Spatiotemporal metabolic networks in a preclinical model of pancreatic cancer and associated cachexia

Blanca Majem; Min-Sik Lee; Kyung Cheul Shin; Insia Naqvi; Courtney Dennis; Lucas Dailey; Sarah Jeanfavre; Mari Mino-Kenudson; Joao B Xavier; Clary B Clish; Deepti Mathur; Nada Y Kalaany

doi:10.1038/s41467-026-77745-0

. 2026 Sep 18;17:10949. doi: 10.1038/s41467-026-77745-0

Spatiotemporal metabolic networks in a preclinical model of pancreatic cancer and associated cachexia

Blanca Majem ^1,², Min-Sik Lee ^1,², Kyung Cheul Shin ^1,², Insia Naqvi ¹, Courtney Dennis ³, Lucas Dailey ³, Sarah Jeanfavre ³, Mari Mino-Kenudson ^4,⁵, Joao B Xavier ⁶, Clary B Clish ³, Deepti Mathur ^1,^2,^6,^8,^✉, Nada Y Kalaany ^1,^2,^7,^✉

PMCID: PMC13680358 PMID: 42897170

Abstract

Cachexia or body wasting is a significant contributing factor to the morbidity and mortality of pancreatic cancer. Despite an unmet need, its metabolic underpinnings remain poorly understood. Here, we characterize a mouse model that recaptures the longitudinal progression of human cachexia during pancreatic cancer development. We profile the metabolomes of pre-, early- and late-stage cancer-associated cachexia in male and female pancreas, its interstitial fluid (IF), plasma, liver, adipose tissue and skeletal muscle. We find that each tissue has a unique metabolome and trajectory across stages and reveal a marked lipid enrichment in the tumor IF. Using mathematical modeling, we identify metabolites participating in cross-tissue networks and in vivo validate a lactate-glucose conversion. We find systemic metabolic changes prior to weight loss and use feature selection algorithms to identify early detection markers. Our study provides a resource for system-wide metabolic evolution of pancreatic cancer and associated cachexia, highlighting the potential for timely diagnostics and interventions.

Subject terms: Cancer metabolism, Pancreatic cancer

Cachexia commonly develops in patients with pancreatic cancer and is closely linked to increased morbidity and mortality. Here, the authors provide an in-depth longitudinal characterization of systemic metabolomic alterations in a mouse model of pancreatic cancer–associated cachexia.

Introduction

Pancreatic ductal adenocarcinoma (PDA), the major form of pancreatic cancer, is a lethal malignancy with rising incidence and no effective therapies^1,2. Compounding the dismal outcomes in PDA is the manifestation in over 80% of cases of cancer-associated cachexia³, or “CAC”. Cachexia is a life-threatening syndrome of skeletal muscle wasting frequently accompanied by fat wasting that occurs as a consequence of chronic disease, including cancer⁴. CAC causes pronounced frailty in patients, preventing them from receiving anti-cancer therapy and from enrolling in clinical trials that could extend survival.

The cachectic syndrome, core definition of which involves notable loss of muscle mass that cannot be reversed by conventional nutrition⁴, is often burdened by systemic inflammation, anorexia, loss of fat mass, and enhanced energy expenditure^4,5. Despite this core defining metabolic feature, however, the metabolic aspect of cachexia has not been thoroughly characterized. We posited that a longitudinal systemic analysis of metabolic network alterations that occur during the progression of cancer and its associated cachexia would be highly valuable for identifying promising strategies for the prevention, therapy, and reversibility of CAC.

Akin to cachexia being contingent on pre-existing disease, CAC is inexorably linked to tumor presence^6–8. Although inclusion of non-cachexia-inducing tumor controls is invaluable for studies aiming to identify causative CAC initiating factors, this approach unavoidably introduces confounding variables that are tumor type-specific. We therefore first aimed to characterize the systemic metabolic networks that evolve during cancer progression and its associated body wasting in an autochthonous inducible mouse model of cancer, where non-tumor-bearing mice serve as negative controls. We then also leveraged the variability in CAC strength in the tumor-bearing mice to identify metabolic changes specific to weight loss, independent of tumor presence. We chose as a model PDA, a cancer with distinctive metabolic features where CAC prevails. We extensively profiled the metabolomes of various tissues across time to provide an unparalleled view of spatiotemporal changes that characterize the progression of pancreatic cancer development and associated body wasting. In this study, we address the following open questions in the field: 1) Can we recapture natural CAC progression that occurs in humans, using a model of PDA? 2) How early is systemic metabolism affected by tumor presence, and how do inter-tissue metabolic networks change across space and time? 3) Can we identify potential metabolites that circulate from peripheral tissues to benefit tumor growth? 4) Can we leverage our metabolic data to find early diagnostic markers and predictive indicators of weight loss?

Results

An autochthonous mouse model of pancreatic cancer cachexia recapitulates human disease progression

CAC is defined in humans as a syndrome of functional impairment that progresses through three continuous clinically relevant stages⁴ (Fig. 1a). To model this pathological state, we used a well-established, inducible, genetically engineered mouse model of PDA⁹, hereafter referred to as “iKras”. In this strain⁹, tumor induction is controlled by pancreas-specific Cre recombinase that mediates the homozygous loss of p53 while enabling doxycycline (Dox)-inducible expression of oncogenic Kras^G12D. To avoid genetic or epigenetic confounding variables introduced when using non-cachexia inducing tumors and to also control for potential Dox effects on metabolism¹⁰, we used Dox-treated iKras control (Ctrl) littermates that lack Cre recombinase, and are hence unable to initiate tumor formation.

In the iKras model, tumors are typically induced at a young 3-5 week old post-weaning age, using a Dox dose of 2 g l⁻¹ in drinking water^9,11, herein termed “high Dox”. This ensures full penetrance and provides a robust fast model of PDA with a median survival of less than 5 weeks⁹. When modeling pancreatic CAC however, we considered the limitations that pancreatic cancer is a disease of older humans¹² and that PDA and hence cachexia progress at a relatively slower pace than in iKras mice⁴. We therefore aimed to separately test and optimize each of these factors: 1) age at time of induction through use of “mature adult” mice¹³ that are 3 months of age at time of “high Dox” treatment and 2) strength of tumor induction through the administration to younger–yet “sexually mature”—5.5 week old mice¹³ of either the high Dox dose or a 5-fold lower dose (0.4 g l⁻¹), herein referred to as “low Dox” (Supplementary Fig. 1a–j). The overall patterns of weight loss were similar among all three experimental conditions (Supplementary Fig. 1b, e, h), providing confidence in our experimental model. However, there was significant variability in the extent and importantly the onset of weight loss in mature adult mice on high Dox or younger sexually mature mice on low Dox compared to high Dox conditions, indicating a decrease in penetrance of tumor induction (Supplementary Fig. 1b, e, h). This variability complicates further analyses; we thus chose to use the younger sexually mature animals on high Dox to model CAC (Fig. 1b and Supplementary Fig. 1g-i).

All mice gained overall BW (Fig. 1b and Supplementary Fig. 1h), including lean and fat mass (Fig. 1c) until days 17–18 post-Dox, when iKras started to progressively lose weight, reaching at week 4 a moribund state that necessitates euthanasia (Fig. 1b and Supplementary Fig. 1i). Based on the reproducibility of this phenotype and a robust pattern observed in over 20 independent experimental cohorts, we defined three stages of CAC in iKras mice: pre-CAC (PC) with 0% BW loss; early-CAC (EC) with <10% BW loss; and late-CAC (LC) with >10% BW loss (Fig. 1a, b and Methods). Notably, loss of lean mass preceded that of fat mass in iKras males, whereas females exhibited the opposite trend (Fig. 1c and Supplementary Fig. 1j). This is consistent with an observed earlier decrease in BW in males (Fig. 1b), and reminiscent of the sexual dimorphism reported in humans, where male patients experience higher prevalence of CAC and more pronounced muscle loss, impairment, and fatigue compared to female patients^14,15. Food intake remained stable in iKras mice until LC (Fig. 1d); this enabled us to decouple nutrition from the onset of weight loss.

As expected in CAC⁵, the levels of many cytokines including IL-6, GDF-15 and TNFα increased in EC and LC in iKras mice compared to controls (Supplementary Fig. 1k). Interestingly, IL-6 was decreased in PC, indicative of distinct signaling prior to weight loss (Supplementary Fig. 1k). Enhanced ambulatory activity was detected in iKras mice at the EC stage, particularly during the light period, when mice are expected to rest (Supplementary Fig. 1l, middle). This was accompanied by a trend in increased energy expenditure (EE) (Supplementary Fig. 1m, n, middle). Conversely, a marked decrease in EE was noted in LC stage (Supplementary Fig. 1m, n, right), when suppressed food intake (Fig. 1d) accompanied a late trend of decreased activity (Supplementary Fig. 1l, right). The respiratory exchange ratio (RER) was also lower in LC stage (Supplementary Fig. 1o, right), consistent with a prior report of a transplant model of lung CAC¹⁶, and indicative of a switch in the use of lipids rather than carbohydrates as a source of energy in iKras mice at later stages of CAC. Furthermore, CAC stages in iKras were accompanied by a progressive decline in body condition (BC) score, a reliable assessment of health status in mice¹⁷ (Fig. 1a). Thus, CAC stages in the iKras model mirror the human disease, capturing a critical period prior to weight loss (PC) that gradually progresses to a moribund state (LC)^4,18. Throughout these stages, tumor burden in iKras mice gradually increased, with larger tumor areas noted in males than females at PC stage (50% vs. 20%, Fig. 1e). The steady increase in pancreas mass (Fig. 1f) inversely correlated with a progressive decrease in the weights of peripheral tissues: inguinal and gonadal white adipose tissues (iWAT and gWAT), representing subcutaneous and visceral fat pads, respectively, brown adipose tissue (BAT), and three distinct skeletal muscles: two glycolytic (i.e., quadriceps, gastrocnemius) and one oxidative (i.e., soleus) (Fig. 1g). Histological analysis validated the pathological wasting described in metabolic tissues at all three stages of CAC (Supplementary Fig. 1p-t), demonstrating a progressive decrease in the size of muscle fibers with earlier onset in males at PC stage (Supplementary Fig. 1p, q), and a decrease in the size and number of white adipocytes, with earlier onset in females (Supplementary Fig. 1r, s).

Consistent with PDA commonly metastasizing to the liver in 90% of cases^12,19, hepatic lesions were observed in iKras mice at the advanced LC stage (Fig. 1e). At this time, liver weight was reduced, particularly in males (Fig. 1f), reflecting a degree of tissue wasting that exceeds increased mass from local metastases. Overall, the pathophysiological features of the iKras model recapture human PDA and its associated tissue wasting, providing a reliable tool to track systemic metabolic network evolution throughout the cachectic stages.

Systemic metabolism is altered in PDA tumor-bearing mice prior to body weight loss

Cachexia is a multi-organ pathological syndrome. We therefore sought to understand the metabolism systemically in the iKras model, throughout CAC progression. Using three distinct methods of liquid chromatography-mass spectrometry (LC-MS) to quantify polar metabolites and lipids, we profiled in an untargeted manner the metabolomes of 10 distinct tissues from 12 different experimental mouse cohorts: male and female, tumor-bearing iKras and non-tumor-bearing controls (Ctrls), across three stages of cachexia. Computational analyses were conducted to dissect metabolic patterns across time and space (Fig. 2a and Methods).

Fig. 2 — a Graphical abstract depicting metabolomic analysis pipeline. Created in BioRender. Xavier, J. (https://BioRender.com/d1v5zr7). b Unsupervised clustering of all HILIC-neg data. Each point represents a single independent sample; t-SNE plots are colored according to sample source (top), cachexia stage (middle), or sex (bottom). c Stacked bar plots of the number of significant metabolite alterations in iKras vs. Ctrls per stage per tissue, for each sex from all 3 LC-MS methods (HILIC-pos, HILIC-neg, and lipids C8-pos). d Volcano plots of plasma (top) and interstitial fluid (IF, bottom) samples in EC stage relative to time-matched Ctrls. All significantly altered metabolites are colored pink as indicated within the boxed legend, but only the top 5 differentially enriched or depleted metabolites are labeled when ranked first by P value and then by fold-change (FC). Comprehensive lists of ranked metabolites for all volcano plots are provided as Source Data. In c, d metabolite changes were considered significant if P < 0.05 and FC > 10% in iKras mice compared to time-matched Ctrls. Statistical significance was determined using two-tailed Student’s t test.

We first visualized the high-dimensional data: t-distributed stochastic neighbor embedding (t-SNE) demonstrated that the primary source of variation is tissue type, superseding tumor presence, stage or sex (Fig. 2b and Supplementary Fig. 2a). Thus, each tissue has a specific metabolome that remains distinct throughout its trajectory across the stages. Independent of type, skeletal muscles cluster together yet distinctly from WAT, which in turn clusters separately from BAT and other tissues (Supplementary Fig. 2b). Close clustering was observed for the liver and plasma lipidomes (Supplementary Fig. 2b, top left), consistent with a major role of the liver in the control of systemic lipid homeostasis²⁰. Principal component analysis (PCA) showed that with few exceptions (Supplementary Fig. 2c), the metabolomes of Ctrl non-tumor-bearing male and female mice are similar (Supplementary Fig. 2c, plasma HILIC-neg as representative). This indicates that although sexual dimorphism may preexist in a small number of tissues, major differences between sexes during cachexia development are the result of tumor-induced metabolic divergence.

Out of a total of 476 identified metabolites (284 polar metabolites and 192 lipids), the number of differentially altered metabolites increased with disease progression, reaching their highest total per tissue analyzed at the LC stage (Fig. 2c and Supplementary Fig. 2d). At the earliest PC stage, the greatest numbers of changes were in the pancreas and its IF, as expected from metabolic alterations resulting from Kras-driven transformation and early tumor formation¹. This is also consistent with the observation that the primary metabolic divergence in the pancreas and its IF is between tumor-bearing iKras and their Ctrls, as evident in the distinct clustering of their metabolomes by PCA (Supplementary Fig. 2e). Importantly, prior to any BW loss or tissue wasting, metabolic changes were noted in anatomical locations both close to and further away from the pancreas and its microenvironment. These include: the liver (Fig. 2c and Supplementary Fig. 2d), where metabolic alterations were detected prior to any micrometastases (Fig. 1e), suggesting tumor-induced systemic changes that may contribute to the establishment of a hepatic pro-metastatic niche¹⁹; and more distant peripheral tissues i.e., adipose and skeletal muscle (Fig. 2c and Supplementary Fig. 2d). Sexual dimorphism was also evident at the PC stage, with metabolic changes in liver and muscle being more biased towards males, and those in adipose towards females (Fig. 2c and Supplementary Fig. 2d), consistent with differential loss of whole body lean mass and fat mass in males vs. females (Fig. 1c and Supplementary Fig. 1j). Differentially altered metabolites were scarce in plasma at the PC stage but became more prominent at later EC and LC stages, pointing at potential challenges in identifying biomarkers for CAC at the earliest stages of tumor formation (Fig. 2c and Supplementary Fig. 2d). This was also evident by PCA where the primary metabolic divergence in plasma was at the LC stage (Supplementary Fig. 2e). While overall metabolomic shifts were evident in the plasma, pancreas and IF (Supplementary Fig. 2e), for all other tissues the lack of clustering of samples by type indicated that individual metabolite changes at each stage needed to be identified through more complex analyses (Supplementary Fig. 2e, iWAT as representative).

A systemic lipid signature dominates the PDA cachexia phenotype

To identify specific metabolite alterations and track their evolution through PDA growth and cachexia progression, we performed pathway analysis for each tissue at every stage. Among all tissues and stages analyzed, the top enriched polar pathways included those of nucleotides, amino acids (AAs) and redox control, particularly in the pancreas and IF, sites of tumor formation (Supplementary Fig. 2f). Enrichment of these pathways was mostly limited to later stages in plasma, liver, and muscle. By contrast, alterations in lipid classes dominated all tissues analyzed, especially di- and triglycerides (DGs and TGs, Supplementary Fig. 2g), which have been reported to decrease in the sera of cachectic pancreatic cancer patients²¹. Moreover, in most tissues, except for glycolytic skeletal muscle, a distinct signature was observed for sphingolipids and glycerophospholipids, which include phosphatidylcholines (PCs) and phosphatidylethanolamines (PEs), major constituents of cell membranes, and their catabolic intermediates lysoPCs (LPCs) and lysoPEs (LPEs) (Supplementary Fig. 2g).

Consistent with pathway analysis, hierarchical clustering showed enrichment of PEs in IF and enrichment in both IF and plasma of sphingomyelins (SMs), PCs, LPCs, and depletion in plasma of TGs with increased cholesterol esters (CEs) particularly at the LC stage (Supplementary Fig. 2h). Not only did plasma TGs anti-correlate with CEs, but also with SMs (females) and with carnitines or CARs (males), which are known to transfer long chain fatty acids into mitochondria for oxidation (Supplementary Fig. 2i). Many of these identified lipid species (SMs, PCs, LPCs, PEs, LPEs, TGs and CEs) ranked as top 5 differentially abundant or depleted in plasma and IF among all polar and non-polar metabolites across stages (Fig. 2d and Supplementary Fig. 2j, k, and Source Data).

Notably, plasma hexoses (fructose/glucose/galactose) significantly decreased in level at all stages in males and females (Supplementary Fig. 2h), ranking among the top 5 differentially altered metabolites in male EC and female EC and LC (Fig. 2d and Supplementary Fig. 2j). This is consistent with prior reports of altered glucose metabolism in cachectic patients and mouse models^22–25. Importantly, our data revealed systemic alterations in the presence of pancreatic tumors, as early as PC, with further changes throughout later stages (Fig. 2 and Supplementary Fig. 2). We therefore aimed to understand how these alterations could interact between tissues.

Systemic metabolic rewiring in PDA cachexia reveals prospective cross-tissue networks

A predominant theory posits that pancreatic tumors trigger peripheral tissue wasting, releasing nutrients and catabolic intermediates into the circulation. These can then be either taken up directly by the tumor to fuel growth, or by the liver for gluconeogenesis, where newly generated glucose is secreted back into the circulation to feed the tumor. We therefore sought to identify metabolites that may be participating in such cross-tissue metabolic networks.

Networks between “source” tissues, plasma and “sink” tissues

Assuming that metabolites released into the circulation will be detected in plasma, we considered only metabolites that are altered in the plasma of iKras relative to Ctrls. We intersected this subset with metabolites concurrently altered in the pancreas or liver, suggesting possible uptake by these “sink” tissues. We further determined whether these metabolites were also concurrently altered in “source” tissues, namely adipose or skeletal muscle. The resulting digraphs—herein referred to as “umbrella plots”—reveal potential cross-tissue metabolic networks through progressive stages of CAC (Fig. 3a, b and Supplementary Fig. 3a–f).

Fig. 3 — a, b Polar (HILIC-pos and HILIC-neg) metabolites in pre-CAC (a) and lipid metabolites in late-CAC (b) that were concurrently altered in the plasma and pancreas or in the plasma, pancreas and liver (and in peripheral tissues) in female or male tumor-bearing iKras mice vs. time-matched Ctrls, and are represented as umbrella plots. Blue/red dots indicate metabolites altered in the plasma; connecting lines indicate which of these metabolites were altered in other tissues, represented by black nodes. c Lipid metabolites that were concurrently altered in the pancreatic IF and pancreas and/or plasma in female or male pre-CAC iKras mice vs. time-matched Ctrls, and are represented as dandelion plots. Blue/red dots indicate metabolites altered in the IF; connecting lines indicate which of these metabolites were altered in the pancreas and/or plasma, represented by black nodes. In a–c metabolite alterations were considered significant if P < 0.05 and FC > 10%, determined by two-tailed Student’s t test. Color of dots/lines indicate direction of change: red, up and blue, down. In a, b the asterisk * marks metabolites of relevance later described in Fig. 5e.

For both sexes, PC presents with the smallest number of interactions, as would be expected from early tumor formation primarily leading to metabolic alterations in the tissue of origin (pancreas) and a nearby central organ (liver), where the blood from the pancreas drains through the portal vein (Fig. 3a and Supplementary Fig. 3e, f). Among the metabolites consistently depleted in male and female plasma across the stages, including PC (i.e., prior to BW loss), are hexoses, which were concurrently enriched in the pancreas with either no change (at PC) or depletion (at EC, LC) in muscle (Fig. 3a and Supplementary Fig. 3b-d).

As the cachectic stages evolve to EC and LC, the numbers of plasma interactions with “source” and “sink” tissues progressively increase. Within polar networks, polyamines and their precursor ornithine, which hold pro-tumorigenic roles and are enriched in cancers including PDA^11,26 were concomitantly elevated in plasma and pancreas of male EC (putrescine), LC (putrescine, ornithine, with spermidine only in plasma) and female LC (acisoga, product of spermidine catabolism), independent of their levels in other tissues (Supplementary Fig. 3b–d); Pantothenate, or vitamin B5, the precursor for de novo synthesis of coenzyme A, was increased in male plasma at both EC and LC stages but concurrently depleted in muscle at LC stage (Supplementary Fig. 3c, d), indicative of its decreased uptake and usage by tissues in later CAC, and consistent with a systemic state of suppressed anabolism²⁷. Cross-tissue AA networks were also prevalent in both sexes, highlighting their depletion in plasma at early stages, and enrichment in pancreas. These include N-acetylmethionine, asparagine, isoleucine, phenylalanine, serine, threonine and tryptophan (Fig. 3a and Supplementary Fig. 3a–d).

Among lipid cross-tissue networks, a prominent signature of metabolite depletion from the oxidative soleus muscle at EC stage was evident in males (Supplementary Fig. 3f, bottom), followed by depletion from all muscles at the LC stage in both sexes (Fig. 3b), with concomitant enrichment of a vast majority of lipids in plasma. Among these, SM 18 was enriched in plasma, but depleted in liver and pancreas in LC (Fig. 3b). The major exception to plasma lipid enrichment at LC stage are TGs, which were largely depleted in plasma, muscle and BAT, and consistently elevated in pancreas in both sexes, indicative of enhanced flux from peripheral tissues to the tumor (Fig. 3b).

Although we posit that these metabolites were released from distal “source” tissues and taken up by “sinks”, we do acknowledge that the direction of flow in these networks is undetermined. Moreover, many metabolites altered in the plasma and “sink” tissues were not altered in “source” tissues, even at later stages of CAC. Although this could indicate an exclusive exchange between the plasma and the liver or pancreas, it is possible that additional plasma exchanges with “source” tissues were missed because these tissues were not sampled in our study. It is also possible that these metabolites were altered in small amounts in more than one “source” tissue that we sampled. As a result, the total level in the plasma was significantly changed, but our ability to detect subtle changes in other tissues was limited.

Networks between plasma, IF and pancreas

A highly dynamic informative exchange between the pancreas and its microenvironment involves the IF, a rich source of metabolic information indicative of influx/efflux from the tumor and/or stromal cells, and a gateway for metabolites to/from the circulation. To understand how the IF participates in metabolic networks, we considered metabolites altered in iKras IF compared to time-matched Ctrls, and intersected this subset with metabolites concurrently changed in the plasma and/or pancreas. The resulting digraphs—herein termed “dandelion” plots—revealed: a- metabolites that may be involved in only local tumor/stromal interactions, b- metabolites that may be exchanged between the IF and plasma that are either derived from or excluded from the pancreas, and c- metabolites that may be moving from the circulation to the tumor via the IF (Fig. 3c and Supplementary Fig. 3g–p).

Among polar metabolites, prevalent network exchanges include hexoses (all stages) and branched chain AAs (BCAAs) isoleucine, leucine and valine, which were enriched in IF and pancreas (PC and EC stages), but depleted from plasma (EC stage), suggestive of movement of nutrients from the circulation into the tumor via the IF; additional networks include glutamine, polyamines and their precursor ornithine, which were enriched in IF, pancreas and plasma, consistent with an enhanced flux of glutamine into ornithine and polyamine synthesis in PDA¹¹ (Supplementary Fig. 3g, h, j, l, n, p). Strikingly and with very few exceptions, the vast majority of lipids, including those identified in the umbrella plots as being systemically altered (e.g., multiple species of SM 18 and CE 18), were elevated in IF, independent of stage or sex, and independent of their decrease or increase in plasma or pancreas (Fig. 3c and Supplementary Fig. 3i, k, m, o). This suggests that the IF may serve as a lipid-entrapping milieu where intake from plasma occurs at a faster rate than its release back into the circulation, allowing on-demand use by the tumor of lipids as fuel or building blocks for its growth.

Importantly, these digraphs also give us insight into the temporal changes through CAC, leveraging the longitudinal data. For example, tryptophan in females was altered in only the plasma (decreased) and IF (increased) during PC but also increased in the pancreas in EC and LC, suggesting movement to the tumor in later stages of disease (Supplementary Fig. 3g, j, l).

Overall, the connections in both umbrella and dandelion plots indicate that systemic metabolic rewiring in distal tissues—fats and muscles—could lead to changes in abundances of the same species in the circulation and consequently in the tumor, either directly or through the liver or pancreatic IF. Such spatial metabolic networks occur as early as PC and increase in complexity across time. It is noteworthy that these digraphs are hypotheses-generating, with each node and its connecting edges representing a testable hypothesis for future studies. We next experimentally validated one of the hypothesized metabolic networks.

Altered hexose metabolism in pancreatic cancer and cachexia progression

Among all quantified metabolites, hexoses were consistently differentially altered in iKras mice compared to Ctrls (Figs. 2d, 3a, Supplementary Figs. 2h, j and 3b–d, g, h, j, l, n, p), with their levels decreased in plasma and increased in IF and pancreas across the stages (Fig. 4a and Supplementary Fig. 4a, b). As glucose is the most abundant monosaccharide in the circulation, its levels are expected to mirror those of hexoses. This was confirmed using an enzymatic approach, given that specific hexoses cannot be distinguished by LC-MS (Fig. 4b and Supplementary Fig. 4c). Blood insulin levels tended to be higher in iKras mice in most stages (Supplementary Fig. 4d, e), suggestive of peripheral insulin resistance, a phenotype previously reported in CAC^28,29. Nevertheless, no significant difference was detected in systemic insulin tolerance tests (Supplementary Fig. 4f). Moreover, the iKRAS mice displayed a seemingly contradictory faster clearance of glucose following its administration compared to Ctrls, particularly at EC and LC stages (Supplementary Fig. 4g). It is noteworthy however, that these systemic responses to insulin and glucose tolerance tests (Supplementary Fig. 4f, g) are likely partly due to the tumor acting as a glucose sink.

Because lactate was unaltered in plasma, but elevated in the IF and pancreas of iKras mice (Fig. 4a and Supplementary Fig. 4a, b), this prompted us to posit that the glycolytic PDA tumors are consuming glucose while producing lactate^9,30. In addition to feeding oxidative metabolism in normal tissues^31,32, lactate can serve as substrate for gluconeogenesis under conditions of high glucose usage (e.g., in muscle under strenuous exercise) in a recycling process known as the Cori cycle. To test if this is occurring in the presence of PDA tumors, we intravenously infused ¹³C₃-lactate (Fig. 4c) into Ctrl non-tumor bearing mice and iKras mice at a pre-cachectic (PC) stage and a cachectic stage -herein termed “Cac”- where both EC and LC stages were combined. We then traced the fate of lactate into glucose. Steady-state plasma enrichment of ¹³C₃-lactate was reached within 90 min in all mice and maintained throughout the 3-hour infusion (Fig. 4d). At this endpoint, ¹³C-lactate was detected at significantly higher levels in Cac mice than PC or Ctrls (Fig. 4e and f, left). This was particularly evident in brown and white adipose tissues (Supplementary Fig. 4h, left) and in the liver (Fig. 4g, left), indicating a higher influx and/or lower consumption of lactate in these tissues at later cachexia stages. During hepatic glucose production, lactate-derived ¹³C₃-pyruvate would be expected to bypass the TCA cycle to produce 3C-labeled glucose through pyruvate carboxylase (PyC, Fig. 4h, left). Whereas in all peripheral tissues, ¹³C-hexoses were significantly lower in Cac (Fig. 4f, right), the liver displayed instead a trend for increased ¹³C-hexoses, with maintenance of 3C-labeled hexose levels (Fig. 4e, right). Given the marked depletion of total hexoses in the liver at the Cac stage (Fig. 4i, right), this amounted to ~4.8-fold higher fraction of hepatic 13C-labeled hexoses and a 3.7-fold specific increase in 3C-labeled hexose fraction in Cac mice compared to Ctrls (Fig. 4g, right). Interestingly, not only was this hepatic increase contributed by 3C-labeled hexoses, but it was equally derived from two additional 1C- and 2C-labeled hexose fractions (4–6% each, Fig. 4g, right). This indicates the contribution to gluconeogenesis of TCA cycle intermediates derived from the oxidation of 3C-labeled pyruvate and hence 2C-labeled acetyl CoA (Fig. 4h, right and Supplementary Fig. 4i) likely also generated in other tissues^31,32. The pancreas displayed a significant increase at PC stage in the levels of ¹³C-hexoses, followed by a marked drop at the Cac stage, consistent with uptake at a pre-cachectic stage, with heavier consumption by the tumor at later cachexia stages (Fig. 4e, g and i, right).

These data validate our network maps (Fig. 3 and Supplementary Fig. 3), providing evidence for the presence in cachectic iKras mice of enhanced hepatic gluconeogenesis through the Cori cycle (Fig. 4c, h), which use as a substrate circulating lactate to partially compensate for a systemic glucose deficit that can in turn benefit the PDA tumor.

Distinct metabolite patterns accompany cachexia progression

To better understand metabolite evolution during disease progression, we tracked longitudinal changes of individual metabolites by comparing each cachectic stage relative to the prior stage: i.e., LC vs. EC, EC vs. PC, PC vs. Ctrl (Fig. 5a and Supplementary Fig. 5a, b). With this temporal analysis, differentially evolving metabolite groups were identified, some of which are distinct from those identified as top altered compared to time-matched Ctrls (Fig. 2d and Supplementary Fig. 2j, k). These include the conjugated bile acids taurocholate and taurodeoxycholate, CARs and ceramides (Cers) that were altered across multiple tissues, particularly during the transition to early- and late-stage disease. This analysis also revealed metabolic patterns that depend on the tissue source. We classified these patterns into 4 categories: 1- “stage-specific”: metabolites that are only altered in one stage; 2- “fluctuating”: metabolites that are altered in more than one stage but in different directions; 3- “stable”: metabolites that maintain an alteration across at least 2 stages; 4- “progressive”: metabolites that continue to change in the same direction in more than one stage (Fig. 5b).

Fig. 5 — a 3D Stem plots where axes are fold-changes (log₂FC) comparing metabolites at each CAC stage to the prior stage in iKras mice (PC is compared to Ctrls): Each dot is a single metabolite, and the location indicates its specific position on the x and z axes. Direction and length of the stem indicate its position on the y axis. Metabolites are colored by significance along each axis; alterations were considered significant if P < 0.05 and FC > 10% in either direction as determined by two-tailed Student’s t-test. Metabolites that appear to be positioned distinctly from the overall cluster (FC > 2.5 on at least one axis) are labeled. Tissues depicted: plasma and iWAT in females (top) and males (bottom). Interactive plots to view individual metabolites are available in the Mendeley repository (see “Methods”). b Metabolite patterns classifications relative to Ctrl: Stage-specific if altered significantly (up or down) in a single stage; Fluctuating if altered in different directions in different stages; Stable if altered in the same direction and magnitude across >1 stage; Progressive if altered in the same direction and increasing magnitude in > 1 stage. c Relative proportion of metabolites falling into each of the categories from b, for each tissue and sex. d Pipeline for pre-cachexia detection analysis: a feature selection algorithm using a nearest neighbor method was used to determine the top 10 (stable or progressive) polar and lipid metabolites that distinguish cachectic (EC and LC) from Ctrl samples. A trained model using these features was then tested on PC vs. Ctrl samples. Created in BioRender. Xavier, J. (https://BioRender.com/48c29zr). e Polar (HILIC-pos and HILIC-neg) and lipid metabolites determined by the feature selection algorithm to distinguish cachectic mice from Ctrls. Female (top) and Male (bottom). f Confusion matrices depicting how the model called true and false positive (PC) samples, and true and false negative (Ctrl) samples, in females (top) and males (bottom). Overall accuracy = number of correctly called samples over total number of samples. Shown here: model considering polar and lipid metabolites together.

We quantified the proportions of metabolic alterations per tissue per sex that fall under each of these classifications. The pancreas and IF had mostly stable changes, while stage-specific alterations dominated peripheral tissues (Fig. 5c). Half of plasma altered metabolites followed stable (30%), progressive ( > 10%) or fluctuating ( < 10%) patterns. Fluctuating metabolites were less numbered overall and more prevalent in plasma and in female peripheral tissues (except quadriceps) but almost absent in pancreas, IF and liver. This analysis revealed that not only do metabolites have different patterns throughout cachexia progression depending on the tissue (Fig. 5a and Supplementary Fig. 5a-b), but tissues also display different dominant patterns of longitudinal metabolic shifts (Fig. 5b, c) that can be leveraged for early detection as described below.

Lipid and polar metabolites predict cachexia progression

A fundamental question remains: can we predict cachexia development before the manifestation of symptoms? A realistic clinical method for early detection of cachexia would employ a blood test in patients at the time of cancer diagnosis. However, at current resolution, relatively few metabolic changes are detected in plasma at the PC stage (Fig. 2c). To overcome this limitation, we used our classification of metabolite patterns (Fig. 5b, c) to select only the plasma metabolites that follow stable or progressive patterns during CAC (EC and LC stage), given that these metabolites would be more likely to be consistently detected among CAC patients and along a continuous timeline of disease progression. We used a feature selection algorithm and trained the model only on CAC and Ctrl data (Fig. 5d). The results of the model yielded a subset of polar and lipid metabolites that are predicted to be the optimal set of features that separate Ctrl from CAC (Fig. 5e and Supplementary Fig. 5c–g). Most of these metabolites were also purported to be part of cross-tissue networks (Fig. 3 and Supplementary Fig. 3), suggesting that some of the key metabolic changes in CAC plasma are not lone or arbitrary metabolites, but have functional systemic importance.

We then asked whether these subsets of metabolites and the model—trained only on CAC samples—would be capable of detecting pre-cachexia among Ctrl and PC samples, thus mimicking early detection in an asymptomatic population. When applied to test data (Fig. 5d), the model detected PC in females and in males with accuracies (88% and 83%, respectively, Fig. 5f) that are higher than when polar and lipid metabolites were considered separately (Supplementary Fig. 5h, i). Notably, while hexoses and 4 other polar metabolites and some TG species overlapped between females and males (Fig. 5e), sexual dimorphism in other predictive markers indicated the need for sex-specific diagnostics of pancreatic CAC.

Model results show similarities with human cachectic markers

Cachexia is linked to pre-existing disease. However, a significant fraction of cancer patients, including ~ 20% pancreatic, do not experience body wasting³³. To represent this patient subset and distinguish the effects of tumor growth from CAC, we took advantage of the natural variation of weight loss in our model (Fig. 1b) and separated EC stage samples ( < 10% BW loss) into “Mild Cac” or “Strong Cac” based on % BW loss that is below or above the median, respectively (Supplementary Fig. 6a). We then repeated some key metabolic analyses, including assessing the top altered metabolites across tissues and sexes between iKras tumor-bearing mice exhibiting Mild vs. Strong BW loss (Fig. 6a, Supplementary Fig. 6b and Source Data). We found that many of the top alterations overlap with those identified between iKras and Ctrl samples in several of our prior analyses (Fig. 2d and Supplementary Figs. 2h–k and 3a–p). This includes significant changes in the same direction in plasma TGs, hexoses, putrescine and SM 18 in males and in pyridoxal, proline and asparagine in females (Source Data file with overlapping metabolites bolded in spreadsheet of Fig. 6a). Importantly, we found that BW loss does not necessarily correlate with tumor size (Supplementary Figs. 6c and 7a). We thus repeated the ¹³C₃-lactate tracing analysis and found that stronger cachexia tends to associate with enhanced gluconeogenesis, with more statistically significant increases in liver ¹³C-hexoses observed in Strong Cac than Mild Cac, when compared to PC or Ctrl (Supplementary Fig. 7b–d). Thus, our data support that systemic metabolic alterations identified in our study are at least in part specifically due to weight loss. We then used a feature selection algorithm to identify metabolic alterations that specifically predict weight loss independent of tumor presence: key metabolites included amino acids and their derivatives in females and TGs in males (Fig. 6b, c and Supplementary Fig. 7e, f), leading to 90% and 80% prediction accuracy, respectively.

To validate the translational implications of our findings, we sought to compare metabolic alterations found in the circulation of iKras mice to those of pancreatic cancer patients who did or did not exhibit cachexia. Studies of cachectic humans are limited and—with only a few exceptions³⁴—generally include a low number of participants that can impact the conclusiveness of the results. Even fewer studies have reported metabolic profiling in cachectic PDA patients, with one in particular, Chakedis et al., that provided source data for independent analysis³⁵. Of the 459 metabolites detected in plasma in our mouse study, 62 were also profiled in human sera in this dataset. We found that 65% of metabolites in females and 74% in males correlate between iKras mice and weight-losing patients (Fig. 6d); this correlation is highly unlikely to be achieved by chance ( < 2% likelihood female, 0% male). The strongest correlations were also in metabolites that had the greatest fold-change (FC) values in mouse data; non-correlated metabolites may thus be in part due to underpowered detection. Of the metabolites altered in weight-losing patients compared to healthy controls, ~60% (28 in females, 29 in males) were also altered in weight-losing patients compared to weight-stable patients, indicating that the majority of observed metabolic changes in our model are attributable to weight loss, rather than mere tumor presence (Supplementary Fig. 7g).

Strikingly, some of the metabolites altered in human CAC were found in our study to be model-selected metabolites for early detection (Fig. 6d, stars), or predictors of weight loss independent of tumor presence (Fig. 6d, daggers). These include threonate, phenylalanine and hexoses/D-glucose (Fig. 5e), Cer 18:1, glutamine and lysine (Supplementary Fig. 7f) in females; pantothenate, uric acid and hexoses/D-glucose (Fig. 5e), creatine / creatinine and ketoglutarate (Supplementary Fig. 7f) in males and some TG species in both females (Fig. 5e) and males (Fig. 5e and Supplementary Fig. 7f). We can thus draw metabolic parallels between our model and patient data, including the biomarkers likely associated with CAC. Overall, these data validate both our mouse model of pancreatic CAC and prediction model, underscoring their translational significance for early detection and timely intervention.

Discussion

Physical wasting in disease has been described for centuries; it however remains incompletely understood⁵. A resurged interest in the etiology of CAC led this past decade to the identification of circulating mediators that can induce certain aspects of body wasting⁵. Translating these advances into the clinic, however, has thus far met little to no success, emphasizing the multi-leveled complexity of CAC, where a single contributing factor is insufficient to recapitulate the entire syndrome or capture its full spectrum of severity. As the majority of these studies focused on identifying causative factors or described trends in body composition changes in cancer patients^{16,21,34,36–40}, a gap in our knowledge of the systemic metabolic alterations that naturally occur throughout cancer progression has become ever apparent. We here provide a systemic analysis of the evolution of cross-tissue metabolic networks throughout progressive stages of pancreatic cancer and its associated cachexia in an autochthonous mouse model of PDA that faithfully recapitulates the human disease. Tissue-specific metabolic alterations were recently examined in a study of 15 patients with gastrointestinal cancers, including PDA, that compared weight-stable patients to those with severe >10% or less severe (5–10%) weight loss⁴¹. Some notable similarities with our findings were observed, including lipids and amino acid alterations, validating the applicability of our model. However, our model and data add an unprecedented value to the field. This includes: 1- a concurrent spatial and temporal metabolic assessment, including capturing the pre-cachectic stage, which is not possible in a retrospective human study. Not only is the added dimension of time in our study critical for a- understanding the rapidly changing landscape of PDA progression, but it is also crucial for b- our predictive analysis of CAC and c- our discovery that organ-specific metabolic alterations change across time; 2- our work experimentally validates purported metabolic networks and profiles their evolution across disease progression; 3- perhaps most relevant for health care is our equal assessment of both females and males, where we uncover significant tumor-induced metabolic divergence and sexual dimorphism, which manifest in distinct metabolic predictors for each sex. This insight will undoubtedly have significant impact on cancer disparities research and application, especially in differential diagnoses and early detection for women and men.

In profiling different organs across time, the liver in particular stands out as an early response to tumor development and CAC, with a notable number of metabolic alterations detected at the PC stage, prior to metastasis. This may be due to the anatomic proximity of the liver to the pancreas, where the tumor is growing and where dynamic exchange of circulating factors takes place through the portal vein, perhaps contributing to the formation of a pre-metastatic niche¹⁹. It would therefore be interesting to explore a connection between tissue wasting and metastatic potential.

Among the most notable evolving phenotypes throughout the cachectic stages is a marked decrease in circulating glucose and AAs detected as early as PC, accompanied by a strikingly enriched lipid signature. This draws a larger picture of a self-centric glucose-avid tumor that shunts its preferred nutrient, i.e., glucose, away from peripheral tissues, which in turn waste to provide intermediates of protein and triglyceride breakdown: i.e., AAs and glycerol for glucose production. It is noteworthy that although elevated levels of BCAAs in blood were previously reported to serve as predictors of future pancreatic cancer diagnosis⁴², this elevation was short-lived within subclinical PDA^37,42. Importantly, it was described in a genetically engineered mouse model (KP^-/-C) with a significantly longer latency and slower rate of tumor formation⁴². It is therefore likely that the rapidity of the iKras PDA model and a higher flux of AAs from the periphery towards the pancreas promotes instead the depletion of BCAAs in the circulation and their concomitant accumulation in the PDA tumor, both at PC and EC stages (Supplementary Fig. 3a, c, g, h, j, n).

We find alterations in the metabolic composition of the tumor IF of particular interest, as they infer differential rates of metabolic exchange between the tumor, its microenvironment and the circulation. These are indicative of the physiological relevance of the IF milieu where a slower exchange of metabolites with the circulation (e.g., glucose and TGs) can filter in or entrap nutrients for use by the tumor either as building blocks or for energy production. Indeed, we find that lactate, which could be derived from the tumor where it accumulates in its IF or from other organs, serves as a substrate through the Cori cycle, for hepatic gluconeogenesis, contributing to partial replenishment of glucose depletion in cachectic mice, as compared to pre-cachectic mice and non-tumor-bearing controls.

Studies investigating the etiology of CAC identified secreted factors involved in the regulation of energy balance and tissue catabolism through suppression of appetite^34,40, decreased lipid absorption³⁷, enhanced white adipose browning¹⁶, or antagonism of insulin/IGF-1 signaling³⁹. Although these conditions convey a fasting-like systemic metabolic state, the metabolic profiles of cachectic mice in our study do not completely overlap with reported profiles of acute or chronic fasting states in healthy humans^43–45, but rather underscore a cachexia-specific signature that is contingent on disease presence. For example, some TG species identified as predictors of PC in our model were also found to be depleted in the fasted state, suggesting that measuring blood after an overnight fast could lead to some false positives for cachexia prediction, compared to a non-fasted state. However, CAC-specific signatures enriched in lipids, such as the predictive marker SM 18:1 show that there are metabolic alterations specific to chronic rather than acute conditions, including the pre-cachectic trajectory. Although it remains to be investigated what roles these lipids play in cachexia, it is tempting to speculate that they contribute to host protection in disease states. For instance, in females, who generally have a higher percentage of body fat than males, cachexia is less prevalent than in males. Indeed, we find that the decrease in body fat is an earlier event in females, perhaps providing catabolic intermediates from fat depots as a cushion to prevent or delay muscle degradation, maintaining overall organismal homeostasis. In a previously described model of PDA-associated body wasting³⁷, adipose loss was found to result from exocrine insufficiency, where mice displayed lower energy expenditure at earlier stages of tumor formation with no change in food intake. Although exocrine insufficiency affects 20-60% of pancreatic cancer patients that have not undergone surgical resection⁴⁶ it does not seem to manifest at early stages in the iKras mice, as decreased EE is not detected till LC stage, when mice display suppressed food intake.

Our findings provide a resource for the systemic metabolic evolution of pancreatic cancer and its associated cachexia. They also provide a foundation for further investigation into the metabolic alterations identified in CAC and how they may be targeted for prevention or treatment. It would also be of value to investigate whether similar or distinct patterns of metabolic alterations can be detected in cachectic models of other cancers or in non-cancer models of body wasting.

Methods

Research in this study complies with the Institutional Biosafety Committee at Boston Children’s Hospital. Methods include: key resources table, method details, statistics and reproducibility.

Key Resources Table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Chemicals and Stable isotopes
D-(+)-Glucose	Sigma-Aldrich	Cat# 8270
Doxycycline hyclate (98%, HPLC)	Sigma-Aldrich	Cat# D9891-100G
Humulin R 100 Units per ml	Eli Lilly	NDC-0002-8215-01
IHC Zinc Fixative	BD Biosciences	Cat# 550523
Sodium L-lactate (¹³C₃, 98%)	Cambridge Isotopes	Cat# CLM-1579-N-0.1MG
Reagents
FisherBrand Heparinized Micro-Hematocrit Capillary Tubes	Fisher Scientific	Cat# 22-362-566
Microvette 500 EDTA K3E	SARSTEDT	Cat# 20.1341.102
Spectra/Mesh Screen Disks, 47 mm, Nylon, 20 µm	Repligen	Cat# 148134
Critical Commercial assays
Proteome Profiler Mouse XL Cytokine Array	R&D Systems, by bio-techne	Cat# ARY028
CONTOUR NEXT EZ Blood Glucose Meter (Glucometer)	Bayer Ascensia Diabetes Care	Model# 7160 P
CONTOUR NEXT Blood Glucose Test Strips	Ascensia	Cat# 7278
Infinity^TM Glucose Oxidase Liquid Stable Reagent	Thermo Scientific	Cat# TR15221
General Chemistry Calibrators	Pointe Scientific	Cat# C7506-50
Ultra Sensitive Mouse Insulin ELISA kit	Crystal Chem	Cat# 90080
Deposited data
Untargeted Metabolite Profiling, 3 methods: HILIC-neg, HILIC-pos and C8-pos	This paper	Metabolomics Workbench Project ID PR001802
Experimental models: organisms/strains
Mouse strain, FVB background: p48-Cre;ROSA_rtTA;tetO_LSL-Kras^G12D; p53^fl/fl	Ying et al.⁹	Acquired from Alec Kimmelman lab
Software and algorithms
Adobe Illustrator v. 24.3	https://www.adobe.com	N/A
Adobe Photoshop	https://www.adobe.com	N/A
BioRender	@BioRender	https://www.biorender.com
CalR v. 1.3	Amir et al.⁴⁷	http://calrapp.org
Metaboanalyst 5.0	Pang et al.^48,49	https://www.metaboanalyst.ca
NDP.view 2	Hamamatsu.com	https://nanozoomer.hamamatsu.com
GraphPad Prism v. 10	GraphPad Software	https://www.graphpad.com/ scientific-software/prism/
R	R Development Core Team⁵⁰	http://www.r-project.org/
MATLAB v. 2020b	MathWorks	https://www.mathworks.com/products/matlab.html
Other
Clinical Manual System Microscope	Olympus	BX43
DEXA Scan Lunar PIXImus2	GE Lunar Corp.	N/A
SpectraMax Multimode Microplate-Reader	Molecular Devices	M3
Multi-Syringe Infusion Syringe Pump	KD Scientific	KDS-230

Open in a new tab

Method details

Animal work

All animal studies and procedures were approved by and performed in compliance with the Institutional Animal Care and Use Committee at Boston Children’s Hospital. Tumor burden was assessed by weighing the pancreas following euthanasia and tissue harvest at each CAC stage (PC, EC and LC); the total maximal tumor burden permitted (15% body weight) was not exceeded in this work, even at the LC stage. Humane experimental endpoints were set such that each mouse did not lose more than 20% body weight compared to its weight at day 0 when Dox was administered for tumor initiation. Euthanasia was performed earlier than expected if a mouse displayed signs of pain or distress. For all experiments, the Dox-inducible genetically engineered mouse model of PDA⁹, herein termed iKras, was used (p48-Cre;ROSA_rtTA;tetO_LSL-Kras^G12D; p53^fl/fl) in FVB background. Mice of the same strain and genotype, frequently iKras littermates, that however lack Cre expression were used as controls (Ctrls). All mice were fed ad libitum a chow diet (LabDiet, Prolab IsoPro RMH 3000 #5P76, 3.44 kcal g⁻¹ physiological fuel value) and were maintained on a 14:10 light/dark cycle at a housing temperature of 71 °F and 35–70% humidity. Unless otherwise indicated, both iKras (p48-Cre^+/-) and Ctrl (p48-Cre^-/-) young, sexually mature mice of average age 5.5 weeks were treated until the experimental endpoint, with a “high Dox” dose of 2 g l⁻¹ in the drinking water. When indicated, a “low Dox” dose of 0.4 g l⁻¹ was used instead to treat these young sexually mature mice. Alternatively, and where stated, “older mice” of average age 12.4 weeks were administered the “high Dox” dose, until the experimental endpoint. In all experiments, Dox water was changed twice per week. Mice were assigned randomly to experimental groups according to genotype, and studies were not blinded. Age- and sex-matched Ctrls, and whenever possible, littermates were used for each experiment.

Assessment of body weight change and definition of cachexia stages

Body weight (BW) change was monitored across time (twice per week) and assessed with reference to the initial BW of iKras or Ctrl mice at Day 0 of Dox administration. Percent BW change was assessed using the highest BW reached as a reference for each mouse. Three stages of CAC were defined in iKras mice following tumor induction, starting one week post-Dox administration: Mice were considered at PC stage or “pre-cachexia” with 0% BW loss (days 7–13), at EC or “early-cachexia” with <10% BW loss (days 14-20), and at LC or “late-cachexia” with >10% BW loss (days 21 or after). These cutoffs were determined empirically: with the rapidity of weight loss in the iKras model compared to humans, the stages of CAC were binned using a 10% rather than 5% cutoff.

Food intake

Mice were housed as 2–3 per cage. Average daily food intake per mouse (g of food per mouse per day) was assessed weekly and at endpoint by weighing the food provided and subtracting the weight of the remaining food in the cage.

Body condition score

The BC score was estimated in mice across the cachectic stages, according to Ullman-Cullere and Foltz, 1999, where a score paradigm of 1–5 is followed¹⁷. A BC score of 5 (obese) indicates the mouse is smooth and bulky and bone structure disappears under flesh and subcutaneous fat; a score of 4 (over-conditioned) indicates the spine is a continuous column and vertebrae are palpable only with firm pressure. A BC score of 3 (well-conditioned) indicates the vertebrae and dorsal pelvis are not prominent, but palpable with slight pressure; a BC score of 2 (under-conditioned) indicates the vertebral column segmentation is evident and pelvic bones readily palpable; and a BC score of 1 (emaciated) indicates the skeletal structure is highly prominent and vertebrae are distinctly segmented.

Body composition measurement

Total lean mass and fat mass were obtained using DEXA (dual-energy X-ray absorptiometry; Lunar PIXImus2, GE Lunar Corp) at timepoints prior to (Day 0, initial) and subsequent to Dox administration. Absolute values were normalized to initial values to calculate percent change relative to Day 0.

Metabolic Cages

Mice were singly housed in metabolic cages in OxymaxR- CLAMS (Comprehensive Lab Animal Monitoring System) at thermoneutrality (28–30 °C). Oxygen consumption (VO₂) and carbon dioxide (CO₂) production rates (ml hour⁻¹) for individual mice were recorded. Respiratory exchange ratio (RER) was calculated as: RER = VCO₂ / VO₂. Energy Expenditure (kcal hour⁻¹) was calculated using indirect calorimetry as: EE = 3.815 × VO₂ + 1.232 × VCO₂. Mouse activity was measured using beam break counts, recorded by infrared light sources and optical sensors. Ambulatory activity was assessed by movement, from one point to another, of beam breaks on both the X and Y axes. Measurements (1 every 24 min) were recorded throughout periods of 2-light, 3-dark cycles for PC stage and 3-light, 4-dark cycles for EC and LC stages. EE regression analysis was plotted to represent kcal hour⁻¹ vs. BW in grams of each mouse at the beginning of each period (PC, EC and LC). Mice were maintained in the same room at the same temperature when not recording between stages. All data were analyzed and plotted using CalR⁴⁷ version 1.3.

Insulin tolerance test (ITT) and glucose tolerance test (GTT)

Two separate cohorts of mice dedicated for either GTT or ITT were treated with 0.4 g l⁻¹ doxycycline (low Dox). Low Dox rather than high Dox dose was used to expand the timeline of each CAC stage (PC, EC and LC), allowing each mouse to fully recover from an overnight fast before fasting and testing again at the subsequent stage. For GTT, mice were fasted overnight (16 hours) before receiving an intraperitoneal injection of D-glucose (0.25 g ml⁻¹ in 0.9% saline) at 1 g kg⁻¹ BW. At timepoints 0, 15, 30, 45, 60, and 120 min post-injection, 5–10 µl of blood was drawn from the tail and blood glucose levels were assayed using a Glucometer. A similar procedure was performed on the ITT mouse cohort, except that insulin (Humulin R, 0.1875 U ml⁻¹ in saline) rather than glucose was injected at 0.75 U kg⁻¹ BW.

Necropsy–collection of blood, different tissues and pancreatic interstitial fluid

Mice were euthanized at the start of the light cycle. Humane experimental endpoints were chosen prior to mice reaching a moribund state. With the exception of ¹³C₃-lactate infused mice (section below), blood was drawn from the tail vein of non-anesthetized mice (for all metabolomic profiling to avoid any potential anesthesia-related metabolic effects⁵¹), or retro-orbitally from anesthetized mice (for non-LC-MS analyses); then plasma was prepared, snap-frozen on dry ice and stored at −80 °C. For all tissues except pancreas (i.e., liver, adipose and skeletal muscle): samples were collected, weighed and either snap-frozen in liquid nitrogen for future metabolic analysis by LC-MS, or fixed in formalin or Zinc Fixative (adipose tissues) for later processing and histological assessment. The “n” sample number for tissue weights in Fig. 1f, g were as follows: f pancreas iKras, female: n = 10, 12, 18; male: n = 10, 11, 23 for PC, EC, LC, respectively, with corresponding Ctrls, female: n = 13, 14, 18; male: n = 13, 9, 17. Liver iKras, female: n = 9, 10, 9; male: n = 6, 6, 14 for PC, EC, LC respectively, with corresponding Ctrls, female: n = 6, 7, 8; male: n = 5, 4, 7. g inguinal white adipose tissue (iWAT; iKras female PC, EC, LC n = 9, 10, 17; Ctrl PC, EC, LC n = 6, 7, 14; iKras male PC, EC, LC n = 6, 5, 20; Ctrl PC, EC, LC n = 5, 4, 13), gonadal WAT (gWAT; iKras female PC, EC, LC n 10 = 9, 10, 15; Ctrl PC, EC, LC n = 6, 6, 14; iKras male PC, EC, LC n = 6, 6, 20; Ctrl PC, EC, LC n = 5, 4, 13), brown adipose tissue (BAT; iKras female PC, EC, LC n = 9, 10, 17; Ctrl PC, EC, LC n = 6, 7, 15; iKras male PC, EC, LC n = 9, 10, 21; Ctrl PC, EC, LC n = 5, 4, 13) and 3 types of skeletal muscles (quadriceps, gastrocnemius and soleus; iKras female PC, EC, LC n = 9, 10, 21; Ctrl PC, EC, LC n = 6, 7, 18 for each muscle type; iKras male PC, EC, LC n = 9, 10, 23; Ctrl PC, EC, LC n = 5, 4, 14 for quadriceps, iKras male PC, EC, LC n = 6, 6, 22; Ctrl PC, EC, LC n = 5, 4, 14 for gastrocnemius, iKras male PC, EC, LC n = 6, 6, 21; Ctrl PC, EC, LC n = 5, 4, 15 for soleus).

For pancreas, the interstitial fluid (IF) was collected first as previously described⁵²; whether tumor-devoid (Ctrl mice) or tumor-burdened (iKras mice), the pancreas was quickly dissected and rinsed in saline, blotted on filter paper (Whatman, 1001) and placed on top of a 20 μm nylon filter (Spectrum Labs, 148134), centrifuged (10 min, 200 g, 4 °C), then snap-frozen in liquid nitrogen. Pancreatic / tumor tissue from which the IF was extracted was then immediately frozen in liquid nitrogen. As needed, samples were pooled to obtain sufficient volume for LC-MS analysis: all Ctrl pancreatic IF samples were pooled from 3–5 mice each; a smaller number of iKras pancreatic IF–mostly from PC and less from EC stage—were pooled from 2-3 mice each. A third of plasma samples were pooled from 2 mice each. The rest was processed without pooling.

Plasma glucose and insulin measurements

Non-fasting plasma glucose levels were measured using a colorimetric glucose oxidase reagent (Infinity^TM, Thermo Scientific, TR15221) and a calibrator for standards (Pointe Scientific, C7506-50). Non-fasting and fasting insulin levels were quantified using the Ultra Sensitive Mouse Insulin ELISA kit (Crystal Chem, Cat# 90080).

Histopathology

Following an overnight fixation in either formalin (pancreas, liver, skeletal muscles) or Zinc Fixative 1X (iWAT, gWAT, BAT), tissues were placed in 70% ethanol or PBS, respectively, for paraffin embedding, sectioning and staining with Hematoxylin and eosin (H&E). Whole section imaging was performed with a NanoZoomer-2.0 HT C9600 digital scanner (Hamamatsu) coupled to a mercury lamp unit L11600-05 and using NDP.scan 3.4 software U10074-03 (Hamamatsu). Images were acquired using the 20X objective of the scanner, where one pixel corresponds to 0.46 μm. NDP.view2 (v.2.9.29, Hamamatsu) software was used to generate snapshots from representative areas at 40X. To enable accurate identification of cellular features over background, brightness and contrast were equally adjusted in Photoshop (−35% each) for all white adipose (iWAT and gWAT) images at all stages, and for brown adipose (BAT) at EC and LC stages; for gastrocnemius, brightness only was adjusted at LC stage ( + 35%). A pathologist semi-quantitatively assessed in a blinded fashion the percent tumor area in the pancreas at each stage. In parallel, the presence or absence of metastatic lesions in the liver was assessed using a scoring system that ranged from none (for no lesions observed) to low, moderate, or high.

Infusion of ¹³C-labeled lactate

Anesthetized mice were infused through the jugular vein^53,54 with ¹³C₃-lactate (Cambridge Isotope Laboratories, CLM-1579-N-0.1MG) as a continuous infusion of 0.005 mg g⁻¹ min⁻¹ for 180 min. Tail blood was collected at 0, 15, 30, 90 and 180 min post-infusion and used to quantify plasma enrichment of labeled nutrients by LC-MS. Mice were euthanized after a 3 h-infusion and tissues were rapidly collected, weighed and snap-frozen in liquid nitrogen for processing and metabolite quantification using LC-MS Method 2 (HILIC-neg), as described in the specified section below. All ¹³C-isotopologues were corrected for natural abundance and normalized by protein concentration. They are plotted in figures as “relative abundance”, where each metabolite is compared among the different conditions. The relative abundance of total pools indicates the sum of all unlabeled and ¹³C-labeled isotopologues of a specific metabolite, corrected for natural abundance, that was then normalized to protein concentration. Plots for “% labeled” metabolites indicate the percent of ¹³C-labeled isotopologues of a specific metabolite out of its own total pool. Detailed legend for Fig. 4h representing a diagram of gluconeogenesis from lactate: During hepatic glucose production, pyruvate is shunted away from the tricarboxylic (TCA) cycle (i.e., from pyruvate dehydrogenase, PDH) towards pyruvate carboxylase (PyC) for direct production of oxaloacetate (OAA), substrate for synthesis of the gluconeogenic precursor phosphoenolpyruvate (PEP) by pyruvate carboxykinase (PCK). In this process, ¹³C₃-pyruvate derived from ¹³C₃-lactate through activity of lactate dehydrogenase (LDH) would lead to the production of 3C-labeled glucose (m + 3). However, in liver and other tissues, OAA can also be produced in the TCA cycle following activity of PDH on ¹³C₃-pyruvate, generating 2C-labeled acetyl CoA; after the first turn of the TCA cycle, two distinct isotopomers of OAA are produced with equal likelihood, each carrying two ¹³C labeled carbons, located at either end of the molecule. Subsequent activity by PCK results in the equal probability of producing either 1C-labeled or 2C-labeled PEP, in turn leading to 50:50 fractions of 2-carbon-labeled (m + 2) or one-carbon-labeled glucose (m + 1).

Liquid chromatography-mass spectrometry analyses

A combination of three LC-MS methods was used to profile metabolites; two methods that measure polar metabolites (HILIC-pos and HILIC-neg), and a lipid profiling method (C8-pos). Data were acquired using LC-MS systems comprised of Nexera X2 U-HPLC systems (Shimadzu Scientific Instruments; Marlborough, MA) coupled to Exactive series orbitrap mass spectrometers (Thermo Fisher Scientific; Waltham, MA). The method details are summarized below.

LC-MS method 1–hydrophilic interaction liquid chromatography (HILIC)-pos

Positive ion mode MS analyses of polar metabolites: LC-MS samples were prepared from 10 µl of plasma or aqueous tissue homogenate suspension via protein precipitation using 90 µl of 74.9: 24.9: 0.2 v/v/v acetonitrile/methanol/formic acid containing stable isotope-labeled internal standards (valine-d8, Isotec; and phenylalanine-d8, Cambridge Isotope Laboratories; Andover, MA). The samples were centrifuged (10 min, 10,000 × g, 4 °C), and the supernatants injected directly onto a 150 × 2 mm Atlantis HILIC column (Waters; Milford, MA). The column was eluted isocratically at a flow rate of 250 µl min⁻¹ with 5% mobile phase A (10 mM ammonium formate and 0.1% formic acid in water) for 1 minute followed by a linear gradient to 40% mobile phase B (acetonitrile with 0.1% formic acid) over 10 min. MS analyses were carried out using electrospray ionization in the positive ion mode using full scan analysis over m/z 70–800 at 70,000 resolution and 3 Hz data acquisition rate. Additional MS settings were: ion spray voltage, 3.5 kV; capillary temperature, 350 °C; probe heater temperature, 300 °C; sheath gas, 40; auxiliary gas, 10; and S-lens RF level 40.

LC-MS Method 2–hydrophilic interaction liquid chromatography (HILIC)-neg

Negative ion mode MS analysis of polar metabolites: LC-MS samples were prepared from 30 µl of plasma or aqueous tissue homogenate suspension via protein precipitation using 120 µl of 80% methanol containing inosine-¹⁵N₄, thymine-d4 and glycocholate-d4 internal standards (Cambridge Isotope Laboratories; Andover, MA). The samples were centrifuged (10 min, 10,000 × g, 4 °C) and the supernatants were injected directly onto a 150 × 2.0 mm Luna NH2 column (Phenomenex; Torrance, CA). The column was eluted at a flow rate of 400 µl min⁻¹ with initial conditions of 10% mobile phase A (20 mM ammonium acetate and 20 mM ammonium hydroxide in water) and 90% mobile phase B (10 mM ammonium hydroxide in 75:25 v/v acetonitrile/methanol) followed by a 10 min linear gradient to 100% mobile phase A. MS analyses were carried out using electrospray ionization in the negative ion mode using full scan analysis over m/z 60–750 at 70,000 resolution and 3 Hz data acquisition rate. Additional MS settings were: ion spray voltage, −3.0 kV; capillary temperature, 350 °C; probe heater temperature, 325 °C; sheath gas, 55; auxiliary gas, 10; and S-lens RF level 50.

LC-MS method 3 – C8-pos

Lipids (polar and nonpolar) were extracted from 10 µl of plasma or aqueous tissue homogenate suspension using 190 µl of isopropanol containing 1-dodecanoyl-2-tridecanoyl-sn-glycero-3-phosphocholine as an internal standard (Avanti Polar Lipids; Alabaster, AL). After centrifugation (10 min, 10,000 × g, ambient temperature), supernatants (10 µl) were injected directly onto a 100 × 2.1 mm ACQUITY BEH C8 column (1.7 µm; Waters; Milford, MA). The column was eluted at a flow rate of 450 µl min⁻¹ isocratically for 1 minute at 80% mobile phase A (95:5:0.1 vol/vol/vol 10 mM ammonium acetate/methanol/acetic acid), followed by a linear gradient to 80% mobile-phase B (99.9:0.1 vol/vol methanol/acetic acid) over 2 minutes, a linear gradient to 100% mobile phase B over 7 minutes, and then 3 minutes at 100% mobile-phase B. MS analyses were carried out using electrospray ionization in the positive ion mode using full scan analysis over m/z 220-1100 at 70,000 resolution and 3 Hz data acquisition rate. Additional MS settings were: ion spray voltage, 3.0 kV; capillary temperature, 300 °C; probe heater temperature, 300 °C; sheath gas, 50; auxiliary gas, 15; and S-lens RF level 60.

Metabolomics quality control and data processing

Samples were randomized prior to LC-MS analyses. Data quality was assured by monitoring internal standard performance in every sample during data acquisition and by regular analysis of pooled quality control (QC) samples that were prepared by combining aliquots from each study sample. Pairs of pooled QC samples were analyzed after intervals of ~20 study samples. One pooled QC sample from each pair was used to correct for MS sensitivity drift across the run using “nearest-neighbor” scaling. The other pooled QC was used to calculate coefficients of variation (CVs) for every metabolite and unknown as a measure of analytical precision for the feature. Raw data were processed using TraceFinder 3.3 software (Thermo Fisher Scientific; Waltham, MA) and Progenesis QI (Nonlinear Dynamics; Newcastle upon Tyne, UK). Metabolite identities in the HILIC-pos and HILIC-neg methods were confirmed using authentic reference standards. Lipids were identified using representative standards from each lipid class.

Metabolomics data analysis

Only known annotated metabolites were considered. After adjusting for pooling and quality control, the final number of samples analyzed were as follows: 3 Ctrl and 6 iKras for each stage, sex, and method in pancreas, liver, BAT, gWAT, iWAT, quadriceps, and gastrocnemius; 3 PC Ctrl, 5 PC iKras, 4 EC Ctrl, 4 EC iKras, 4 LC Ctrl, 4 LC iKras for soleus female C8-pos lipidomics; 4 PC Ctrl, 4 PC iKras, 4 EC Ctrl, 4 EC iKras, 5 LC Ctrl, 5 LC iKras for soleus male C8-pos lipidomics; 3 PC Ctrl, 6 PC iKras, 3 EC Ctrl, 5 EC iKras, 3 LC Ctrl, 6 LC iKras for soleus female and male HILIC-pos; 3 PC Ctrl, 5 PC iKras, 5 EC Ctrl, 4 EC iKras, 5 LC Ctrl, 4 LC iKras for soleus female HILIC-neg; 4 PC Ctrl, 5 PC iKras, 5 EC Ctrl, 1 EC iKras, 6 LC Ctrl, 4 LC iKras for soleus male HILIC-neg; 10 PC Ctrl, 6 PC iKras, 16 EC Ctrl, 9 EC iKras, 16 LC Ctrl, 6 LC iKras for plasma female all methods, 14 PC Ctrl, 10 PC iKras, 18 EC Ctrl, 10 EC iKras, 17 LC Ctrl, 13 LC iKras for plasma male all methods, 4 PC Ctrl, 10 PC iKras, 4 EC Ctrl, 12 EC iKras, 4 LC Ctrl, 10 LC iKras for IF female all methods, 4 PC Ctrl, 10 PC iKras, 6 EC Ctrl, 13 EC iKras, 5 LC Ctrl, 12 LC iKras for IF male all methods.

Unless otherwise specified, metabolite alterations when comparing two groups were considered significant if P < 0.05 using a two-tailed two-sample t test.

Unsupervised analyses

All three metabolomics methods used in this study were considered separately. Data from each tissue were combined, considering the metabolites that were detected in common among all tissues. Both principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE) clustering were performed, and data were labeled by tissue source, sex, or stage. In addition to original code, the following MATLAB file exchange was used to plot ellipses: “Plot ellipse on scattered 2D data”, Version 1.2.4 by Zoltan Gillay.

Supervised analyses

The numbers of metabolic alterations per tissue per sex per stage relative to time-matched control were tabulated. Lipids and polar metabolites were counted together (Fig. 2c) as well as separately (Supplementary Fig. 2d). Individual metabolite alterations were graphed as volcano plots with lipids and polar metabolites combined (Figs. 2d, 6a and Supplementary Figs. 2j, k and 6b). Each tissue, sex, and stage was considered separately. The top 5 differentially enriched or depleted metabolites were labeled when ranked first by P-value and then by FC. Associated Source Data file lists p-values, FC, and FDR-adjusted q-values for each comparison. To detect correlation between plasma metabolites at EC stage for each sex (Supplementary Fig. 2i), lipids and polar metabolites were combined. FC relative to time-matched Ctrl for each metabolite was calculated. Metabolites with a false discovery rate <0.05 were selected. The pairwise linear correlation coefficient between each pair of metabolite FCs was calculated and plotted as a scaled color-map. Select groups of metabolic classes, which had positive or negative correlations, were labeled.

Pathway analyses

Each tissue, stage, and sex was considered separately. Polar metabolites—HILIC-pos and HILIC-neg—were processed by Metaboanalyst 5.0, with “Globaltest” pathway enrichment analysis method. Resulting pathway P values and false discovery rates were graphed. To summarize the pathway analyses across all tissues, all pathways that passed FDR < 0.05 for maximum stringency were selected. Pathways were tabulated according to the number of times altered (across all tissues/stages/sexes). The top 25 pathways were selected: plot indicates whether each of these pathways was altered in a given tissue, stage, and sex. Many lipid species in our data were not matched to compounds in the Metaboanalyst library. Therefore, a custom MATLAB script was created to separately analyze lipid metabolites with overrepresentation analysis: first, metabolites were categorized according to lipid class. A hypergeometric probability density function was used to determine whether the number of altered metabolites (relative to time-matched Ctrl) in a given class was greater than chance. Classes that were overrepresented with P < 0.05 in a given tissue/stage/sex were plotted.

All digraphs (Umbrella and dandelion plots)

For each stage, lipids and polar metabolic networks were analyzed separately. Females and males were also analyzed separately. For a more stringent analysis, a metabolite was considered to be “altered” if its P value was <0.05 AND its FC was > 10% in either direction compared to time-matched controls. For brevity, lipids were grouped according to their number of carbon atoms, disregarding the number and placement of double bonds (e.g., TG 51:2 was encoded as TG 51). If more than one lipid belonged to a specific grouping, the mode of the FC directions—i.e., the direction of change for the majority of lipid subspecies in that grouping—was used. All plots were created using custom MATLAB scripts. Different digraph types, i.e., umbrella and dandelion plots are described in further detail below.

Umbrella plots

For each stage, metabolites significantly altered in the plasma in iKras vs. Ctrl were selected. Of these, metabolites that were also altered in the pancreas or liver (herein termed “sink” tissues) in iKras vs. Ctrl were sub-selected. Metabolites that were significantly altered in any muscle or fat (herein termed “source” tissues) in iKras vs. Ctrl and that intersected with this sub-selected group were also plotted. Our definition of “source” and “sink” is not based on arterial or venous blood (blood from tail vein was drawn for all mice); rather, these terms refer to computational predictions where “source” is a tissue that could release nutrients into circulation, and “sink” a tissue that could uptake and process these nutrients for the benefit of the tumor. To generate umbrella plots (Fig. 3a, b and Supplementary Fig. 3a–f), data from each tissue were encoded in an adjacency matrix, where a “1” indicated a common alteration between plasma and another tissue. The matrix was graphed as a digraph with a “layered” layout. Nodes (dots) are plasma metabolites and tissues. Edges (connecting lines) are alterations of these metabolites in other tissues. The digraph was colored according to the direction of change relative to Ctrl: metabolite nodes are colored according to the direction of change in the plasma. Edges connecting a metabolite to a tissue are colored according to direction of change of that metabolite in that tissue.

Dandelion plots

For each stage, metabolites significantly altered in the IF in iKras vs. Ctrl were selected. Of these, metabolites that were also altered in the pancreas or plasma in iKras vs. Ctrl were sub-selected. To generate dandelion plots (Fig. 3c and Supplementary Fig. 3g-p), data from each of these 3 tissues were encoded in an adjacency matrix, where a “1” indicated a common alteration between IF and pancreas and/or plasma. The matrix was graphed as a digraph with a “force” layout. Nodes (dots) are IF metabolites, pancreas, and plasma. Edges (connecting lines) are alterations of the sub-selected metabolites in the pancreas and/or plasma. The digraph was colored according to the direction of change relative to Ctrl: metabolite nodes are colored according to the direction of change in the IF. Edges connecting a metabolite to a tissue are colored according to direction of change of that metabolite in that tissue.

3D stem plots

Each tissue and sex was analyzed separately. Metabolite levels were compared across 3 dimensions: LC vs. EC, EC vs. PC, and PC vs. Ctrl. Each FC for each metabolite was plotted using the “stem3” MATLAB function: dots denote position on the X and Z axes, and stems denote position on the Y axis. Metabolites were colored according to which axes the FC passed the significance threshold.

Prediction models

Females and males were analyzed separately. Lipid and polar metabolites were considered in separate models as well as together in a single model. Only metabolites altered in the plasma were considered, sub-selecting metabolites that followed a stable or progressive pattern when considering just EC and LC stages. The ReliefF feature selection algorithm using a nearest-neighbor method was used to rank predictors. We adjusted the number of neighbors in accordance with n at stage 1 as follows: 0 <n < 50 #neighbors = 10, 50 ≤n ≤ 70 #neighbors = 20; n > 70 #neighbors = 30. The top 10 predictors by weight were used to train a generalized linear model with a binomial distribution, using data from EC and LC with time-matched controls. This reduced the number of features to 10, less than the number of samples (47 female 58 male). The resulting model and data from only metabolites selected was then used to predict sample type in test data: Ctrl vs. PC (thus, PC samples were not used in the pipeline prior to testing). We then calculated the true positive, true negative, false positive, false negative, and overall accuracy from the predictions. Confidence intervals of prediction accuracy were calculated using the Clopper-Pearson method, fitting a binomial distribution to predicted values. Model training and predictions were performed using custom MATLAB scripts.

Assessment of mild versus strong cachectic phenotype

To understand the metabolic alterations specifically attributable to weight-loss, the natural variation in our model was used to classify mice in EC stage as exhibiting “Mild” or “Strong” cachexia: below or above the median weight loss within EC stage, respectively. This was decoupled from tumor size, allowing specific analysis of metabolic alterations due to weight loss using volcano plots as described above. Associated Source Data file (for Figs. 2d, 6a and Supplementary Figs. 2j, k, 6b) lists p-values, FC, and FDR-adjusted q-values for each comparison. Analysis of ¹³C-hexoses in Mild or Strong cachectic mice was repeated as above. The prediction model as described above was modified to detect predict Mild vs. Strong cachectic trajectories. To overcome the sample size limitation, the leave-out method was employed, where for each of 50 iterations one sample was randomly selected as the test sample. The model was trained on the remaining samples (as above, the nearest-neighbor method was used to select top metabolites, these subsequently were used to fit a generalized linear model). The resulting model was tested on the “left out” sample. This sample was then “replaced” before the next iteration. Feature selection was thus performed within each resampling loop. Since each iteration yielded a slightly different model due to a slightly different subset of training data, the metabolites that were chosen in every iteration were plotted: logFC Strong/Mild weight loss.

Comparison to human data

Publicly available human serum metabolomics data³⁵ were compared to our own generated mouse plasma data. Data for females and males were analyzed separately. First, overlapping metabolites between the two datasets were selected, based on PubChem CID (a unique integer identifier for a specific chemical structure in the PubChem database), resulting in a total of 62 metabolites. Human samples from weight-losing pancreatic cancer patients vs. tumor-free healthy controls, as well as weight-losing vs. weight-stable pancreatic cancer patients, were compared to each stage vs. Ctrl in our mouse plasma data. Metabolic alterations were binarized, and correlated metabolites (changing in the same direction) between human and mouse data (from at least one stage) were selected. Correlated metabolites were plotted as Venn diagrams (Supplementary Fig. 7g), separately considering those that were increased or decreased in serum (human data) or plasma (our mouse data). For each metabolite, the FC of metabolite level in weight-losing compared to weight-stable pancreatic cancer patients was plotted in Fig. 6d, as well as the FC in iKras mouse plasma compared to time-matched Ctrls. In the case that the metabolite was altered in more than one stage in mouse plasma, the stage with the greatest FC absolute value was plotted. The statistical significance of the correlation was tested by chi-square analysis. Samples were then randomly shuffled, and the correlation percentage was recalculated: this was repeated for 50 iterations in order to calculate the likelihood that this correlation was achieved by chance.

Statistics and reproducibility

Data in all time-course graphs represent the mean ± s.e.m or the mean ± s.d, as indicated in the figure legends; P-values were obtained by two-way ANOVA followed by Šídák’s or —where indicated— Tukey’s multiple comparisons post-hoc test. For all dot-plots and bar graphs, non-parametric Mann-Whitney test was performed. For survival analyses, Kaplan-Meier was performed, followed by one-sided chi-square log-rank (Mantel-Cox) test. For all statistical tests, *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001. All figures represent two to seven independent experiments where data were combined, as indicated in the figure legends, and n represents the number of independent biological replicates. Multiple hypothesis correction was not performed unless stated otherwise for specific figures, to be comprehensive in our exploratory analyses. This allowed us to capture biologically relevant changes, including those that correlated with human studies, given the power of our sample size. No statistical method was used to predetermine sample size. No data were excluded from the analyses. Mouse experiments as well as LC-MS samples were randomized. Except for the pathologist (see Histopathology section), the investigators were not blinded to allocation during experiments and outcome assessment.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Supplementary information

Supplementary Information^{(13.9MB, pdf)}

Reporting Summary^{(81.4KB, pdf)}

Transparent Peer Review file^{(217.4KB, pdf)}

Source data

Source Data - NEW^{(4.1MB, xlsx)}

Acknowledgements

We thank members of the N.Y.K. Laboratory, the Division of Endocrinology at Boston Children’s Hospital and S. Biddinger for helpful comments and discussions.

Author contributions

B.M. designed and performed all wet-lab experiments. D.M. designed and performed all computational analyses. M.-S.L., with help from B.M., performed the in vivo infusion and analyzed tracing data. I.N. assisted with mouse work. K.C.S. helped with mouse work, performed the plasma glucose and cytokine assays and together with B.M. prepared the manuscript tissue images. C.D., L.D. and S.J. analyzed and quantified all metabolite data, HILIC-pos, HILIC-neg, and C8-pos, respectively. C.B.C. supervised and discussed the metabolite quantification analyses. M.M.-K. completed pathological assessment and scoring. D.M. conceived, designed and performed the prediction model analysis. J.B.X. discussed and provided feedback on all computational work. N.Y.K. conceived, designed and supervised the study and obtained research funding for B.M., M.-S.L., K.C.S., I.N. and N.Y.K. The manuscript was written by D.M., B.M. and N.Y.K. with feedback from all authors.

Peer review

Peer review information

Nature Communications thanks Stephan Herzig and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by the NIH/National Cancer Institute (grant R01 CA270110 to N.Y.K.) and by funds from Boston Children’s Hospital. D.M and J.B.X were supported by R01 CA266068 and F32 CA260841. The Metabolomics Workbench repository was supported by NMDR grant# U2CDK119886, Common Fund Data Ecosystem (CFDE, grant# 3OT2OD030544) and Metabolomics Consortium Coordinating Center (M3C, grant # 1U2CDK119889).

Data availability

Source data files are provided with this paper. All metabolic profiling data from this study have been deposited at the NIH Common Fund’s National Metabolomics Data Repository (NMDR), Metabolomics Workbench: (https://www.metabolomicsworkbench.org) under Project ID PR001802, where they can be accessed at (https://doi.org/10.21228/M8SD9K). Raw data for iKras mouse characterization (Source Data for Fig. 1b–d, f, g and Supplementary Fig. 1a–o); Comprehensive lists of ranked metabolites from all Volcano plots (Source Data for Figs. 2d, 6a and Supplementary Figs. 2j, k and 6b); Comprehensive lists of pathways from pathway analysis (Supplementary Fig. 2f); Interactive plots including Heatmaps (Supplementary Fig. 2h), Correlation plots (Supplementary Fig. 2i) and Stem plots (Fig. 5a and Supplementary Fig. 5a, b); as well as Tracing data (Source Data for Figs. 4d–g, i, Supplementary Figs. 4h–i and 7b–d) have all been deposited at Mathur, Deepti; Kalaany, Nada, Mendeley Data, V3, https://doi.org/10.17632/bg5htws7sf.3 and can be accessed at https://data.mendeley.com/datasets/bg5htws7sf/3.

Code availability

Original code (Figs. 3a–c, 5a, 5e, f, 6b, c and Supplementary Figs. 2g, 3a–p, 5a–h) has been deposited at Mathur, Deepti; Kalaany, Nada (2026), Mendeley Data, V3, https://doi.org/10.17632/bg5htws7sf.3 and can be accessed at https://data.mendeley.com/datasets/bg5htws7sf/3

Competing interests

The authors declare no conflicts of interest.

Footnotes

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Contributor Information

Deepti Mathur, Email: mathur.30@osu.edu.

Nada Y. Kalaany, Email: nada.kalaany@childrens.harvard.edu

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77745-0.

References

1.Halbrook, C. J., Lyssiotis, C. A., Pasca di Magliano, M. & Maitra, A. Pancreatic cancer: advances and challenges. Cell186, 1729–1754 10.1016/j.cell.2023.02.014 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
2.Rahib, L., Wehner, M. R., Matrisian, L. M. & Nead, K. T. Estimated projection of US cancer incidence and death to 2040. JAMA Netw. Open4, e214708 10.1001/jamanetworkopen.2021.4708 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
3.Bachmann, J. et al. Cachexia worsens prognosis in patients with resectable pancreatic cancer. J. Gastrointest. Surg.12, 1193–1201 10.1007/s11605-008-0505-z (2008). [DOI] [PubMed] [Google Scholar]
4.Fearon, K. et al. Definition and classification of cancer cachexia: an international consensus. Lancet Oncol.12, 489–495 10.1016/S1470-2045(10)70218-7 (2011). [DOI] [PubMed] [Google Scholar]
5.Ferrer, M. et al. Cachexia: a systemic consequence of progressive, unresolved disease. Cell186, 1824–1845 10.1016/j.cell.2023.03.028 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
6.Gallagher, I. J. et al. Suppression of skeletal muscle turnover in cancer cachexia: evidence from the transcriptome in sequential human muscle biopsies. Clin. Cancer Res18, 2817–2827 10.1158/1078-0432.CCR-11-2133 (2012). [DOI] [PubMed] [Google Scholar]
7.Salazar-Degracia, A. et al. Reduced lung cancer burden by selective immunomodulators elicits improvements in muscle proteolysis and strength in cachectic mice. J. Cell Physiol.234, 18041–18052 10.1002/jcp.28437 (2019). [DOI] [PubMed] [Google Scholar]
8.Williams, J. P. et al. Effect of tumor burden and subsequent surgical resection on skeletal muscle mass and protein turnover in colorectal cancer patients. Am. J. Clin. Nutr.96, 1064–1070 10.3945/ajcn.112.045708 (2012). [DOI] [PubMed] [Google Scholar]
9.Ying, H. et al. Oncogenic Kras maintains pancreatic tumors through regulation of anabolic glucose metabolism. Cell149, 656–670 10.1016/j.cell.2012.01.058 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Ahler, E. et al. Doxycycline alters metabolism and proliferation of human cell lines. PLoS One8, e64561 10.1371/journal.pone.0064561 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Lee, M. S. et al. Ornithine aminotransferase supports polyamine synthesis in pancreatic cancer. Nature616, 339–347 10.1038/s41586-023-05891-2 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Park, W., Chawla, A. & O’Reilly, E. M. Pancreatic Cancer: A Review. Jama326, 851–862 10.1001/jama.2021.13027 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Flurkey, K., Currer, J.M. & Harrison, D. E. TheMouse in Biomedical Research. 3, Ch-20 (Elsevier, Academic Press, 2007).
14.Stephens, N. A. et al. Sexual dimorphism modulates the impact of cancer cachexia on lower limb muscle mass and function. Clin. Nutr.31, 499–505 10.1016/j.clnu.2011.12.008 (2012). [DOI] [PubMed] [Google Scholar]
15.Wallengren, O., Iresjo, B. M., Lundholm, K. & Bosaeus, I. Loss of muscle mass in the end of life in patients with advanced cancer. Support Care Cancer23, 79–86 10.1007/s00520-014-2332-y (2015). [DOI] [PubMed] [Google Scholar]
16.Kir, S. et al. Tumour-derived PTH-related protein triggers adipose tissue browning and cancer cachexia. Nature513, 100–104 10.1038/nature13528 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
17.Ullman-Cullere, M. H. & Foltz, C. J. Body condition scoring: a rapid and accurate method for assessing health status in mice. Lab Anim. Sci.49, 319–323 (1999). [PubMed] [Google Scholar]
18.Oken, M. M. et al. Toxicity and response criteria of the Eastern Cooperative Oncology Group. Am. J. Clin. Oncol.5, 649–655 (1982). [PubMed] [Google Scholar]
19.Lee, J. W. et al. Hepatocytes direct the formation of a pro-metastatic niche in the liver. Nature567, 249–252 10.1038/s41586-019-1004-y (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Heeren, J. & Scheja, L. Metabolic-associated fatty liver disease and lipoprotein metabolism. Mol. Metab.50, 101238 10.1016/j.molmet.2021.101238 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
21.Klatte, D. C. F. et al. Temporal trends in body composition and metabolic markers prior to diagnosis of pancreatic ductal adenocarcinoma. Clin. Gastroenterol. Hepatol. 10.1016/j.cgh.2024.03.038 (2024). [DOI] [PubMed]
22.Holroyde, C. P., Gabuzda, T. G., Putnam, R. C., Paul, P. & Reichard, G. A. Altered glucose metabolism in metastatic carcinoma. Cancer Res35, 3710–3714 (1975). [PubMed] [Google Scholar]
23.O’Connell, T. M. et al. Metabolomic analysis of cancer cachexia reveals distinct lipid and glucose alterations. Metabolomics4, 216–225 10.1007/s11306-008-0113-7 (2008). [DOI] [Google Scholar]
24.Rohdenburg, G. L., Bernhard, A. & Krehbiel, O. Sugar tolerance in cancer. Jama, 1528–1530 (1919).
25.Winter, A., MacAdams, J. & Chevalier, S. Normal protein anabolic response to hyperaminoacidemia in insulin-resistant patients with lung cancer cachexia. Clin. Nutr.31, 765–773 10.1016/j.clnu.2012.05.003 (2012). [DOI] [PubMed] [Google Scholar]
26.Casero, R. A. Jr., Murray Stewart, T. & Pegg, A. E. Polyamine metabolism and cancer: treatments, challenges and opportunities. Nat. Rev. Cancer18, 681–695 10.1038/s41568-018-0050-3 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
27.Dibble, C. C. et al. PI3K drives the de novo synthesis of coenzyme A from vitamin B5. Nature608, 192–198 10.1038/s41586-022-04984-8 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Glicksman, A. S. & Rawson, R. W. Diabetes and altered carbohydrate metabolism in patients with cancer. Cancer9, 1127–1134 (1956). [DOI] [PubMed] [Google Scholar]
29.Honors, M. A. & Kinzig, K. P. The role of insulin resistance in the development of muscle wasting during cancer cachexia. J. Cachexia Sarcopenia Muscle3, 5–11 10.1007/s13539-011-0051-5 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Warburg, O. On respiratory impairment in cancer cells. Science124, 269–270 (1956). [PubMed] [Google Scholar]
31.Faubert, B. et al. Lactate metabolism in human lung tumors. Cell171, 358–371 e359 10.1016/j.cell.2017.09.019 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Hui, S. et al. Glucose feeds the TCA cycle via circulating lactate. Nature551, 115–118 10.1038/nature24057 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
33.Stewart, G. D., Skipworth, R. J. & Fearon, K. C. Cancer cachexia and fatigue. Clin. Med (Lond.)6, 140–143 10.7861/clinmedicine.6-2-140 (2006). [DOI] [PMC free article] [PubMed] [Google Scholar]
34.Al-Sawaf, O. et al. Body composition and lung cancer-associated cachexia in TRACERx. Nat. Med.29, 846–858 10.1038/s41591-023-02232-8 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
35.Chakedis, J. M. et al. Identification of circulating plasma ceramides as a potential sexually dimorphic biomarker of pancreatic cancer-induced cachexia. JCSM Rapid Commun.5, 254–265 10.1002/rco2.68 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
36.Bilgic, S. N. et al. EDA2R-NIK signalling promotes muscle atrophy linked to cancer cachexia. Nature617, 827–834 10.1038/s41586-023-06047-y (2023). [DOI] [PubMed] [Google Scholar]
37.Danai, L. V. et al. Altered exocrine function can drive adipose wasting in early pancreatic cancer. Nature558, 600–604 10.1038/s41586-018-0235-7 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
38.Goncalves, M. D. et al. Fenofibrate prevents skeletal muscle loss in mice with lung cancer. Proc. Natl. Acad. Sci. USA115, E743–E752 10.1073/pnas.1714703115 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
39.Kwon, Y. et al. Systemic organ wasting induced by localized expression of the secreted insulin/IGF antagonist ImpL2. Dev. Cell33, 36–46 10.1016/j.devcel.2015.02.012 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
40.Lockhart, S. M., Saudek, V. & O’Rahilly, S. GDF15: a hormone conveying somatic distress to the brain. Endocr Rev.41 10.1210/endrev/bnaa007 (2020). [DOI] [PMC free article] [PubMed]
41.Sun, N. et al. Inter-organ cross-talk in human cancer cachexia revealed by spatial metabolomics. Metabolism161, 156034 10.1016/j.metabol.2024.156034 (2024). [DOI] [PubMed] [Google Scholar]
42.Mayers, J. R. et al. Elevation of circulating branched-chain amino acids is an early event in human pancreatic adenocarcinoma development. Nat. Med.20, 1193–1198 10.1038/nm.3686 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
43.Couch, M. et al. Cancer cachexia syndrome in head and neck cancer patients: part I. Diagnosis, impact on quality of life and survival, and treatment. Head. Neck29, 401–411 10.1002/hed.20447 (2007). [DOI] [PubMed] [Google Scholar]
44.Steinhauser, M. L. et al. The circulating metabolome of human starvation. JCI Insight3 10.1172/jci.insight.121434 (2018). [DOI] [PMC free article] [PubMed]
45.Townsend, M. K. et al. Reproducibility of metabolomic profiles among men and women in 2 large cohort studies. Clin. Chem.59, 1657–1667 10.1373/clinchem.2012.199133 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
46.Capurso, G., Traini, M., Piciucchi, M., Signoretti, M. & Arcidiacono, P. G. Exocrine pancreatic insufficiency: prevalence, diagnosis, and management. Clin. Exp. Gastroenterol.12, 129–139 10.2147/CEG.S168266 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
47.Mina, A. I. et al. CalR: a web-based analysis tool for indirect calorimetry experiments. Cell Metab.28, 656–666 e651 10.1016/j.cmet.2018.06.019 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
48.Pang, Z. et al. MetaboAnalyst 5.0: narrowing the gap between raw spectra and functional insights. Nucleic Acids Res49, W388–W396 10.1093/nar/gkab382 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
49.Pang, Z. et al. Using MetaboAnalyst 5.0 for LC-HRMS spectra processing, multi-omics integration and covariate adjustment of global metabolomics data. Nat. Protoc.17, 1735–1761 10.1038/s41596-022-00710-w (2022). [DOI] [PubMed] [Google Scholar]
50.The R Development Core Team. R: A language and environment for statistical computing. (R Foundation for Statistical Computing, Vienna, Austria, 2016).
51.Ghini, V. et al. Metabolomics profiling of pre-and post-anesthesia plasma samples of colorectal patients obtained via Ficoll separation. Metabolomics11, 1769–1778 10.1007/s11306-015-0832-5 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
52.Sullivan, M. R. et al. Quantification of microenvironmental metabolites in murine cancers reveals determinants of tumor nutrient availability. eLife 8, e44235 10.7554/eLife.44235 (2019). [DOI] [PMC free article] [PubMed]
53.Marin-Valencia, I. et al. Analysis of tumor metabolism reveals mitochondrial glucose oxidation in genetically diverse human glioblastomas in the mouse brain in vivo. Cell Metab.15, 827–837 10.1016/j.cmet.2012.05.001 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Bartman, C. R. et al. Slow TCA flux and ATP production in primary solid tumours but not metastases. Nature614, 349–357 10.1038/s41586-022-05661-6 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Information^{(13.9MB, pdf)}

Reporting Summary^{(81.4KB, pdf)}

Transparent Peer Review file^{(217.4KB, pdf)}

Source Data - NEW^{(4.1MB, xlsx)}

Data Availability Statement

[CR1] 1.Halbrook, C. J., Lyssiotis, C. A., Pasca di Magliano, M. & Maitra, A. Pancreatic cancer: advances and challenges. Cell186, 1729–1754 10.1016/j.cell.2023.02.014 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR2] 2.Rahib, L., Wehner, M. R., Matrisian, L. M. & Nead, K. T. Estimated projection of US cancer incidence and death to 2040. JAMA Netw. Open4, e214708 10.1001/jamanetworkopen.2021.4708 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR3] 3.Bachmann, J. et al. Cachexia worsens prognosis in patients with resectable pancreatic cancer. J. Gastrointest. Surg.12, 1193–1201 10.1007/s11605-008-0505-z (2008). [DOI] [PubMed] [Google Scholar]

[CR4] 4.Fearon, K. et al. Definition and classification of cancer cachexia: an international consensus. Lancet Oncol.12, 489–495 10.1016/S1470-2045(10)70218-7 (2011). [DOI] [PubMed] [Google Scholar]

[CR5] 5.Ferrer, M. et al. Cachexia: a systemic consequence of progressive, unresolved disease. Cell186, 1824–1845 10.1016/j.cell.2023.03.028 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR6] 6.Gallagher, I. J. et al. Suppression of skeletal muscle turnover in cancer cachexia: evidence from the transcriptome in sequential human muscle biopsies. Clin. Cancer Res18, 2817–2827 10.1158/1078-0432.CCR-11-2133 (2012). [DOI] [PubMed] [Google Scholar]

[CR7] 7.Salazar-Degracia, A. et al. Reduced lung cancer burden by selective immunomodulators elicits improvements in muscle proteolysis and strength in cachectic mice. J. Cell Physiol.234, 18041–18052 10.1002/jcp.28437 (2019). [DOI] [PubMed] [Google Scholar]

[CR8] 8.Williams, J. P. et al. Effect of tumor burden and subsequent surgical resection on skeletal muscle mass and protein turnover in colorectal cancer patients. Am. J. Clin. Nutr.96, 1064–1070 10.3945/ajcn.112.045708 (2012). [DOI] [PubMed] [Google Scholar]

[CR9] 9.Ying, H. et al. Oncogenic Kras maintains pancreatic tumors through regulation of anabolic glucose metabolism. Cell149, 656–670 10.1016/j.cell.2012.01.058 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR10] 10.Ahler, E. et al. Doxycycline alters metabolism and proliferation of human cell lines. PLoS One8, e64561 10.1371/journal.pone.0064561 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR11] 11.Lee, M. S. et al. Ornithine aminotransferase supports polyamine synthesis in pancreatic cancer. Nature616, 339–347 10.1038/s41586-023-05891-2 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR12] 12.Park, W., Chawla, A. & O’Reilly, E. M. Pancreatic Cancer: A Review. Jama326, 851–862 10.1001/jama.2021.13027 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR13] 13.Flurkey, K., Currer, J.M. & Harrison, D. E. TheMouse in Biomedical Research. 3, Ch-20 (Elsevier, Academic Press, 2007).

[CR14] 14.Stephens, N. A. et al. Sexual dimorphism modulates the impact of cancer cachexia on lower limb muscle mass and function. Clin. Nutr.31, 499–505 10.1016/j.clnu.2011.12.008 (2012). [DOI] [PubMed] [Google Scholar]

[CR15] 15.Wallengren, O., Iresjo, B. M., Lundholm, K. & Bosaeus, I. Loss of muscle mass in the end of life in patients with advanced cancer. Support Care Cancer23, 79–86 10.1007/s00520-014-2332-y (2015). [DOI] [PubMed] [Google Scholar]

[CR16] 16.Kir, S. et al. Tumour-derived PTH-related protein triggers adipose tissue browning and cancer cachexia. Nature513, 100–104 10.1038/nature13528 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR17] 17.Ullman-Cullere, M. H. & Foltz, C. J. Body condition scoring: a rapid and accurate method for assessing health status in mice. Lab Anim. Sci.49, 319–323 (1999). [PubMed] [Google Scholar]

[CR18] 18.Oken, M. M. et al. Toxicity and response criteria of the Eastern Cooperative Oncology Group. Am. J. Clin. Oncol.5, 649–655 (1982). [PubMed] [Google Scholar]

[CR19] 19.Lee, J. W. et al. Hepatocytes direct the formation of a pro-metastatic niche in the liver. Nature567, 249–252 10.1038/s41586-019-1004-y (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR20] 20.Heeren, J. & Scheja, L. Metabolic-associated fatty liver disease and lipoprotein metabolism. Mol. Metab.50, 101238 10.1016/j.molmet.2021.101238 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR21] 21.Klatte, D. C. F. et al. Temporal trends in body composition and metabolic markers prior to diagnosis of pancreatic ductal adenocarcinoma. Clin. Gastroenterol. Hepatol. 10.1016/j.cgh.2024.03.038 (2024). [DOI] [PubMed]

[CR22] 22.Holroyde, C. P., Gabuzda, T. G., Putnam, R. C., Paul, P. & Reichard, G. A. Altered glucose metabolism in metastatic carcinoma. Cancer Res35, 3710–3714 (1975). [PubMed] [Google Scholar]

[CR23] 23.O’Connell, T. M. et al. Metabolomic analysis of cancer cachexia reveals distinct lipid and glucose alterations. Metabolomics4, 216–225 10.1007/s11306-008-0113-7 (2008). [DOI] [Google Scholar]

[CR24] 24.Rohdenburg, G. L., Bernhard, A. & Krehbiel, O. Sugar tolerance in cancer. Jama, 1528–1530 (1919).

[CR25] 25.Winter, A., MacAdams, J. & Chevalier, S. Normal protein anabolic response to hyperaminoacidemia in insulin-resistant patients with lung cancer cachexia. Clin. Nutr.31, 765–773 10.1016/j.clnu.2012.05.003 (2012). [DOI] [PubMed] [Google Scholar]

[CR26] 26.Casero, R. A. Jr., Murray Stewart, T. & Pegg, A. E. Polyamine metabolism and cancer: treatments, challenges and opportunities. Nat. Rev. Cancer18, 681–695 10.1038/s41568-018-0050-3 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR27] 27.Dibble, C. C. et al. PI3K drives the de novo synthesis of coenzyme A from vitamin B5. Nature608, 192–198 10.1038/s41586-022-04984-8 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR28] 28.Glicksman, A. S. & Rawson, R. W. Diabetes and altered carbohydrate metabolism in patients with cancer. Cancer9, 1127–1134 (1956). [DOI] [PubMed] [Google Scholar]

[CR29] 29.Honors, M. A. & Kinzig, K. P. The role of insulin resistance in the development of muscle wasting during cancer cachexia. J. Cachexia Sarcopenia Muscle3, 5–11 10.1007/s13539-011-0051-5 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR30] 30.Warburg, O. On respiratory impairment in cancer cells. Science124, 269–270 (1956). [PubMed] [Google Scholar]

[CR31] 31.Faubert, B. et al. Lactate metabolism in human lung tumors. Cell171, 358–371 e359 10.1016/j.cell.2017.09.019 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR32] 32.Hui, S. et al. Glucose feeds the TCA cycle via circulating lactate. Nature551, 115–118 10.1038/nature24057 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR33] 33.Stewart, G. D., Skipworth, R. J. & Fearon, K. C. Cancer cachexia and fatigue. Clin. Med (Lond.)6, 140–143 10.7861/clinmedicine.6-2-140 (2006). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR34] 34.Al-Sawaf, O. et al. Body composition and lung cancer-associated cachexia in TRACERx. Nat. Med.29, 846–858 10.1038/s41591-023-02232-8 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR35] 35.Chakedis, J. M. et al. Identification of circulating plasma ceramides as a potential sexually dimorphic biomarker of pancreatic cancer-induced cachexia. JCSM Rapid Commun.5, 254–265 10.1002/rco2.68 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR36] 36.Bilgic, S. N. et al. EDA2R-NIK signalling promotes muscle atrophy linked to cancer cachexia. Nature617, 827–834 10.1038/s41586-023-06047-y (2023). [DOI] [PubMed] [Google Scholar]

[CR37] 37.Danai, L. V. et al. Altered exocrine function can drive adipose wasting in early pancreatic cancer. Nature558, 600–604 10.1038/s41586-018-0235-7 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR38] 38.Goncalves, M. D. et al. Fenofibrate prevents skeletal muscle loss in mice with lung cancer. Proc. Natl. Acad. Sci. USA115, E743–E752 10.1073/pnas.1714703115 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR39] 39.Kwon, Y. et al. Systemic organ wasting induced by localized expression of the secreted insulin/IGF antagonist ImpL2. Dev. Cell33, 36–46 10.1016/j.devcel.2015.02.012 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR40] 40.Lockhart, S. M., Saudek, V. & O’Rahilly, S. GDF15: a hormone conveying somatic distress to the brain. Endocr Rev.41 10.1210/endrev/bnaa007 (2020). [DOI] [PMC free article] [PubMed]

[CR41] 41.Sun, N. et al. Inter-organ cross-talk in human cancer cachexia revealed by spatial metabolomics. Metabolism161, 156034 10.1016/j.metabol.2024.156034 (2024). [DOI] [PubMed] [Google Scholar]

[CR42] 42.Mayers, J. R. et al. Elevation of circulating branched-chain amino acids is an early event in human pancreatic adenocarcinoma development. Nat. Med.20, 1193–1198 10.1038/nm.3686 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR43] 43.Couch, M. et al. Cancer cachexia syndrome in head and neck cancer patients: part I. Diagnosis, impact on quality of life and survival, and treatment. Head. Neck29, 401–411 10.1002/hed.20447 (2007). [DOI] [PubMed] [Google Scholar]

[CR44] 44.Steinhauser, M. L. et al. The circulating metabolome of human starvation. JCI Insight3 10.1172/jci.insight.121434 (2018). [DOI] [PMC free article] [PubMed]

[CR45] 45.Townsend, M. K. et al. Reproducibility of metabolomic profiles among men and women in 2 large cohort studies. Clin. Chem.59, 1657–1667 10.1373/clinchem.2012.199133 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR46] 46.Capurso, G., Traini, M., Piciucchi, M., Signoretti, M. & Arcidiacono, P. G. Exocrine pancreatic insufficiency: prevalence, diagnosis, and management. Clin. Exp. Gastroenterol.12, 129–139 10.2147/CEG.S168266 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR47] 47.Mina, A. I. et al. CalR: a web-based analysis tool for indirect calorimetry experiments. Cell Metab.28, 656–666 e651 10.1016/j.cmet.2018.06.019 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR48] 48.Pang, Z. et al. MetaboAnalyst 5.0: narrowing the gap between raw spectra and functional insights. Nucleic Acids Res49, W388–W396 10.1093/nar/gkab382 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR49] 49.Pang, Z. et al. Using MetaboAnalyst 5.0 for LC-HRMS spectra processing, multi-omics integration and covariate adjustment of global metabolomics data. Nat. Protoc.17, 1735–1761 10.1038/s41596-022-00710-w (2022). [DOI] [PubMed] [Google Scholar]

[CR50] 50.The R Development Core Team. R: A language and environment for statistical computing. (R Foundation for Statistical Computing, Vienna, Austria, 2016).

[CR51] 51.Ghini, V. et al. Metabolomics profiling of pre-and post-anesthesia plasma samples of colorectal patients obtained via Ficoll separation. Metabolomics11, 1769–1778 10.1007/s11306-015-0832-5 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR52] 52.Sullivan, M. R. et al. Quantification of microenvironmental metabolites in murine cancers reveals determinants of tumor nutrient availability. eLife 8, e44235 10.7554/eLife.44235 (2019). [DOI] [PMC free article] [PubMed]

[CR53] 53.Marin-Valencia, I. et al. Analysis of tumor metabolism reveals mitochondrial glucose oxidation in genetically diverse human glioblastomas in the mouse brain in vivo. Cell Metab.15, 827–837 10.1016/j.cmet.2012.05.001 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR54] 54.Bartman, C. R. et al. Slow TCA flux and ATP production in primary solid tumours but not metastases. Nature614, 349–357 10.1038/s41586-022-05661-6 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

Spatiotemporal metabolic networks in a preclinical model of pancreatic cancer and associated cachexia

Blanca Majem

Min-Sik Lee

Kyung Cheul Shin

Insia Naqvi

Courtney Dennis

Lucas Dailey

Sarah Jeanfavre

Mari Mino-Kenudson

Joao B Xavier

Clary B Clish

Deepti Mathur

Nada Y Kalaany

Abstract

Introduction

Results

An autochthonous mouse model of pancreatic cancer cachexia recapitulates human disease progression

Fig. 1. Pancreatic CAC evolves as three stages of progressive body wasting.

Systemic metabolism is altered in PDA tumor-bearing mice prior to body weight loss

Fig. 2. Systemic metabolism is altered prior to the onset of BW loss in pancreatic CAC.

A systemic lipid signature dominates the PDA cachexia phenotype

Systemic metabolic rewiring in PDA cachexia reveals prospective cross-tissue networks

Networks between “source” tissues, plasma and “sink” tissues

Fig. 3. Concurrent metabolite alterations in plasma, pancreatic IF and distinct tissues infer prospective metabolic networks in pancreatic CAC.

Networks between plasma, IF and pancreas

Altered hexose metabolism in pancreatic cancer and cachexia progression

Fig. 4. Lactate-derived hepatic gluconeogenesis is enhanced in cachectic PDA tumor-bearing mice.

Distinct metabolite patterns accompany cachexia progression

Fig. 5. Algorithmic selection of metabolite predictors allows accurate early detection of pre-cachexia.

Lipid and polar metabolites predict cachexia progression

Model results show similarities with human cachectic markers

Fig. 6. Key metabolic distinctions between Mild and Strong cachexia predict weight loss in iKras mice and correlate with patient data.

Discussion

Methods

Method details

Animal work

Assessment of body weight change and definition of cachexia stages

Food intake

Body condition score

Body composition measurement

Metabolic Cages

Insulin tolerance test (ITT) and glucose tolerance test (GTT)

Necropsy–collection of blood, different tissues and pancreatic interstitial fluid

Plasma glucose and insulin measurements

Histopathology

Infusion of 13C-labeled lactate

Liquid chromatography-mass spectrometry analyses

LC-MS method 1–hydrophilic interaction liquid chromatography (HILIC)-pos

LC-MS Method 2–hydrophilic interaction liquid chromatography (HILIC)-neg

LC-MS method 3 – C8-pos

Metabolomics quality control and data processing

Metabolomics data analysis

Unsupervised analyses

Supervised analyses

Pathway analyses

All digraphs (Umbrella and dandelion plots)

Umbrella plots

Dandelion plots

3D stem plots

Prediction models

Assessment of mild versus strong cachectic phenotype

Comparison to human data

Statistics and reproducibility

Reporting summary

Supplementary information

Source data

Acknowledgements

Author contributions

Peer review

Peer review information

Funding

Data availability

Code availability

Competing interests

Footnotes

Contributor Information

Supplementary information

References

Associated Data

Infusion of ¹³C-labeled lactate