Modeling lipid homeostasis using stable isotope tracing and flux analysis

SUMMARY

Lipids enable compartmentation and coordinate membrane-localized signaling events in cells, and dysregulation of lipid metabolism is linked to many disease states. However, limited tools are available for quantifying metabolic fluxes across the lipidome. To measure fluxes encompassing lipid homeostasis in cells and tissue slices, we apply stable isotope tracing, liquid chromatography-high-resolution mass spectrometry, and network-based isotopologue modeling to non-small cell lung cancer (NSCLC) models. Lipid metabolic flux analysis (Lipid-MFA) enables quantitation of fatty acid synthesis, elongation, headgroup assembly, and salvage reactions within virtually any biological system. Using Lipid-MFA, we observed decreased fatty acid synthase and very long-chain fatty acid (VLCFA) elongation fluxes, along with increased sphingolipid recycling, in p53-deficient versus liver kinase B1 (LKB1)-deficient NSCLC tumors using precision-cut lung slice culture. We also apply Lipid-MFA to demonstrate the unique trafficking of ceramides with distinct n-acyl chain lengths, highlighting the utility of this approach in elucidating molecular mechanisms in lipid homeostasis.

In brief

Wessendorf-Rodriguez et al. develop a pipeline for applying metabolic flux analysis to quantify lipid homeostasis. Lipid-MFA elucidates differences in synthesis and salvage fluxes in precision-cut lung slices harboring tumors with common driver mutations of lung adenocarcinoma. Application of Lipid-MFA reveals distinct trafficking of ceramide species into the sphingolipidome.

Graphical Abstract

INTRODUCTION

Lipids are the most abundant and diverse group of metabolites in cells and tissues. They facilitate numerous biological processes that include membrane compartmentalization, bioenergetics, signal transduction, protein function, and cell-cell interactions.^1–6 Membrane phospholipids and sphingolipids (SLs) are highly metabolized and processed in the Lands cycle and endolysosomal system, respectively.⁷ They are generally synthesized from a head group and acyl chains that are further modified and catabolized by diverse enzymes spanning numerous organelles. The physical properties and function of each lipid, therefore, depend on the backbone, head group chemistry, and acyl-chain structure (chain length, desaturation, hydroxylation). For example, ceramides (Cer) containing long-chain fatty acids versus very long-chain fatty acids (VLCFAs) have distinct functions and disease relevance.^8–11 The ubiquity of lipids in cellular biology places their metabolic regulation as a key factor in health and disease. Indeed, biological processes such as autophagy and vesicle secretion are intimately tied to lipid metabolism,^12–14 suggesting that measurements of lipid metabolic flux can be informative for cell biology.

Given the bioenergetic costs of lipid synthesis as well as their intermediate molecular size relative to small molecule metabolites and peptides, “compound” or “complex” lipids are synthesized de novo and extensively recycled by organisms.^14–16 Exogenous lipids are also consumed, metabolized, and incorporated into cellular lipid pools such that membrane biochemistry can also reflect the cellular microenvironment or diet.^17–21 This inherent plasticity confers some resilience when cells have limited resources but also allows environmental lipids to influence membrane biology.^22,23 The vast, interconnected network of enzymes responsible for headgroup and acyl-chain synthesis, elongation, lipid assembly, and catabolism poses a challenge to the identification of disease mechanisms associated with lipid homeostasis. However, recent advances in high-resolution mass spectrometry and computational metabolic flux analysis (MFA) are now enabling more reliable measurement of these processes.

¹³C MFA is the most advanced and reliable approach for quantifying biochemical fluxes in living organisms.^24–28 The metabolism of stable isotope-labeled nutrients generates complex labeling patterns or “mass isotopomer distributions” (MIDs) that are measured by mass spectrometry or other approaches. In turn, these data are analyzed in the context of molecular networks and thermodynamic constraints to estimate fluxes across the network. ¹³C MFA applications have typically addressed questions in central carbon metabolism and biopolymer synthesis,^25,27,29 with more targeted approaches used to examine lipid metabolism by examining distinct mass isotopomers.³⁰ As lipid metabolism becomes increasingly relevant in biology,^31–33 new tools are needed to extend MFA beyond these applications. In addition to multi-step synthesis pathways, lipids are highly catabolized to facilitate the recycling and re-use of acyl chains and headgroups in the lysosome, peroxisome, endoplasmic reticulum (ER), Golgi, and mitochondria. MFA models comprised of compartmentalized and/or reversible reactions within a comprehensive biochemical network are required to resolve such metabolic exchange fluxes,³⁴ as well as synthesis in one organelle (ER) and breakdown in another (lysosome). Although commonly overlooked in favor of “rate limiting” or unidirectional reactions within biochemistry texts, high-exchange-flux reactions promote the resilience of such systems to biochemical stressors and fluctuations in nutrient availability. For example, oxidative and reductive fluxes through isocitrate dehydrogenases support the use of different substrates for acetyl-coenzyme A (CoA) provision in normoxia and hypoxia.^35–37 The malate-aspartate shuttle and pyruvate cycling similarly allow cells to negotiate anaplerosis and electron transfer within mitochondria, highlighting the importance of these metabolic exchange nodes. In the context of SL metabolism, the endolysosomal network facilitates recycling of sphingoid bases and represents an important node in bioactive lipid homeostasis. Approaches that enable researchers to quantify the balance between membrane lipid synthesis and recycling in cells and tissues will be critical to elucidate disease mechanisms in the coming decades.

Here, we develop and implement Lipid-MFA to elucidate reaction fluxes involved in lipid homeostasis, applicable to a myriad of biologically relevant models. This analytical and modeling framework leverages high-resolution mass spectrometry of intact lipids, metabolic networks encompassing lipid synthesis and recycling, and established MFA software³⁸ to resolve fluxes across the lipidome. We demonstrate the responsiveness of distinct lipogenic enzymes to serum availability. We further validate this approach in cultured cells and in precision-cut lung slice (PCLS) cultures using Kras mutant non-small cell lung cancer (NSCLC) models that are deficient in either tumor protein 53 (Trp53) or liver kinase B1 (Lkb1). Both tumor suppressors influence lipid synthesis and recycling pathways,^31,39,40 and Lipid-MFA enables resolution of flux changes in these models. This framework, therefore, enables reliable quantitation of metabolic fluxes across the lipidome in virtually any biological system, including homogeneous cells in culture and PCLS cultures cultured with stable isotope tracers, providing a versatile tool for researchers to quantify fluxes through numerous enzymes in lipid metabolism.

DESIGN

MFA modeling requires a stoichiometric network and mass isotopomer data to estimate relative fluxes and confidence intervals representing the metabolism of a biological system.²⁵ For initial testing and implementation of Lipid-MFA, we developed a pipeline for generating metabolic networks that encompass both glycerolipid and SL metabolism (Figure 1A). To this end, we cultured the A549 NSCLC cell line in the absence of tracers and measured quantifiable lipid species of interest that were free of co-eluting, isobaric metabolites and displayed MIDs that aligned with calculated theoretical natural abundances (Figure S1A). Using ultra-high pressure liquid chromatography coupled to high-resolution mass spectrometry, we identified a broad range of species across several lipid classes with confirmed MS2 fragments that were suitable for modeling, focusing on abundant species such as phosphatidylcholines (PCs), phosphatidylethanolamines (PEs), phosphatidylserines (PSs), ether lipids (pPEs), triglycerides (TGs) and SLs (Table S1). For simplicity, we focused on de novo-synthesized lipids with high enrichment and only included species containing poly-unsaturated fatty acids (PUFAs) when isobaric species could be chromatographically separated and validated by MS2 fragmentation. However, these lipid networks and data inputs can be readily tailored to incorporate different molecular species, reactions, tracers (e.g., isotopically labeled choline or PUFAs), and measurements depending on the pathway of interest.

Figure 1. Design, implementation, and assumptions.

(A) Workflow for the implementation of Lipid-MFA. (Created with Biorender.com).

(B) Mass isotopomer distribution (MID) of PC 32:0 highlighting characteristic labeling patterns from FASN and GPAT flux from a [U-¹³C₆]glucose tracing experiment (n = 3 per group).

(C) MID of Cer 18:1;O2/24:0 highlighting characteristic labeling patterns from FASN and ELOVL1/6 flux from [U-¹³C₆]glucose tracing experiment (n = 3 per group).

(D) MID of GM2 18:1;O2/24:0 from a [U-¹³C₆]glucose tracing experiment (n = 3 per group).

(E) Depiction of GM2 18:1;O2/24:0 with colored carbon atoms that represent potential sources of labeling that contribute to the MID in (D) (n = 3 per group).

(F) Fractional labeling (M3) of G3P from [U-¹³C₆]glucose over 24 h from A549 cells cultured in 10% FBS (n = 3 per group).

Next, we conducted tracer studies using [U-¹³C₆]glucose, which yielded complex labeling patterns after 24 h that provide information on various reactions within the pathway. MIDs for PC and Cer generated in A549 cells are depicted in Figures 1B and 1C, with key reactions and mass isotopomers highlighted for reference. Complex glycolipids, such as the ganglioside GM2 18:1;O2/24:0, are synthesized by many reactions and thus generate more complicated distributions (Figures 1D and 1E). While the mass isotopomer patterns are visually distinct for some lipids, such as PCs (Figure 1B) and Cer (Figure 1C), as the number of building blocks increases, these data become increasingly more complex for larger lipid species (Figure 1E). By defining the atom transitions within a reaction network encompassing lipid synthesis and catabolism, the MIDs noted above can be used to estimate fluxes within the system using INCA,^{38 13}C-FLUX2,⁴¹ OpenFLUX-2,⁴² or other software.⁴³ While we observed robust labeling in abundant lipids when culturing with [U-¹³C₆]glucose, limited enrichment was observed in free sphingosine (SPB 18:1;O2, or SO) and sphinganine (SPB 18:0;O2, or SA) in whole cell lysates (Figures S1C and S1D). These species are derived from catabolism of unlabeled membrane SLs within the lysosome and cytosol.^44–46 Thus, to observe robust flux through the serine palmitoyltransferase (SPT) complex, which synthesizes a long-chain base (LCB) from serine and palmitoyl-CoA, we cultured A549 cells in the presence of [U-¹³C₃]serine. This tracer readily labels dihydroceramide (DHCer), Cer, and sphingomyelin pools in cell lysates, while minimally labeling fatty acids and lipids aside from PS and PE (Figures S1D and S1E). The inclusion of parallel serine tracing data facilitated measurement of LCB de novo synthesis versus salvage fluxes, as detailed below.⁸

Next, we incorporated these data into a mechanistic ¹³C MFA model using INCA,³⁸ a flux analysis tool that allows the user to input reaction networks, atom transitions, tracers, and data to estimate fluxes and confidence intervals for a given experiment. MFA software packages such as INCA establish a stoichiometric matrix and iteratively simulate labeling data to identify the flux network that best fits experimentally determined results. Confidence intervals are subsequently determined via sensitivity analysis to translate labeling data (and/or pool sizes) to flux measurements. In the case of Lipid-MFA, isotopic labeling of precursor pools (e.g., glycerol 3-phosphate [G3P], acetyl-CoA, serine) is confirmed or assumed to be at pseudo-steady state (Figures 1F and S1G). While enrichment of lipids in biomass does not reach isotopic steady state, precursor labeling and the time-dependent fractional synthesis, g(t), of each lipid are calculated.^29,47 These assumptions and solutions are routinely used for in vivo and in vitro flux measurements such as isotopomer spectral analysis (ISA) or mass isotopomer distribution analysis (MIDA).^47,48 The reaction networks and flux estimates used in this manuscript are presented as supplemental tables referenced throughout the text and can be found in Data S3. Separate compartments of acetyl-CoA for the cytosol and ER were defined to deconvolute fatty acid synthase (FASN) from elongation (ELOVL) flux, accounting for the differential compartmentalization of FASN and the ELOVL enzymes. Overall, by incorporating mass isotopomer data from diverse lipid classes, we could resolve key fluxes in membrane lipid biosynthesis, recycling, and interconversions between lipid classes.

For all MFA models, we either (1) fix the final measured metabolite pool to 100 pmol without constraining intermediate pools or (2) use measured pool sizes to quantify picomolar flux values necessary to maintain the observed membrane lipid pools. Finally, parameter continuation is used to determine the 99% confidence interval flux bounds for each reaction. Below, we apply Lipid-MFA to quantify fluxes across the lipidome within NSCLC models, exploiting distinct tumor genetics to highlight the ability of this tool for quantifying lipid synthesis and salvage fluxes within the ER, Golgi, and lysosomal network.

RESULTS

Lipid availability strongly impacts acetyl-CoA and FASN flux in tissue culture

First, we determined whether Lipid-MFA could resolve flux changes driven by culturing cells in media with varying lipid availability. We performed parallel [U-¹³C₆]glucose and [U-¹³C₃]serine tracing experiments for 24 h in the presence of either 1%, 10%, or 20% fetal bovine serum (FBS). To determine which SL and glycerolipids to include in the model, we compared the enrichment and abundances of lipid species in four main lipid classes: SLs, PCs, PEs, and TGs (Figures S2A–S2D; Tables S2 and S3). This step allowed us to focus on highly metabolized lipids over those species simply taken up from the media. While SM pools were only mildly impacted (Figure S2A), the most affected lipid pools were selected PUFA-containing phospholipids and TGs, which were decreased in 1% FBS and increased in 20% FBS (Figures S2B–S2D). These changes are expected, given the presence of essential fatty acid-containing lipids in FBS. With respect to isotope enrichment, increasing the media serum resulted in lower isotope enrichment across most (but not all) sphingomyelin species (Figures 2A and 2B). For analysis of SL metabolism, we built a reaction network that encompassed the de novo biosynthesis and recycling of SM 42:1;O2 and incorporated mass isotopomer data for SPB 18:1;O2, Cer 18:0;O2/24:0, Cer 18:1;O2/24:0, and SM 42:1;O2 generated from parallel tracer experiments using [U-¹³C₆]glucose and [U-¹³C₃]serine. These metabolites allowed us to measure flux through FASN, ELOVLs, SPT, ceramide synthase 2 (CERS2) (the predominant CERS isoform expressed in A549 cells^49,50), delta-4- desaturase sphingolipid 1 (DEGS1), sphingomyelin synthase 1 (SGSM1), sphingomyelin phosphodiesterase 1 (SMPD1), and N-acylsphingosine amidohydrolase 1 (ASAH1). The network, labeling data, and flux measurements for this model are reported in Tables S4–S6.

Figure 2. Lipid availability strongly impacts acetyl-CoA and FASN flux in tissue culture.

(A) Mole Percent Enrichment (MPE) of sphingomyelin species from [U-¹³C₆]glucose in A549 cells cultured in 1%, 10%, or 20% FBS (n = 3 per group).

(B) Percent M2 labeling of sphingomyelin from [U-¹³C₃]serine (n = 3 per group).

(C) Fractional contribution of [U-¹³C₆]glucose to lipogenic acetyl-CoA.

(D) Fraction of newly synthesized fatty acids used to maintain SM 42:1;O2 pools.

(E) Maximal exponential growth rate of A549 cells cultured in 1%, 10%, and 20% FBS.

(A) and (B) show MPE data as mean ± standard error of mean (SEM) and were analyzed using two-way ANOVA. **p < 0.01, ***p < 0.001, ****p <0.0001. Data shown in (C) and (D) represent a comparison of fluxes in Tables S4–S6 with 99% confidence intervals with parameter continuation. (E) show mean, and 99% confidence intervals from the calculated growth rate from cell number counts on Incucyte SX5.

Increasing media serum reduced the contribution of glucose to lipogenic acetyl-CoA, supplying FASN and ELOVL fluxes (Figure 2C). Similar trends were observed in the contribution of FASN flux to the SM 42:1;O2 pool. This is consistent with reduced TGs, suggesting that cells cultured in lipid-depleted conditions access stored lipids and use available lipogenic acetyl-CoA for palmitate biosynthesis (Figures 2D and S2D). Notably, cells cultured in 1% FBS exhibited reduced synthesis of VLCFAs in SM compared with cells cultured in 10% or 20% FBS, a trend that correlated closely with cell growth rates (Figures 2D and 2E). These data suggest that VLCFAs are minor components of FBS and tightly linked to cell growth.

Next, we generated a glycerolipid reaction network to examine how FBS restriction and supplementation impacted these pathways (Tables S7–S9). Again, we selected relatively abundant, high enrichment lipid species for data input and flux modeling, including PC 34:1, PC 36:1, PE 34:1, PE 36:1, and TG 16:0_18:1_18:0. All of these lipid species are synthesized from DG 34:1 and DG 36:1 which we validated from MS2 fragmentation to be primarily composed of DG 16:0_18:1, and DG 18:0_18:1. This combination of fatty acids allowed us to determine FASN flux for glycerolipids as well as ELOVL6 and SCD1 flux as a response to different media lipid availabilities. While increasing FBS in the culture medium reduced the glucose-to-acetyl-CoA and FASN fluxes, it had no impact on G3P labeling (Figure S2F). Cells cultured in 1% FBS had more newly synthesized palmitate and stearate incorporated into the membrane lipids modeled by this reaction network. In all conditions, more than 50% of the network lipids turned over during the 24-h tracing period (Figure S2G), with the highest enrichment on M3 indicating high rates of headgroup turnover (e.g., within the Lands cycle) even when fatty acid enrichment was modest (Figure S2H). Interestingly, cells cultured in 1% FBS showed greater PC turnover but lower PE turnover, suggesting differential responses of these lipid classes to serum starvation, potentially due to PE becoming limited at lower FBS concentrations (Figure S2G). Triglyceride turnover (e.g., TG 52:1) was high at all FBS concentrations, with >90% of the pool containing some newly synthesized building block after 24 h (Figure S2G). These results demonstrate the dynamic nature of lipid metabolism in cells and validate the ability of Lipid-MFA to assess changes in specific fluxes under lipid-replete versus restricted conditions. Specifically, we observed that FBS availability had expected effects on acetyl-CoA generation and FASN flux but distinct impacts on elongation of VLCFAs present in membrane lipids.

LKB1 regulates LCB recycling through the lysosome

A549 cells lack the LKB1 tumor suppressor and exhibit aberrant AMPK signaling, which influences lipid homeostasis and lysosomal metabolism.^29,51–53 This lack of metabolic plasticity may account for the high rates of de novo biosynthesis and the stress sensitivity observed in LKB1-deficient cancer cells.^29,51,54,55 To determine whether Lipid-MFA could resolve dysregulated salvage fluxes in NSCLC, we stably expressed wild-type LKB1 (+LKB1) or an empty vector (+EV) in A549 cells (Figure S3A) and cultured each line in 10% FBS containing either [U-¹³C₆]glucose or [U-¹³C₃]serine for 24 h. Cellular lipids were then extracted and measured for isotope enrichment, and these data were analyzed using the sphingomyelin (Figure S3B; Tables S10 and S11) and glycerolipid (Tables S12 and S13) networks described above.

We observed no difference in growth or in the contribution of glucose to the acetyl-CoA or G3P pools used for Lipid-MFA (Figures S3C and S3D), similar to prior studies.²⁹ However, the re-introduction of functional LKB1 in A549 cells decreased the contribution of FASN to the palmitate pool and the synthesis of glycerolipids containing palmitate, namely PC 16:0_18:1 (Figures 3A and S3D–S3F). While the overall flux through CERS2 was comparable between these cells, LCB salvage flux to SL pools was increased 1.8-fold in +LKB1 cells (Figures 3B, 3C, and S3B). Specifically, salvage of SPB 18:1;O2 contributed more picomoles to CERS2-mediated SL biosynthesis in +LKB1 cells compared with +EV cells. Prior to Lipid-MFA, these shifts in de novo and salvage fluxes could not be resolved.⁵⁵ Changes in lipid pools were consistent with these flux results, as we observed a reduction in Cer 18:0;O2/24:0 and TG 52:1 pools as well as an increase in PC 34:1 and PC 36:1 pools within LKB1+ cells (Figure S3G). Lastly, we observed a striking increase in globoside pools in LKB1-proficient cells, suggesting a potential tie between LKB1 and globoside biosynthesis (Figure S3H). These data suggest that LKB1-proficient cells exhibit higher plasma membrane lipid exchange, consistent with their role in AMPK signaling and lysosomal biology.

Figure 3. LKB1 regulates long-chain base recycling through the lysosome.

(A) Fraction of newly synthesized fatty acids used to maintain SM 42:1;O2 network in A549 cells expressing an empty vector (+EV) or functional LKB1 (+LKB1).

(B) Comparison of de novo (SA) and salvage (SO) metabolic contribution to CERS2 based on sphingoid base acylation to maintain SM 42:1;O2 network in A549 cells expressing an empty vector (+EV) and functional LKB1 (+LKB1).

(C) Schematic showing enzymatic flux contributing to sphingoid base acylation along the de novo pathway (acylation of SA) or the salvage pathway (acylation of SO). (Created with Biorender.com).

(D) Fraction of newly synthesized fatty acids used to maintain SM 42:1;O2 network in A549 cells or H1299 cells.

(E) Comparison of de novo (SA) and salvage (SO) metabolic contribution to CERS2 based on sphingoid base acylation to maintain SM 42:1;O2 network in A549 cells and H1299 cells.

(F and G) Normalized cell number in A549 cells (F) and H1299 cells (G) cultured with ARC39 (SMPD1 inhibitor) (n = 4 wells per group).

In (A)–(D), data shown are a comparison of fluxes from SM 42:1;O2 network MFA with 99% confidence intervals in Tables S10 and S11 (A549 cells +EV and +LKB1) and Tables S14 and S15 (A549 and H1299 cell lines in 10% FBS), ** p < 0.01.

To evaluate whether Lipid-MFA could resolve these LCB synthesis and salvage fluxes across independent cell lines, we next compared fluxes in the SM 42:1;O2 network between A549 and H1299 cells (Tables S14 and S15), In contrast to P53 wild-type, LKB1-deficient A549 cells, the H1299 cell line expresses functional STK11 alleles but has homozygous deletion of TP53.⁵⁶ While there was no difference in glucose contribution to lipogenic acetyl-CoA flux (Tables S14 and S15), H1299 cells exhibited reduced palmitate and lignoceric acid (FA 24:0) synthesis, reduced de novo SPT flux, and increased salvage contributing to SM synthesis compared with A549 cells (Figures 3D and 3E). In general, A549 cells exhibited higher flux through lipid biosynthetic pathways spanning FASN through SPT, whereas salvage and recycling fluxes were greater in the H1299 cell line.

The above MFA data suggest that H1299 cells are more dependent on lipid salvage to support membrane lipid homeostasis as compared with A549 cells during normal growth. To determine whether these flux changes are functionally relevant, we cultured both lines at varying concentrations of ARC39, an inhibitor of sphingomyelin phosphodiesterase 1 (SMPD1), a lysosomal sphingomyelinase.⁵⁷ While we observed only a small reduction in growth rate in A549 cells at 1 μM treatment, H1299 cell growth was compromised at 1 μM and reduced at 100 nM ARC39s (Figures 3F and 3G). We confirmed target engagement after 24 h of treatment with 1 μM ARC39 by measuring sphingomyelin accumulation (Figures S3I and S3J). These observations provide evidence that Lipid-MFA can resolve functionally relevant changes in lipid homeostasis in cells with distinct genetic backgrounds.

Application of Lipid-MFA to PCLS culture

Next, we examined whether Lipid-MFA could resolve the same flux changes in a more complex, physiologically relevant microenvironment by applying PCLS culture to mice bearing NSCLC tumors.^58,59 We initiated lung tumors in Kras^LSL-G12D/+; Stk11^flox/flox (KL) and Kras^LSL-G12D/+; Trp53^flox/flox (KP) mice via intratracheal intubation of lentiviral Cre as previously described.^54,60 Approximately 16 weeks after tumor initiation, lungs were inflated with agarose and prepared for slicing and sub-culture as noted in the methods. After overnight recovery in fresh medium, slices were cultured in media containing either [U-¹³C₆]glucose or [U-¹³C₃]serine for 24 h (Figure 4A). Finally, slices were processed and analyzed for Lipid-MFA using the SM 42:1;O2 reaction network.

Figure 4. Application of Lipid-MFA to precision-cut lung slice culture.

(A) Schematic of the generation of precision-cut lung slices from GEMMs harboring KL and KP mutations. (Created with Biorender.com).

(B) Schematic of SM 42:1;O2 MFA with the net fluxes estimated from ¹³C Lipid-MFA for precision-cut lung slices harboring KL and KP tumors. Fluxes represent the picomoles needed to maintain a pool of 100 pmol of SM 42:1;O2. Black arrows represent overlapping confidence intervals, whereas colored arrows represent higher flux values with non-overlapping 99% confidence intervals in blue or red for tumors harboring KL and KP mutations, respectively. (Created with Biorender.com). Note: for simplicity, R32 captures the flux of SMPD1 and ASAH1 occurring in a single step.

(C) Fractional contribution of [U-¹³C₆]glucose to lipogenic acetyl-CoA in SM 42:1;O2 network.

(D) Fraction of newly synthesized fatty acids used to maintain membrane lipids in SM 42:1;O2 network.

(E) Comparison of de novo (SA) and salvage (SO) metabolic contribution to CERS2 based on sphingoid base acylation to maintain SM 42:1;O2 network lipids.

(F) Fractional turnover of SM 42:1;O2 in KL and KP PCLS after 24 h of culture.

Relative abundance is calculated by normalizing to an internal standard specific to lipid class.

In (C)–(F), the data shown are a comparison of fluxes estimated by ¹³C MFA with 99% confidence intervals from Tables S16 and S17, **p < 0.01.

The LKB1 tumor suppressor (encoded by Stk11) regulates AMPK and a lysosomal network encompassing numerous SL catabolic enzymes.^40,54 We therefore hypothesized that KL tumors would exhibit compromised LCB salvage and increased fatty acid biosynthesis compared with KP tumors. Model output results largely reflected the expected changes in flux across these tumor genotypes (Figure 4B; Tables S16 and S17). Lung slice cultures from mice bearing KP tumors exhibited a slight decrease in glucose flux to acetyl-CoA pools (Figure 4C), but more dramatic effects on fatty acid synthesis and SL salvage (Figures 4D and 4E). KP tumors showed high rates of salvage flux fueling CERS2 as well as reduced contributions of newly synthesized palmitic acid and lignoceric acid toward the synthesis of SM 42:1;O2 (Figures 4D and 4E). In contrast, KL tumor slice cultures revealed higher biosynthetic flux through FASN and SPT, lower recycling of LCBs, and an overall decrease in SM turnover compared with KP tumors (Figures 4D–4F). These data suggest that membrane lipid turnover is significantly higher in KP tumors than in KL tumors, which are unable to effectively regulate AMPK signaling and lysosomal biogenesis.^40,54 Notably, these fluxes are not resolvable through systemic approaches such as administration of ²H₂O, as we have previously tested in similar models.⁵⁵ Finally, we observed that KL tumors maintained higher SL pools in the SM 42:1:O2 synthesis axis, with significant increases in SA, DHCer, and SM pools, headgroup labeling from [U-¹³C₃]serine, and acyl-chain enrichment from [U-¹³C₆]glucose (Figures S4A–S4G). These data support the elevated SPT flux and reduced salvage fluxes resolved by Lipid-MFA in KL tumor-bearing PCLS cultures. Our modeling approach can therefore elucidate flux changes in microenvironments containing diverse cell types, which are likely to support lipid metabolism through distinct enzymatic activities, thereby better recapitulating tissue- or tumor-level homeostatic processes.

Lipid-MFA resolves the isozyme specificity of drugs

Enzymes involved in lipid homeostasis are highly druggable clinical targets,^61–67 but promiscuity and redundancies in these pathways complicate their analysis. Pharmacological agents may also have varied affinity for isozymes, leading to off-target effects that could impact lipid metabolism more broadly. Fumonisin B1 (FuB1) is a commonly used “pan-inhibitor” of CERS isozymes^68–71; however, it is often used at concentrations far greater than its reported IC₅₀ of 0.7 μM.^18,68 To investigate the overall impact to lipid metabolism from a moderate FuB1 dose, we modeled lipid metabolism in A549 cells with an expanded reaction network that incorporated mass isotopomer data for intermediates and pool sizes for SM, GM2, PC, PE, PS, and pPE species into an integrated model to estimate fluxes across multiple lipid classes (Figure S5A; Tables S18 and S19). Cells were cultured in media containing 20% FBS and either [U-¹³C₆]glucose or [U-¹³C₃]serine, and treated cells with 2 μM FuB1, a concentration that did not inhibit cell growth (Figure S5B). We quantified changes in flux through two CERS reactions catalyzed by CERS5/6 and CERS2, which are responsible for the N-acylation of SA and SO with palmitate or a VLCFA, respectively.

While glucose flux to lipogenic acetyl-CoA and LCFA synthesis was unchanged, Lipid-MFA resolved significant increases in lignoceric acid and nervonic acid synthesis (Figure 5A). We also observed opposite effects on reactions catalyzed by CERS5/6 versus CERS2. FuB1 reduced flux through CERS5/6, as noted by a reduction in Cer 18:0;O2/16:0, Cer 18:1;O2/16:0, and SM 34:1;O2 synthesis, reduced pools of these SLs, and increased SPB 18:0;O2 and SPB 18:1;O2 abundances (Figures 5B, S5A, S5C, and S5D). On the other hand, we observed a marked increase in flux through CERS2, along with elevated levels of SLs containing the VLCFAs lignoceric acid and nervonic acid (Figures 5B, S5A, and S5D). Similar changes were observed in pool sizes and synthesis fluxes toward GM2 gangliosides (Figures 5B and S5D). These changes in CERS isozyme fluxes caused further alterations throughout the network. For example, with reduced incorporation of palmitate into Cer 18:1;O2/16:0 and SM 34:1;O2, we detected a compensatory increase in PC 32:0 and PE 36:1 synthesis, suggesting reduced SL biosynthesis drives increases of selected phospholipid species (Figure S5A), as we previously observed in vivo with myriocin.¹⁰ We also detected a reduction in PS 34:1 and PS 36:1 pools along with their biosynthesis, but an increase in PS to PE conversion (Figure S5A; Tables S17 and S18). While we did not observe differences in plasmalogen metabolism, their fractional turnover was significantly lower than that of other glycolipids (Tables S18 and S19).

Figure 5. Lipid-MFA resolves the isozyme specificity of drugs.

(A) Contribution of newly synthesized fatty acids used to maintain membrane lipids in the reaction network (Tables S18 and S19).

(B) Newly synthesized pool of sphingomyelins and GM2 calculated from ¹³C MFA.

(C) M2 labeled ceramide pools from [U-¹³C₃]serine in A549 cells treated with vehicle or moderate doses of fumonisin B1 (FuB1) and cultured in 20% FBS for 24 h.

(D) M2 labeled sphingomyelin pools from [U-¹³C₃]serine in A549 cells treated with vehicle or moderate doses of FuB1 and cultured in 20% FBS for 24 h.

Abundances were calculated by normalizing to the appropriate internal standard from Ultimate Splash and μg of protein per sample. In (A) and (B), the data shown are a comparison of estimated fluxes from Tables S18 and S19 of the ¹³C MFA with 99% confidence intervals. Abundance data were analyzed using two-way ANOVA. *p <0.05, **p <0.01, ***p <0.001, ****p <0.0001.

To further understand the impact of FuB1 at moderate doses, we cultured A549 cells with [U-¹³C₃]serine and [U-¹³C₂]glycine and either 2, 5, or 10 μM FuB1 to assess effects on CERS fluxes as well as Cer and sphingomyelin pools. As expected,⁶⁹ we observed a dose-dependent increase in SPB18:0;O2 pools and SPT flux, with labeled pools increasing by 9-fold at 2 μM and over 180-fold at 10 μM FuB1 treatments (Figures 5C, S5E, and S5F). At the highest concentration tested, we observed a reduction in all measured Cer species (Figures 5C and S5G). However, sphingomyelin pools did not follow the same trend, such that sphingomyelins containing a LCFA (i.e., SM 34:0;O2 and SM 34:1;O2) were reduced, but those containing VLCFAs were significantly increased (Figures 5D and S5H). This “divergence” in behavior of long-chain versus very long-chain Cer flux was evident at moderate FuB1 doses, where we observed increases in both the overall and labeled pools of VLCFA-SMs. These results show a shift in flux of sphingoid bases from CERS5/6 toward CERS2, suggesting that at moderate dosing, FuB1 preferentially inhibits synthesis of long-chain Cer.

The above findings add to known distinctions in CERS isozyme function⁶⁹ given the specificity of FuB1 observed here. However, our Lipid-MFA also highlights the unique partitioning of long-chain (Cer 18:1;O2/16:0) versus very long-chain (Cer 18:1;O2/24:0) Cer in downstream sphingomyelin and glycosphingolipid synthesis (Figures 6A and 6B). While these components of lipid-ordered domains are synthesized from the same Cer pool, their fate is dependent on traffic to the Golgi by ceramide transfer protein (CERT) or coat protein complex II (COPII)-mediated vesicular flux.^72–74 Consistent with this mode of regulation, we observed that long-chain Cer were preferentially converted to sphingomyelin rather than to glycosphingolipids, such as GM2 (Figure 6C). On the other hand, very long-chain Cer were evenly partitioned between sphingomyelin and GM2 synthesis (Figure 6C). Lastly, we noted that incorporation of isotopes from [U-¹³C₃]serine was greater in Cer 18:1;O2/24:0 as compared with Cer 18:1;O2/16:0, with similar findings in downstream SM and GM2 pools, despite their similar rates of fractional turnover (Figures S6A–S6D). Therefore, we compared the fluxes through CERS2 and CERS5/6 for SA and SO in Table S18. Strikingly, CERS2 showed a strong preference toward acylation of SA, whereas CERS5/6 acylated SO and SA much more evenly (Figure 6D). These quantitative findings highlight key aspects of the SL pathway architecture and offer mechanistic insights toward therapeutically relevant functions of enzymes throughout the pathway.⁷⁵ Collectively, our results demonstrate the unique ability of Lipid-MFA to quantify fluxes through critical nodes of lipid homeostasis and capture the topology of lipid metabolism beyond the FASN-ACC (acetyl-CoA carboxylase) axis and pool size changes.

Figure 6. Lipid-MFA shows altered ceramide fate preference based on n-acyl chain length.

(A) Depiction of Cer 18:1;O2/16:0 as the hub for sphingomyelin synthesis or glycosphingolipid synthesis, depending on trafficking in A549 cells cultured with vehicle in 20%FBS. (Created with Biorender.com).

(B) Depiction of Cer 18:1;O2/24:0 as the hub for sphingomyelin synthesis or glycosphingolipid synthesis, depending on trafficking in A549 cells cultured with vehicle in 20%FBS. (Created with Biorender.com).

(C) Ratio of UGCG and SGMS fluxes based on n-acyl chain.

(D) Ratio of SA to SO acylation based on CERS isozyme.

In (A)–(D), the data shown are a comparison of estimated fluxes from Table S18 of the ¹³C MFA with 99% confidence intervals, **p < 0.01. Thickness of arrows for (A) and (B) represents the flux for each enzyme.

DISCUSSION

Here, we demonstrate the ability of Lipid-MFA to untangle the complexity of lipid metabolic pathways, providing a quantitative, systems-based approach for studying lipid homeostasis beyond fatty acid biosynthesis. For decades, stable isotope tracing and downstream modeling have provided key insights into biopolymer synthesis in cells, animals, and patients, highlighting critical tissue sites for lipogenesis in numerous disease contexts.^{26,47,76–78} The labeling information and network analysis outlined here can provide quantitative information on headgroup and acyl-chain flux across numerous classes of membrane lipids. In turn, one can identify the most active (or inactive) lipid-metabolizing enzymes within a given biological system, including recycling pathways that may span several organelles or cell types, as well as core pathways such as the Lands cycle. While custom tracers or wash-out approaches can facilitate measurement of fluxes associated with membrane lipid uptake, catabolism, and recycling, their applications are costly and technically challenging.⁷⁹

More accurate and sensitive high-resolution mass spectrometry technology facilitates this method, providing unprecedented detail on lipid homeostasis.^80–82 This framework enables quantification of lipid metabolism beyond the established ACLY-ACC-FASN axis of reactions, which have been the focus of most lipid flux studies to date, with some focusing on glycerol-3 phosphate turnover as a proxy for lipid turnover, while not accounting for fatty acid biosynthesis and elongation.³⁰ Our comparisons of lipid metabolism in PCLS versus cell culture in media with different serum availability also highlight the non-physiological nature of tissue culture. Lipids are readily available from the local microenvironment (e.g., tumor interstitial fluid)⁸³ or circulation in vivo, and our results suggest that lipid synthesis becomes highly active in under 24 h of culture in “typical” FBS-containing culture conditions. These findings contrast with observations of “polar” substrate concentrations (glucose, amino acids, etc.) in typical culture media, which are present at supraphysiological levels,^84,85 since tissue culture conditions are inherently deficient in lipids relative to in vivo conditions. This result may have profound implications for the metabolism of cells adapted to in vitro culture and data generated under these conditions.⁸⁶

The tumor microenvironment contains diverse cell types, and this heterogeneity facilitates metabolic crosstalk and lipid exchange beyond that provided by circulating lipoproteins.^32,87–91 This complex environment provides tumors with an array of pathways to support proliferation. Using PCLS culture, our approach successfully quantified changes in lipid salvage and biosynthesis in NSCLC tumors with specific genotypes. Indeed, autophagy, lipophagy, and other salvage pathways are emerging as critical mediators of tumor progression, neurodegeneration, and aging,^14,15,92 so precise quantitation of these pathways is becoming increasingly important in biology. The redundancy and flexibility of such exchange fluxes and metabolite interconversions provide an innate biochemical resilience that is necessary for cell survival.

While in vitro culture conditions will never perfectly recreate the tumor microenvironment, the molecular detail and tunability afforded by tissue culture Lipid-MFA will be an invaluable tool for mechanistic studies that identify gene function or inhibitor specificity. Many genes encoding lipid-metabolizing enzymes remain unannotated, and it remains challenging to resolve the distinct functionalities of tissue-specific isozymes or disease variants for key enzymes in lipid metabolism. This issue is particularly relevant in characterizing the function of SL enzymes linked to diseases such as amyotrophic lateral sclerosis (ALS) and sensory neuropathy.^93–95 Similarly, identifying potential off-target effects of inhibitors is critical for clinical drug development. Evaluating candidates using methods beyond pool size changes could aid in triaging some molecules.

Finally, the application of Lipid-MFA will enable testing hypotheses focused on specific mechanistic questions while capturing broader lipidomic changes at the systems level. With higher-resolution equipment becoming more available, the application of distinct stable isotope elements can add an additional layer of information. For example, application of [¹³C₆]glucose and labeled PUFAs can enable simultaneous modeling of de novo lipogenesis, essential fatty acid uptake, and their fates across lipid classes.^96,97 The use of spatial mass spectrometry imaging technology can, in turn, provide information on local flux changes (e.g., ACC-FASN) within a tissue,^80,81 and integration of these pipelines into Lipid-MFA may provide deeper information on lipid trafficking in distinct cell types. The comprehensive flux quantitation afforded by Lipid-MFA will prove invaluable for addressing such questions in the future, revealing the dynamic nature of membrane lipid metabolism.

Limitations of the study

In the studies and models outlined here, we focused on abundant lipids that were actively synthesized and salvaged, based on enrichment data from glucose and serine. While measuring and modeling all lipids may be impractical, focused studies on lipid subclasses of interest will provide critical mechanistic insights into lipid homeostasis. Therefore, Lipid-MFA should be viewed as modular in nature, such that focused studies on specific lipid pathways or sub-networks can be executed easily. We quantify differences in fluxes when culturing cells in 1%, 10%, or 20% FBS to highlight the general deficiency of lipids in normal tissue culture, especially as cells become more confluent. However, other nutrients such as glucose and amino acids are present at supraphysiological levels, as highlighted by several recent cancer metabolism studies.^84,98 We also predominantly conducted “steady-state” models assuming precursor enrichment is constant, estimating time-dependent fractional synthesis of lipid pools rather than performing kinetic, non-stationary MFA (INST-MFA). In test cases, we found that sufficient information was available to resolve fluxes using steady-state modeling, and the computational time required for INST-Lipid-MFA modeling and parameter continuation is prohibitive for routine use. However, application of non-stationary models is feasible if required to increase flux resolution further. Nevertheless, our results highlight the broad impacts of Lipid-MFA in assessing lipid homeostasis to answer distinct biochemical questions spanning the lipidome.

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Christian M. Metallo (metallo@salk.edu).

Materials availability

No other unique materials were generated in this study.

Data and code availability

• All raw mass spectrometry data are uploaded to the Metabolomics Workbench (https://doi.org/10.21228/M8FC3V).

• Values used to create all graphs in the paper and uncropped images are available in Data S1 and S3.

• This paper does not report original code, but all MATLAB INCA files with the appropriate reaction networks and input for the Lipid-MFAs can be found in Data S2.

• Data S3 contains the precursor and product ions used to identify lipids, mass isotopomer distributions, and Lipid-MFA results, related to Figures 1, 2, 3, 4, 5, 6, and S1–S6.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

STAR★METHODS

EXPERIMENTAL MODEL DETAILS

Cell culture experiments

A549 and H1299 cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM, Gibco) with 1%, 10%, or 20% FBS and 1% penicillin/streptomycin (P/S). A549 cells were transduced with pBabe (empty vector; Addgene #1764) or pBabe-FLAG-LKB1 (Addgene #8592) LKB1 as previously described.⁹⁹ Cells with the integrated vectors were selected for using 400 ug/ml of Hygromycin in cell culture. Cell lines were confirmed to be free of Mycoplasma (Lonza LT07–318). All media were adjusted to pH = 7.3 and cells were cultured at 37°C with 5% CO₂.

Cell proliferation and ¹³C tracing

Cells were plated in growth media (10% FBS) and allowed to adhere for 24 hours before changing to the specified growth media. Cell counts were performed using the Incucyte SX5 for 66–72 hours after media change. Best fit exponential growth curves were calculated to fit the growth data and 99% confidence intervals for growth rates (hr⁻¹) were calculated.

¹³C isotope tracing media was formulated using a custom Hyclone glucose-, amino acid-, and sodium pyruvate-free DMEM (Cytiva Life Sciences) supplemented with either 25 mM [U-¹³C₆]glucose or 0.4 mM [U-¹³C₃]serine while replacing the remainder amino acids or unlabeled glucose to achieve the necessary media composition and supplemented with the amount of FBS necessary to achieve the condition described in each experiment. Cultured cells were washed with 1 mL of phosphate-buffered saline (PBS) before applying tracing media for 24 hours as indicated in figure legends. Cells were cultured with vehicle or 2 μM, 5 μM, 10 μM fumonisin B1 overnight prior to the application of tracing media. These cells were cultured with 0.4 mM [U-¹³C₃]serine and 0.4 mM [U-¹³C₂]glycine for 24 hours prior to extracting.

Animal Experiments

All animal studies were approved by the Salk Institute Institutional Animal Care and Use Committee. Mouse strains were maintained on FVB/N background. KL (Kras^LSL-G12D/+;Stk11^flox/flox;R26^{LSL-Luc/LSL-Luc}) and KP (Kras^LSL-G12D/+Trp53^flox/flox; R26^{LSL-Luc/LSL-Luc}) mice have been described previously.⁵⁴ Lung tumors were initiated by the delivery of lentiviral Cre recombinase (4 × 10⁵ plaque-forming units per mouse) via intratracheal intubation as described previously⁶⁰ in six month old female mice. Lung tissues were collected immediately for tissue slice culture upon observation of respiratory distress (15–18 weeks after tumor initiation). Mice were fed standard chow and kept at 22°C ambient temperature with 40% humidity and a 12-hour light/dark cycle.

Precision-cut lung slice culture

Low-melt agarose (LMA) was prepared prior to procuring lungs. 3% w/V low-melt agarose Type VII-A (Sigma, A0701) was added to a 0.9% NaCl solution, microwaved to dissolve, and placed into a 42°C water bath. A 3ml syringe tipped with a blunt 21-gauge needle was filled with 2ml of LMA and kept warm.

Mice were sacrificed and trachea was nicked with a 25-gauge needle. Warm LMA was pushed into the trachea using the pre-filled syringe. The lungs were slowly inflated with approximately 2ml of LMA, and the thorax packed with ice to solidify the agarose prior to withdrawing the needle from the trachea. The lungs were removed from the thorax en bloc, the lobes separated out and placed in chilled, sparged (with 95% O₂/5% CO₂ carbogen gas) slicing buffer. The lobes were superglued to a stage with the hilium facing downward, and 300μm sections cut on a Leica VT1200S with Vibracheck. Prior to incubation at 37°C, sections were kept continuously in chilled, sparged slicing buffer (Krebs-Henseleit buffer (Millipore Sigma, Cat# K3753). Lung slices were matched for size and tumor burden, placed into individual wells in a 12 well suspension culture plate (greiner bio-one Cellstar, #665 102) containing warmed, oxygenated resting media (2 ml, RPMI1640 with 2%FBS (GIBCO^™ ThermoFisher Scientific, Cat# 11875093) and incubated overnight while on an orbital shaker.

Following an overnight recovery in RPMI, slices were washed three times with 0.9% saline and tracing DMEM media containing [U-¹³C₆]glucose or [U-¹³C₃]serine was added. After 24 hours, tracing media was collected, and the lung slices rinsed 3 times with 0.9% (w/V) saline. After the final saline rinse was removed, the slices were quick frozen by placing the 12 well culture plate on top of a metal block immersed in a slurry of isopropanol and dry ice. Biopsy punches were used to isolate frozen disks of tumor and adjacent lung tissue for separate analysis.

METHOD DETAILS

Western Blots

Protein lysates were prepared in the following lysis buffer supplemented with protease (Roche, 11836153001) and phosphatase (Roche, 4906837001) inhibitors: 100 mM Tris-HCl (pH 7.4), 300 mM NaCl, 2% Triton X-100, 2 mM EDTA, 0.2% SDS, and 1% Sodium Deoxycholate. Cell lysates in prepared lysis buffer were homogenized by water bath sonication. Snap frozen micro-dissected lung tumors were homogenized on ice in 50 μL lysis buffer by mechanical dissociation followed by microtip sonication. Total protein input was normalized across samples by evaluating protein concentration by BCA (Thermo Fisher Scientific 23225). Denatured protein lysates were resolved on 4–12% Bis-Tris gels (Invitrogen, NP0335BOX) and transferred to PVDF membranes (Thermo Fisher Scientific, 88518). The resulting membranes were blocked in 5% milk in TBS-T 0.1%, incubated overnight at 4°C in diluted primary antibody, washed with TBS-T, incubated for one hour in secondary antibody diluted in in TBS-T plus milk, washed in TBS-T, and developed using Clarity Western ECL Substrate (Bio Rad, 170–5060). Secondary antibodies were anti-rabbit (Millipore, AP132P) and anti-mouse (Millipore, AP124P).

Metabolite extraction and GC-MS analysis

At the conclusion of the tracer experiment, media was aspirated. Then, cells were rinsed twice with 0.9% saline solution and lysed with 500 μL of ice-cold methanol. 10 nanomoles of norvaline was added to each sample along with lipid standards. Cells were scraped from the plate and lysates transferred to Eppendorf tubes. Samples were split in two. One portion was used for polar lipid analysis on LCMS and the other further extracted and separated to generate an aqueous layer containing polar metabolites for GCMS analysis and a hydrophobic phase used for analysis of neutral lipids on LCMS. Separation of polar and organic phases was performed by addition of 200uL of water and 400 μL of chloroform to the 200 μL of methanol left in the Eppendorf tube. Samples were then vortex and centrifuged at 21,000gs for 5 min at 4°C. The organic phase was collected and 2 μL of formic acid was added to the remaining polar phase which was re-extracted with 400 μL of chloroform. Combined organic phases were dried under nitrogen and used for neutral lipid analysis.

Polar lipid samples were centrifugation at 21,000 g for 15 min at 4°C. 300 μL of methanol were collected and evaporated under nitrogen and stored at −80°C until resuspended for analysis. 250 μL of the upper aqueous phase were transferred to a GC vial and dried at 4°C overnight. 250uL of lower organic layer was collected and evaporated under nitrogen at room temperature. Dried polar metabolites were processed for gas chromatography–mass spectrometry (GC-MS) as described previously by Cordes and Metallo.¹⁰⁰ Briefly, polar metabolites were derivatized using a Gerstel MultiPurpose Sampler (MPS 2XL). Methoxime–tert-butyldimethylchlorosilane (tBDMS) derivatives were formed by addition of 15 μL of 2% (w/v) methoxylamine hydrochloride (MP Biomedicals, Solon, OH) in pyridine and incubated at 45°C for 60 min. Samples were then silylated by addition of 15 μL of N-tert-butyldimethylsily-N-methyltrifluoroacetamide (MTBSTFA) with 1% tBDMS (Regis Technologies, Morton Grove, IL) and incubated at 45°C for 30 min.

Derivatized polar samples were injected into a GC-MS using a DB-35MS column (30 m by 0.25 mm i.d. by 0.25 μm; Agilent J&W Scientific, Santa Clara, CA) installed in an Agilent 7890B GC system integrated with an Agilent 5977a MS. Samples were injected at a GC oven temperature of 100°C which was held for 1 min before ramping to 255°C at 3.5°C/min and then to 320°C at 15°C/min and held for 3 min. Electron impact ionization was performed with the MS scanning over the range of 100 to 650 mass/charge ratio (m/z) for polar metabolites. Metabolite levels and mass isotopomer distributions were analyzed with an in-house MATLAB script which integrated the metabolite fragment ions and corrected for natural isotope abundances.

LC-MS/MS analysis

For experiments in Figures 2 and 3, adherent cells were washed and scraped into 500μL of methanol and spiked with deuterated internal standards: Equisplash (Avanti Polar Lipids, Cat# 330731), SPB 18:0;O2 {d7] (Avanti Polar Lipids, Cat# 860658), SPB 18:1;O2 [D7] (Avanti Polar Lipids, Cat# 860657), and G_M3 18:1;O2/18:0 [D3] (Cayman Chemicals, item No. 39226). For experiments in Figures 4, 5, and 6, adherent cells and tumor punches were washed and scraped or homogenized in 500uL of methanol and spiked with deuterated internal standards: Ultimate splash (Avanti Polar Lipids, Cat#330820), SPB 18:0;O2 [D7] (Avanti Polar Lipids, Cat# 860658), SPB 18:1;O2 [D7] (Avanti Polar Lipids, Cat# 860657), glucosylceramide 18:1;O2[D7]-15:0 (Avanti Polar Lipids, Cat# 330729), lactosylceramide 18:1;O2[D7] (Avanti Polar Lipids, Cat# 330727), and G_M3-d₃ 18:1;O2/18:0 [D3] (Cayman Chemicals, item No. 39226). Ten percent of the lysate was transferred to 96 well plate, dried and redissolved in 20 μL of M-PER buffer (Thermo Fisher Scientific Cat. No. 78501) for protein estimation with BCA assay. The remaining methanol extract and the tubes were vortexed and spun at 21000 × g for 15 minutes. The supernatants were transferred to a new tube and dried under nitrogen gas. The dried samples were resuspended in 60 μL of the initial running buffer and then analyzed with LC-MS/MS. 5 μL of sample were injected.

Polar lipids

Chromatographic separation and lipid species identification was performed using Q Exactive orbitrap mass spectrometer with a Vanquish Flex Binary UHPLC system (Thermo Fisher Scientific) equipped with an Kinetex C18 column, 100 × 2.1 mm, 1.7 μm particle (Phenomenex) column at 35 °C. Chromatography was performed using a gradient of 98:2 v/v water: methanol with 5 mM ammonium acetate (mobile phase A) and 50:50 v/v methanol: isopropanol with 5 mM ammonium acetate (mobile phase B), both at a flow rate of 0.2 mL/min. The liquid chromatography gradient ran with the following profile: 0 min, 30% B; 1 min, 30% B; 2 min, 70% B; 11 min, 95%B; 17 min, 30%B; 21.5 min, 30%B; 27 min, 30% B. Lipids were analyzed in positive mode using spray voltage 3.5 kV. Sweep gas flow was 1 arbitrary units, auxiliary gas flow 10 arbitrary units and sheath gas flow 50 arbitrary units, with a capillary temperature of 325 °C. Full mass spectrometry (scan range 220–2,500 m/z) was used at 140,000 resolution with 10E6 automatic gain control and a maximum injection time of 100 ms. Data dependent MS2 (Top 12) mode at 17,500 resolution with automatic gain control set at 10E5 with a maximum injection time of 50 ms was used.

MS2 data acquired in the negative mode were used for confirming the identities of PC and PE molecular species. Lipids were analyzed in negative mode using spray voltage 3.9 kV. Sweep gas flow was 1 arbitrary units, auxiliary gas flow 10.63 arbitrary units and sheath gas flow 46.25 arbitrary units, with a capillary temperature of 250 °C. Full mass spectrometry (scan range 250–1,800 m/z) was used at 70,000 resolution with 10E6 automatic gain control and a maximum injection time of 100 ms. Data dependent MS2 (Top 12) mode at 17,500 resolution with automatic gain control set at 10E5 with a maximum injection time of 50 ms was used.

Neutral lipids

Chromatographic separation and lipid species identification for neutral lipids was performed using Q Exactive orbitrap mass spectrometer with a Vanquish Flex Binary UHPLC system (Thermo Scientific) equipped with an Accucore C30, 150 × 2.1 mm, 2.6 μm particle (Thermo) column at 40 °C. Chromatography was performed using a gradient of 40:60 v/v water: acetonitrile with 10 mM ammonium formate and 0.1% formic acid (mobile phase A) and 10:90 v/v acetonitrile: propan-2-ol with 10 mM ammonium formate and 0.1% formic acid (mobile phase B), both at a flow rate of 0.2 ml min−1. The liquid chromatography gradient ran from 30% to 43% B from 3–8 min, then from 43% to 50% B from 8–9 min, then 50–90% B from 9–18 min, then 90–99% B from 18–26 min, then held at 99% B from 26–30 min, before returning to 30% B in 6 min and held for a further 4 min. Neutral lipids were analyzed in positive mode using spray voltage 3.9 kV. Sweep gas flow was 2 arbitrary units, auxiliary gas flow was 11 arbitrary units, and sheath gas flow was 46 arbitrary units, with a capillary temperature of 325°C. Full MS (scan range, 200 to 1400 m/z) was used at 70,000 resolution with 10E6 automatic gain control and a maximum injection time of 100 ms. Data-dependent MS2 (Top 6) mode at 17,500 resolution with automatic gain control set at 10E5 with a maximum injection time of 50 ms was used.

Analysis of Lipid Mass Isotopomer Data

Data were analyzed using EI-Maven. Mass error was set to 5 ppm for metabolite identification. Mass isotopomers were filtered to elute within 5 seconds of the unlabeled parent ion to ensure all labeled species came from the same metabolite. Lipid species–specific fragments used for identification and quantitation are presented in Table S19. Relative abundance was calculated by normalizing to internal standards specific to the lipid class and protein levels. Absolute abundances were calculated by normalizing to specific lipid species in Ultimate splash or specific deuterated standards added during extraction. Mass isotopomer distributions were analyzed with an in-house MATLAB script which integrated the metabolite fragment ions and corrected for natural isotope abundances.

¹³C Lipid Metabolic Flux Analysis

ISA and MFA were performed to estimate the percent of newly synthesized lipids as well as the contribution of the tracer of interest to lipogenic acetyl-CoA pools, glycerol-3P turnover, serine pools for LCB synthesis, and the metabolic flux necessary to maintain lipid pools. Serine and glycerol-3P were provided in excess with unused metabolites diverted to “sink”. This allowed us to measure relative fluxes to maintain the lipid pools. For experiments in Figures 2, 3, 4, and 5, lipid pools were set to 100, without restricting any intermediates. For the experiment in Figure 6, each lipid pool was entered into the model as a measured pool to capture the nanomolar flux through enzymes to sustain the entered lipidome. A χ² statistical test was applied to assess the goodness-of-fit using α of 0.01. Parameters for contribution of ¹³C tracers to lipogenic acetyl-CoA (D value) and percentage of newly synthesized lipids [g(t) value] and their 99% confidence intervals are then calculated using best-fit model after estimating 50 times using random initial guesses for all fluxes in the network in INCA MFA (isotopomer network compartmental analysis metabolic flux analysis) software.³⁸

Experimental lipid labeling from [U-¹³C₆]glucose and [U-¹³C₃]serine after a 24-hour trace, as indicated in figure legends, was compared to simulated labeling using reaction networks for SM 42:1;O2, glycerophospholipids, or a broader range of lipids as desired and described for the experiment. Experimental replicates were entered as different wells. Estimated flux values and parameter continuation were performed with all replicates active for each of the isotopically labeled substrate used in parallel. Mass Isotopomer distributions were entered into INCA after natural isotope correction using an in-house MATLAB script. When background noise in the mass isotopomer distributions were identified from ¹²C labeling wells or in regions with unexpected labeling, that mass isotopomer value was replaced with “NaN” to ignore the value during flux estimation and parameter continuation. This resulted in less than 5% of all mass isotopomers being removed. This was necessary due to coelution of lipid species with similar masses and the added isotopic complexity resulting from the addition of stable isotope tracing.

MFA data are plotted as 99% confidence intervals. “**” indicates statistical significance by non-overlapping confidence intervals. Reaction networks fluxes can be found in Tables S4–S19, with the MS1 ions used for identification and MS2 ions used for metabolite confirmation found in Table S1 in Data S1. Matlab INCA files used to generate Tables S4–S19 can be found in files Data S2. All MIDs used in the Lipid-MFAs presented can be found in Table S20 of Data S3_Supplemental_Tables as well as in the matlab files provided in Data S2.

¹³C metabolic flux analysis was conducted under the following assumptions:

1. Cells were assumed to be at metabolic steady state

2. Acetyl-CoA, glycerol-3P, and serine, as lipid precursors, are assumed to be at isotopic steady state as they turnover much faster than membrane lipids and triglycerides, whose time-dependent fractional synthesis is modeled as the g(t) parameter. We demonstrate that glycerol-3P from [U-¹³C₆]glucose and the serine pool from [U-¹³C₃]serine reach steady at the beginning of the culture (Figures 2F and S2G). This assumption has been routinely used to model acetyl-CoA contributions to fatty acid and cholesterol synthesis.

3. Cells proliferate exponentially.

Mole percent enrichment (MPE) represents the contribution of carbons from a specific source (i.e. glucose) to a metabolite and is calculated using the following equation:

$$\text{MPE}=\frac{\sum _{i=1}^n{x}_i}{n}\mathrm{*}100$$

Where ${\mathrm{x}}_{\mathrm{i}}$ represents the fractional enrichment of a mass isotopomer $\left({\mathrm{M}}_1+{\mathrm{M}}_2+\dots {\mathrm{M}}_{\mathrm{n}}\right)$ summed over the number of carbons “$\mathrm{n}$”. For example, the MPE of SM 42:1;O2 is calculated using the following equation:

$${\mathrm{M}\mathrm{P}\mathrm{E}}_{\text{SM}42:1;02}=\frac{M_1+M_2+M_3+M_4+M_5+\dots +M_{41}+M_{42}}{42}\mathrm{*}100$$

QUANTIFICATION AND STATISTICAL ANALYSIS

Data are presented as mean ± standard error of mean (SEM) of at least three biological replicates as indicated in figure legends. Statistical analysis was performed with GraphPad Prism 10.3.1 using two-tailed independent t-test to compare two groups, one-way ANOVA with Fisher’s least significant difference (LSD) post hoc test to compare more than two groups, two-way ANOVA with Fisher’s LSD post hoc test to compare two-factor study designs. For all tests, * p< 0.05, **p < 0.01, *** p< 0.001, or **** p<0.0001.

Supplementary Material

SUPPLEMENTAL INFORMATION

Supplemental information can be found online at https://doi.org/10.1016/j.cmet.2026.01.020.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
LKB1 antibody	Cell Signaling Technologies	Cat # 3047; RRID:AB_2198327
p-ACCS79 antibody	Cell Signaling Technologies	Cat # 3661; RRID:AB_330337
ACC antibody	Cell Signaling Technologies	Cat # 3662; RRID:AB_2219400
p-AMPKαT172 antibody	Cell Signaling Technologies	Cat # 2535; RRID:AB_331250
Actin antibody	Sigma Aldrich	Cat # A5441; RRID:AB_476744
anti-rabbit antibody	Sigma Aldrich	Cat # AP132P; RRID:AB_90264
anti-mouse antibody	Millipore Sigma	Cat # AP124P; RRID:AB_90456
Biological samples
Mouse Lungs with tumors	This Paper	N/A
Chemicals, peptides, and recombinant proteins
[U-13C6] D-glucose	Cambridge Isotope Laboratories	CLM-1396
[U-13C3] L-serine	Cambridge Isotope Laboratories	CLM-1574
[U-13C2] L-glycine	Cambridge Isotope Laboratories	CLM-1017
D-glucose	Sigma Aldrich	G8270
L-serine	Sigma Aldrich	S4500
L-glycine	Sigma Aldrich	1295800
DMEM	Thermo Fisher Scientific	Cat#11965-092
DMEM without glucose, L-glutamine, phenol red, sodium pyruvate and sodium bicarbonate	Sigma Aldrich	D5030
DMEM without glucose, amino acids, phenol red, sodium pyruvate, and sodium bicarbonate	Cytiva	Custom order
RPM1640	Thermo Fisher Scientific	11875093
Fetal Bovine Serum (FBS)	Thermo Fisher Scientific	16000-044
Penicillin-streptomycin	Thermo Fisher Scientific	15130-122
Dubelcco’s Phosphate Buffered Saline (DPBS)	Thermo Fisher Scientific	141190-144
Dimethyl sulfoxide	Thermo Fisher Scientific	BP231-100
Fumonisin B1	Enzo Life Sciences	BML-SL220
Sodium Chloride	Sigma Aldrich	S9888
LiChrosolv 2-propanol	Sigma Aldrich	1.02781
LiChrosolv methanol	Sigma Aldrich	1.06035
LiChrosolv acetonitrile	Sigma Aldrich	1.00029
Formic Acid	Sigma Aldrich	5.43804
Water	Sigma Aldrich	270733
Chloroform	Sigma Aldrich	366927
Ammonium formate	Sigma Aldrich	70221
Ammonium acetate	Sigma Aldrich	73594
Methoxylamine hydrocloride	Sigma Aldrich	89803
MSTFA Silylation Reagent	Macherey-Nagel	701270.510
UltimateSPLASH One Mix	Avanti Polar Lipids	330820
EquiSPLASH	Avanti Polar Lipids	330731
SPB 18:0;O2 [d7]	Avanti Polar Lipids	860658
SPB 18:1;O2 [D7]	Avanti Polar Lipids	860657
glucosylceramide 18:1;O2[D7]/15:0	Avanti Polar Lipids	330729
lactosylceramide 18:1;O2[D7]/15:0	Avanti Polar Lipids	330727
GM3-d3 18:1;O2/18:0 [D3]	Avanti Polar Lipids	39226
Mammalian Protein Extraction Reagent	Thermo Fisher Scientific	78501
Protease Inhibitor	Roche	11836153001
Phosphatase Inhibitor	Roche	4906837001
4–12% Bis-Tris gels	Invitrogen	NP0335BOX
PVDF membranes	Thermo Fisher Scientific	88518
Clarity Western ECL Substrate	Bio Rad	170-5060
Novex Tris-Glycine SDS Running Buffer (10X)	Thermo Fisher Scientific	LC2675-5
20X TBS Tween-20 Buffer	Thermo Fisher Scientific	28360
Low-melt agarose Type VII-A	Sigma Aldrich	A0701
Krebs-Henseleit Buffer	Sigma Aldrich	K3753
Bovine Serum Albumin	Sigma Aldrich	A6003
Critical commercial assays
Pierce BCA Kit	Thermo Fisher Scientific	23227
Deposited data
Mass Spectrometry Data (Project ID:	Metabolomics Workbench	https://doi.org/10.21228/M8FC3V
Data S1	This paper	N/A
Data S2	This paper	N/A
Data S3	This paper	N/A
Experimental models: Cell lines
A549 cells	ATCC	RRID: CVCL_0023
H1299 cells	Provided by Adam J. Engler, University of California, San Diego, La Jolla, California, USA	RRID: CVCL_0060
Experimental models: Organisms/strains
KL (KrasLSL-G12D/+;Stk11flox/flox;R26LSL-Luc/LSL-Luc)	Provided by Reuben J. Shaw, Salk Institute for Biological Studies, La Jolla, California, USA	(Shackelford et al.99)
KP (KrasLSL-G12D/+;Trp53flox/flox; R26LSL-Luc/LSL-Luc)	Provided by Reuben J. Shaw, Salk Institute for Biological Studies, La Jolla, California, USA	(Shackelford et al.99)
Software and algorithms
Matlab R2025a	Mathworks	https://www.mathworks.com/products/matlab.html
INCA 2.4	Vanderbilt University	https://mfa.vueinnovations.com/licensing
GraphPad Prism 10.6.1	GraphPad	https://www.graphpad.com/scientific-software/prism/
EL-MAVEN V0.12.1	Elucidata	https://docs.polly.elucidata.io/Apps/Metabolomic%20Data/El-MAVEN.html
Other
DB-35 MS column, 30 m × 0.25 mm i.d. × 0.25 μm	Agilent	122-3832UI
Accucore C30, 150 × 2.1 mm, 2.6 μm particle	Thermo Fisher Scientific	27826-152130
Kinetex C18 100 × 2.1 mM, 1.7 μm particle	Phenomenex	00D-4475-AN
Biorender	Biorender	Biorender.com

Highlights.

• Lipid-MFA measures synthesis and salvage fluxes that support lipid homeostasis

• Precision-cut lung slice culture enables Lipid-MFA in the tumor microenvironment

• Loss of p53 or LKB1 tumor suppressors distinctly impacts sphingolipid flux in NSCLC

• LC- versus VLC-ceramides are differentially trafficked to complex sphingolipids