that utilized a hydrolysis-based method to analyze core lipids of D. alkenivorans, where the proportion of alkylglycerol ether lipids was found to remain relatively stable regardless of the growth temperature (39). Furthermore, as we examined the overarching pattern of ester bond lipids versus ether bond lipids, we observed significant shifts from ester bond lipids to ether bond lipids when transitioning from suboptimal to optimal salinity conditions (Fig. 6C, P < 0.001).

In order to assess the general variation in the lipidome of D. alkenivorans strain across all experimental conditions and replicates, we conducted a principal component analysis (PCA) on the abundances of lipid species, based on peak intensity (Fig. 2B). The first principal component (PC1) explained 30.1% of the variance of all the lipids and the second principal component (PC2) explained 20.6%. The majority of the applied culture conditions clustered closely together, with the exception of the highest temperature (40 °C) and the two lowest ${[\mathrm{P}\mathrm{O}}_4^{3-}]$ concentrations (0.0015 $\mathrm{a}\mathrm{n}\mathrm{d}$ 0.0005 mM). The noticeable separation of the lipidome produced at the highest temperature and the most limiting $[{\mathrm{P}\mathrm{O}}_4^{3-}]$ from those biosynthesized at the other culturing conditions illustrates that these two parameters have the most substantial influence on the D. alkenivorans lipidome (Fig. 2B).

To further evaluate the lipidome plasticity under various physicochemical perturbations and nutrient stress, we applied an information theory framework (43, 44). This framework allowed us to measure both the lipidome diversity ($H_j$ index) and specialization (${\delta }_j$ index) based on the Shannon entropy of the lipidomic frequency distribution. The lipidome diversity ($H_j$ index) represents the extent of variation in its composition at a specific culturing condition. A high value of the $H_j$ diversity index indicates either a larger proportion of lipids detected or a more evenly distributed abundance of lipids under that condition. The lipidome specialization ${\delta }_j$ index reflects the average degree of uniqueness of individual lipids and is thus an indicator for the average lipid specificity for a particular culturing condition. By cross-plotting the specialization and diversity of the D. alkenivorans lipidome, we observed a similar grouping for most culturing conditions, except for those involving the highest temperature employed (40 °C) and the lowest (limiting) ${[\mathrm{P}\mathrm{O}}_4^{3-}]$ (1.5 $\mathrm{a}\mathrm{n}\mathrm{d}$ 0.5 mM$;$ Fig. 2C). Interestingly, cultivation under phosphate limitation displayed the second-most diverse lipidome profile and the highest specialization among all cultivation conditions. High growth temperature (40 °C) exhibited the highest diverse lipidome profile of all culture conditions, while still maintaining a relatively high level of specialization, especially when compared to that of 25 and 30 °C. We subsequently calculated the lipid species specificity ($S_i$ index) for all the lipid species across the various culturing conditions (Fig. 2D). The $S_i$ specificity index allows us to identify which lipids exhibit the highest degree of specificity or sensitivity toward changes in the culturing conditions. Among the lipid species analyzed, a large number of lipids had small values of $S_i$ ($S_i$ < 3.0, 77% of the total), indicating they were either present in low concentrations or uniformly distributed across all the culturing conditions. On the other hand, a few specific lipids, such as PG-DEG (31:0, a summed chain length of 31 carbons with 0 double bond), CL-TetraEG (56:1), CL-TriEG (58:2), GlcA-DEG (31:0), GlcA-AEG (31:0), and BACys-DAG (32:0), demonstrated a higher level of specificity toward particular culturing conditions.

We also applied a hierarchical clustering heatmap to analyze the distribution of major lipid classes across all the culturing conditions of D. alkenivorans (Fig. 2E). This visualization method allowed to assess the similarity of lipid class profiles and rapidly examine major lipid class changes under different culturing conditions. The major lipid classes were grouped into four distinct clusters. The first cluster consisted of all the nonphospholipids (GlcA and the novel BACys). The second cluster comprised phospholipids, specifically PG and PE. The third cluster included CLs with 1-3 ether chains, while the fourth cluster consisted of CL-TetraEGs and isoprenoid quinones.

In order to assess the impact of perturbations in culturing conditions on the remodeling of the membrane lipidome, we quantified the cumulative variability of all lipid species within each range of growth conditions (Fig. 2F). The variability of lipid species was determined by calculating the difference in mean abundance between the optimal growth condition and the sub- and supraoptimal conditions (33). The total lipidomic variability was then obtained by aggregating the absolute value of the changes in abundance for all individual lipid species. Among the range of conditions considered in this study, the availability of phosphate exhibited the most significant effect on lipidome remodeling, accounting for 85% of the total lipidomic variability under P-stress conditions. Additionally, supraoptimal temperature was found to have a substantial impact (~25%), resulting in more than a twofold increase in variability compared to the remaining conditions.

Increasing temperature leads to a higher diversity of lipids in D. alkenivorans, exemplified by the increased relative abundance of CL (Fig. 4C); they increased from 2 to 20% of total lipids when temperature was increased from 30 to 40 °C (Fig. 4C). In addition, a markedly increased number of individual CL species with different ether/ester chain combinations was noted at 40 °C (Fig. 5B). The particularly anionic CL have been proposed to be beneficial in interactions with divalent cations (47, 48). For example, under elevated Mg²⁺ concentrations, the negatively charged CL lipid assists cells in counteracting curvature stress and protects the cell’s interior from an elevated level of toxic Na⁺ (35), but salinity had no large effect On the CL abundance and composition (Fig. 5). CL may also play a compensatory role by counteracting the effects of enhanced phospholipid saturation and is essential for cristae development in environments with low oxygen levels (49). Under high temperature, it is possible that the upregulation in CL production may have assisted D. alkenivorans control curvature stress and optimally balance membrane permeability. The finding of a dominant effect of temperature on lipidome remodeling is similar to that observed in Methylobacterium extorquens (33), suggesting a commonality in lipidome adaptation between these two diverse bacteria.

Previous studies have shown that increasing the number of DBEs per lipid increases lipid mobility and hence counteracts low temperature-induced membrane rigidity (8, 50). In this study, we did not observe a clear relationship between DBEs and growth temperature (Fig. 6). In fact, the lipidome has a highly “saturated” nature (Fig. 6D; DBE average always below 0.04), fully in line with previous work on its core lipid composition (39). Similarly, dominant saturated lipids have also been detected in cultured bacteria from the order Rubrobacterales (16) and Thermodesulfobacteriales (51). This characteristic may serve as a common feature for the membrane lipid building blocks of anaerobic bacteria. Interestingly, IPLs that are highly saturated, particularly DEGs with 0 to 1 DBEs, predominate in deep waters of the euxinic Black Sea (31) and the oxygen minimum zone of the eastern tropical North Pacific Ocean (30).

The chain length adaptation of bacterial core lipids has also been reported at elevated temperatures (8). However, similar to DBEs, a minimal influence of temperature on lipid chain length of IPLs was observed in D. alkenivorans (Fig. 6E). A study evaluating the impact of temperature on the hydrolyzed core lipid composition of sulfate-reducing bacteria including D. alkenivorans, Desulfatibacillum aliphaticivorans, and Thermodesulfobacterium commune also indicated a minimal influence of temperature on lipid chain length (39). The potential influence of other parameters, such as pressure, on chain length and DBE remodels still requires further investigation.

The thermal stability of ether bonds compared to ester bonds, along with the frequent occurrence of nonisoprenoid ether glycerol lipids in thermophilic bacterial strains, has often been regarded as another adaptation to withstand high temperatures (5). For instance, Acididesulfobacillus, which thrives in extreme conditions at temperatures exceeding 80 °C, contains significant amounts of ether lipids (52, 53). However, the production of ether lipids is not limited to thermophilic bacteria (19, 37, 54), and not all thermophilic bacteria synthesize ether lipids (17). In an analysis of the core lipid (ether/ester) content of two mesophilic (D. alkenivorans and its close relative D. aliphaticivorans) and one thermophilic (T. commune) sulfate-reducing bacteria, Vinçon-Laugier et al. (39) observed that the ether/ester lipid content remained remarkably stable regardless of the growth temperature. The major adaptive mechanisms involved modifications in the degree and position of methyl branching within the lipid chains. The present study of the complete intact lipidome of D. alkenivorans confirms that the ratio of summed ether/ester lipid content remains stable with temperature changes (Fig. 6C). However, significant changes in the ether/ester lipid content were observed within a specific lipid class, i.e., CL. CL with more ether chains, such as CL-TetraEGs and CL-DiEGs, increased from less than 2 to ~7% of the total lipids when the temperature was increased from 30 to 40 °C (Fig. 4C). In contrast, CL with four ester chains, CL-TetraAGs, remained relatively stable (<0.5%). This suggests that the adaptation of ether/ester chains has occurred in specific lipid classes to help maintain membrane integrity and functionality. Again, this is in line with previous findings on membrane adaptation of M. extorquens, which showed headgroup-specific effects of acyl chain variability on its membrane properties (33).

When subjected to ${[\mathrm{P}\mathrm{O}}_4^{3-}]$ limitation, D. alkenivorans demonstrates a remarkable metabolic shift characterized by the almost complete replacement of the phospholipids PE, PG, and CL into nonphospholipids, BACys and GlcA (Fig. 5). This adaptive response leads to an increase of both diversity and specificity of the lipidome (Fig. 2). Apparently, D. alkenivorans possesses a remarkable genetic capacity to switch to a completely new set of membrane lipids at times of phosphate scarcity.

Previous research has indicated that certain bacteria, when grown under phosphorous-limited conditions, synthesize non-phospholipids such as betaine lipids (55, 56), ornithine lipids (57, 58), and glycolipids (13, 58) and use them as membrane lipids. The presence of the GlcA polar headgroup, as observed in this study, has also been reported in the free-living marine bacteria of the order Pelagibacterales (SAR11) (13), in microalgae (59) and in higher plants (60), facing phosphorus starvation. The notable difference observed here lies in the ability of D. alkenivorans to incorporate GlcA with various core lipids, ranging from DAGs to AEGs and DEGs. Additionally, this study reports the identification of BACys lipids in bacteria, serving as functional lipids under phosphorus stress. A “sulfur-for-phosphorus” strategy was reported earlier for picocyanobacteria which synthesize SQDG (14, 42), membrane lipids that contain sulfur and sugar, to replace phospholipids in phosphorous-deprived environments. The polar headgroup of BACys lipids distinguishes itself from those of the GlcA and SQDG lipids, in the fact that they are sulfur-containing aminolipids. This indicates that microorganisms can respond to environmental phosphorus shortages through diverse combinations of polar head groups rather than by relying solely on fixed types of non-phospholipids. A class of sulfur-containing aminolipids was found to occur widespread in marine roseobacters (61). These lipids, which have been confirmed as cysteinolides recently (62), play a crucial role in biofilm formation by these bacteria. However, whether they have an impact on adaptation to phosphorus limitation still requires further investigation.

When facing elevated temperature or ${[\mathrm{P}\mathrm{O}}_4^{3-}]$ limitation, the homeoviscous adaptation of D. alkenivorans is not an isolated event that focuses exclusively on the polar headgroups. Our study indicates that “homoviscous adaptation” in D. alkenivorans is finely tuned through the coordinated regulation of polar headgroup composition, the ratio of ether/ester bonds within specific lipid classes, and the degree and position of methyl branching of the alkyl and acyl chains (38), with a much less dominant role for the adjustment of the alkyl/acyl chain length and DBEs. Rather than just individual changes in one factor, the intricate interplay between different lipid characteristics in achieving optimal adaptive responses of the lipidome seems to be the key.

The exploration of how microbes perceive changes in their biophysical characteristics and adapt through lipidome remodeling has gained significant attention (33). Understanding the underlying mechanisms of membrane lipids’ responsiveness would answer a fundamental question regarding the essential requirements for a system to demonstrate lifelike adaptability. Moreover, it could offer valuable insights for designing synthetic membranes from scratch, serving as a blueprint for future engineering endeavors. Our study provides valuable insights into the molecular mechanisms of ether/ester bond lipids, encompassing various polar headgroups that play a crucial role in the functionality of cellular membranes. By examining these lipid compositions, we gain a deeper understanding of the intricate processes that govern membrane stability and function in sulfate-reducing bacteria. Delving deeper into the mechanisms of the ether/ester bond lipid producers holds the potential to uncover the intricate interplay between membrane lipid structure and function.

D. alkenivorans PF2803^T is a mesophilic bacterium belonging to the family Desulfatibacillaceae and is known for its ability to degrade n-alkenes (39). It was isolated from oil-polluted sediments in Fos Harbour, France (38). For cultivation, D. alkenivorans was grown on synthetic sulfate-reducing medium supplemented with octanoate (>99%, Sigma-Aldrich) as the sole carbon and energy source. Cultures (100 mL) were performed in batches for different conditions and were incubated under anaerobic conditions without shaking. The optimal conditions for growth are temperature 30 °C, pH 6.8, [NaCl] 10 g L⁻¹, ${[\mathrm{P}\mathrm{O}}_4^{3-}]$ 1.5 mM, and ${[\mathrm{N}\mathrm{H}}_4^{+}]$ 5.6 mM. KH₂PO₄ and NH₄Cl were added to the culture medium to reach the different concentrations for phosphorus and nitrogen. To investigate the effects of different parameters on the lipidome adaptation of D. alkenivorans, triplicate cultures were established by varying one parameter affecting growth, while the other parameters remained at their optimal values. These variations in growth parameters applied were temperature (25, 30, and 40 °C), pH (6.4, 6.8, and 7.8), and NaCl concentration (3, 10, 25, and 60 g L⁻¹), ${\mathrm{P}\mathrm{O}}_4^{3-}$ (0.0005, 0.0015, 0.015, and 1.5 mM), and ${\mathrm{N}\mathrm{H}}_4^{+}$ (0.056, 0.56, and 5.6 mM) concentrations. The cultures were incubated for a period of 1 to 3 wk, and growth was monitored for each condition by measuring the optical density to ensure that all cultures reached the end of the exponential phase before collection (Dataset S2). The optical density varied between 0.1 and 0.3 according to different culturing conditions, corresponding to approximately 2.5 to 10 × 10⁷ cells/mL, respectively (Dataset S2). The cultures were subsequently filtered on glass microfiber filters (GF/F; Whatman) and stored at −80 °C until extraction of their lipidome.

A detailed description of lipid extraction and measurement can be found in ref. 41. Briefly, the filtered cells were extracted using a modified Bligh–Dyer extraction method (51). Medium blanks and extraction blanks were used to subtract background compounds and contaminants. The samples and blanks were extracted ultrasonically for 10 min, twice in a mixture of methanol, dichloromethane (DCM), and ${\mathrm{P}\mathrm{O}}_4^{3-}$ buffer (2:1:0.8, v/v/v) and twice with a mixture of methanol, DCM, and aqueous trichloroacetic acid solution pH 3 (2:1:0.8, v/v/v). The organic phase was separated by adding more DCM and buffer to achieve a final solvent ratio of 1:1:0.9 (v/v/v). This organic phase was then subjected to three more extractions using DCM and dried using a stream of N₂ gas. The resulting extract was reconstituted using a mixture of methanol and DCM (in a ratio of 9:1, v/v) and filtered through 0.45 μm regenerated cellulose syringe filters (4 mm diameter; Grace Alltech). The filtered extracts were subsequently analyzed using an Agilent 1290 Infinity I UHPLC system coupled to a Q Exactive Orbitrap MS (Thermo Fisher Scientific, Waltham, MA). The generated output data files from the ultra-high-pressure liquid chromatography coupled with high-resolution tandem mass spectrometry (UHPLC-HRMS²) analyses were further processed using MZmine software (63) to extract MS¹ and MS² spectra as well as quantify peaks. This processing involved several steps, such as mass peak detection, chromatogram building, deconvolution, isotope grouping, feature alignment, and gap filling (https://ccms-ucsd.github.io/GNPSDocumentation).

The MS/MS spectra dataset was analyzed using the Feature-Based Molecular Networking tool available on the Global Natural Product Social Molecular Networking (GNPS) platform (64, 65). MS2LDA (66), MolnetenHance r (67), SIRIUS 5 (68), and Spec2Vec (69) were used to assist the searching and identification of lipids. Molecular networking is a data analysis methodology utilized in untargeted metabolomics studies based on MS/MS analysis. It organizes MS/MS spectra into a network-like map, where molecules with similar spectral patterns are grouped together, indicating their structural similarity. Vector similarities were calculated by comparing pairs of spectra based on at least six matching fragment ions (peaks). This comparison takes into account the relative intensities of the fragment ions as well as the difference in precursor m/z values between the spectra (65, 70). The resulting molecular network is generated using MATLAB scripts, where each spectrum is allowed to connect to its top K scoring matches (typically up to 10 connections per node). Edges between spectra are retained if they are among the top K matches for both spectra and if the vector similarity score, represented as a cosine value, exceeds a user-defined threshold. In this study, a cosine value of 0.6 was used, with a cosine value of 1.0 indicating identical spectra.

In cases where an ion component exhibited both [M+H]⁺ and [M+NH₄]⁺ ions, the abundance of the component was calculated as the sum of the abundances of both ions. A minimum of six shared fragment ions was required between pairs of related MS/MS spectra for them to be connected by an edge in the molecular network. Each node in the network was connected to a maximum of 10 analogs. In addition, consensus spectra were searched against the GNPS spectral library, allowing for a maximum analog mass difference of m/z 500. Maximum size of nodes allowed in a single connected subnetwork was 100. In cases where a significant number of related lipids were present in the dataset (greater than 100), they were separated into different subnetworks.

The resulting molecular networks were visualized using Cytoscape version 3.9.1 (71, 72). It is important to note that since many of the lipids detected in this study have not been previously described, authentic standards for absolute quantification were unavailable. The lipids were corrected for sample recovery with a 1,2-dipalmitoyl-sn-glycero-3-O-4′-[N,N,N-trimethyl(d9)]-homoserine (DGTS-d9) internal standard and then examined based on their calibrated peak area responses. A threshold of peak intensity of 1 × 10⁶ was used to filter out lipids with an abundance below 0.01 mol% of total lipids. Consequently, the relative peak areas do not necessarily indicate the actual relative abundance of different lipids. Nevertheless, this method allowed comparison between different cultivation conditions (73).

Given our extraction and analytical methods, and relying on the annotation provided by the GNPS library and putatively identified by ourselves, it is anticipated that the majority of ion components within the molecular network are lipids. Therefore, the term “lipidome” was used in relevant parts of the discussion where ion components were addressed.

For PCA (Fig. 2B), the abundance data of lipid species were initially transformed using the Hellinger distance method (74) to mitigate bias arising from 0 values. These data were then processed and visualized using R software, version 4.1.2. Hierarchical clustering (Fig. 2E) was performed using the “ggplot2” and “pheatmap” packages in R, version 4.3.2.

Information theory framework (Fig. 2 C and D) was used to define lipidome diversity and specialization and individual lipid species specificity (43, 44, 75). Lipids were characterized by their own unique MS/MS spectrum and relative frequency of occurrence among all the culturing samples. Lipidome diversity, the $H_j$ index, was calculated using Shannon entropy of MS/MS (lipid species) frequency distribution derived from the abundance of MS/MS precursors by the following equation

where P_ij correspond to relative frequency of the ith MS/MS (i = 1, 2, …, m) in the jth sample (j = 1, 2, …, t), to illustrate how abundant a specific MS/MS spectrum is relative to all others.

The average frequency of the ith MS/MS among samples was calculated as

Individual lipid species specificity, the $S_i$ index, was defined as the identity of a given MS/MS regarding frequencies among all the samples. The lipid species specificity was calculated as

Individual lipid species specificity of specific sample was defined as $S_{\mathit{ij}}$ index.

The lipidome specialization ${\delta }_j$ index was measured for each jth sample, the average of the MS/MS specificities using the following formula

S.D., V.G., E.C.H., N.J.B., C.C.-L., and J.S.S.D. designed research; S.D. performed research; V.G., E.C.H., and C.C.-L. contributed new reagents/analytic tools; S.D., E.C.H., and N.J.B. analyzed data; J.S.S.D. supervised the research; and S.D. wrote the paper with contributions from all authors.

The authors declare no competing interest.