Hepatocytes promote liver metastasis of pancreatic cancer by providing serine

To examine the serine synthesis pathway (SSP) in PDAC liver metastases, both in cancer cells and in the surrounding metastatic microenvironment, we analysed the expression of PHGDH in human PDAC liver-metastasis samples collected through a rapid-autopsy programme⁸. We found that 10 out of 32 cases (31.3%) were PHGDH-negative in tumour cells (Fig. 1a,b), a frequency comparable to that observed in primary tumours⁶. Notably, hepatocytes adjacent to metastatic lesions, particularly those with weak or negative expression of PHGDH in tumour cells, exhibited strong PHGDH expression, whereas hepatocytes in non-metastatic liver tissue showed weak or negative staining (Fig. 1a,b). Dual immunofluorescence staining for PHGDH and albumin (a hepatocyte marker) confirmed this reciprocal pattern, with strong PHGDH expression in hepatocytes adjacent to PHGDH-negative PDAC metastases but little or no PHGDH expression around PHGDH-strong metastases (Fig. 1c). To validate this finding in vivo, we established liver metastases in nude mice using human PDAC cell lines with differential PHGDH expression: MiaPaCa2 (PHGDH-high) and PaTu8902 (PHGDH-null)⁶. Consistent with the human data, dual immunofluorescence staining revealed minimal expression of PHGDH in hepatocytes surrounding MiaPaCa2 metastases but a marked upregulation of PHGDH in those adjacent to PaTu8902 metastases (Extended Data Fig. 1a). Given the central role of hepatocytes in maintaining metabolic homeostasis, we hypothesized that hepatocytes upregulate serine synthesis to metabolically support exSer-dependent PDAC cells in the liver metastatic niche.

To test the hypothesis that hepatocytes support PDAC cell growth by supplying de-novo-synthesized serine, we co-cultured primary mouse hepatocytes with PDAC cells and quantified PDAC cell growth and proliferation (Fig. 2a–d and Extended Data Fig. 3a,b). Under −SG conditions, the growth of exSer-dependent PDAC cells was markedly impaired, but it was restored by co-culture with hepatocytes (Fig. 2b,d and Extended Data Fig. 3b). This hepatocyte-dependent growth increase was not observed in PHGDH-proficient PDAC cells or in SG-containing medium (Fig. 2b,d and Extended Data Fig. 3b), indicating that hepatocyte-derived serine specifically promotes the proliferation of exSer-dependent PDAC cells in conditions in which extracellular serine is limited. These in vitro findings were mirrored in vivo: sgPhgdh PDAC cells exhibited increased proliferation at the tumour–liver interface, compared with the tumour core, suggesting that there is a spatial relationship between tumour growth and hepatocyte proximity (Extended Data Fig. 3c,d).

To determine whether activation of the SSP leads to functional serine production and transfer to cancer cells, we performed co-culture assays with ¹³C₆-glucose (Fig. 2e,f). Labelled M+3 serine, a product of the SSP, was detected in sgPhgdh PDAC cells only when co-cultured with hepatocytes, and not in monoculture (Fig. 2g and Extended Data Fig. 3e,f), suggesting that hepatocyte-derived serine is synthesized de novo, released into the medium and taken up by PDAC cells.

Consistently, qPCR and RNA sequencing (RNA-seq) analyses revealed an upregulation of SSP genes in hepatocytes that were co-cultured with sgPhgdh HY19636 cells (Fig. 2h), along with a trend towards enrichment of signatures of glycine–serine–threonine metabolism (Extended Data Fig. 3g). These transcriptional changes recapitulate what was observed in patients with liver metastases and in mouse models (Fig. 1 and Extended Data Fig. 1). Notably, −SG conditions alone did not clearly induce the expression of SSP genes in monocultured hepatocytes, whereas co-culture with exSer-dependent PDAC cells did (Fig. 2h), suggesting that PDAC-derived signals have a role in hepatocyte metabolic reprogramming.

We next sought to identify soluble factors secreted by exSer-dependent PDAC cells that induce PHGDH expression in hepatocytes. RNA-seq analysis identified Cxcl5 as a candidate gene, which was upregulated (1) in sgPhgdh HY19636 cells compared with sgTOM HY19636 cells under co-culture with hepatocytes, and (2) in sgPhgdh HY19636 cells grown in co-culture with hepatocytes compared with those grown in monoculture (Fig. 3a and Supplementary Table 1). This upregulation of CXCL5 was further confirmed at the protein level in co-culture systems (Extended Data Fig. 4a) and in tissue interstitial fluids (TIFs) from mouse liver metastases (Fig. 3b). Notably, the increased expression of CXCL5 persisted for at least 14 days in vitro in sgPhgdh HY19636 cells after isolation from liver metastases (Extended Data Fig. 4b,c). CXCL5 concentrations were also increased in human liver-metastasis samples, with the highest levels in tissues bearing PHGDH-weak or PHGDH-negative metastases, lower levels in PHGDH-strong metastases and undetectable levels in normal liver tissue (Fig. 3c).

We previously showed that exSer-dependent PDAC cells in primary tumours rely on nerve growth factor (NGF)-driven tumour innervation and neuronal serine supply⁶. Consistently, sgPhgdh HY19636 cells showed increased production of NGF under –SG conditions, but this induction was blunted by co-culture with hepatocytes (Extended Data Fig. 4d). Moreover, mouse liver metastases showed only a non-significant trend towards increased NGF in TIFs and no increase in tumour innervation (Extended Data Fig. 4e,f). Conversely, sgPhgdh tumours in the pancreas did not upregulate Cxcl5 after dietary serine restriction (Extended Data Fig. 4g). Together, these data suggest that exSer-dependent PDAC cells engage distinct tissue-dependent paracrine programs for serine acquisition, with the NGF–nerve axis predominating at the primary site and CXCL5 emerging as a candidate mediator of tumour–hepatocyte cross-talk in liver metastases.

To delineate the signalling axis that is activated in hepatocytes after co-culture with sgPhgdh HY19636 cells under CXCR2 signalling, we performed over-representation analysis (ORA) in RNA-seq data from primary mouse hepatocytes. Specifically, we compared hepatocytes cultured with or without sgPhgdh HY19636 cells and, under co-culture conditions, with or without the CXCR2 inhibitor SB225002. In both comparisons, the PI3K–AKT signalling pathway was significantly over-represented (Fig. 4a,b). Consistently, co-culture with PDAC cells increased AKT phosphorylation and PHGDH expression in hepatocytes (Extended Data Fig. 5a). Pharmacological inhibition of either CXCR2 or PI3K–AKT blocked this upregulation of PHGDH (Extended Data Fig. 5a) and suppressed de novo serine synthesis from ¹³C₆-glucose (Figs. 2e and 4c,d and Extended Data Fig. 5b). Consequently, the growth-promoting effect of hepatocytes on exSer-dependent PDAC cells under −SG conditions was lost (Extended Data Fig. 5c,d). These results identify the PI3K–AKT pathway as a crucial node that mediates PHGDH upregulation in hepatocytes downstream of the CXCL5–CXCR2 axis.

Among the candidate transcription factors that might regulate Phgdh expression, we focused on FOXO3A for two reasons: (1) FOXO3 is inactivated by PI3K–AKT signalling through phosphorylation-dependent cytoplasmic sequestration and subsequent proteasomal degradation^14,15; and (2) FOXO3 can function as both a repressor and an activator of transcription, depending on the target genes^16,17,18. Consistently, gene set enrichment analysis (GSEA) revealed a significant negative enrichment of the FOXO signalling pathway in hepatocytes co-cultured with sgPhgdh PDAC cells, relative to monoculture, whereas CXCR2 inhibition reversed this pattern to positive enrichment (Extended Data Fig. 5e,f). Similarly, GSEA using transcription factor–target-gene sets showed a negative enrichment of FOXO3 target genes in co-cultured hepatocytes, with a trend towards positive enrichment after inhibition of CXCR2 (Extended Data Fig. 5g), further supporting FOXO3A as a plausible downstream effector of PI3K–AKT signalling in this context.

On the basis of these findings, we hypothesized that nuclear FOXO3 represses Phgdh expression in hepatocytes, and that PI3K–AKT activation downstream of the CXCL5–CXCR2 axis alleviates this repression by promoting FOXO3 phosphorylation and cytoplasmic retention^14,15,19 (Fig. 4e). Supporting this model, pharmacological inhibition of CXCR2 or PI3K in hepatocytes co-cultured with sgPhgdh PDAC cells enhanced FOXO3 nuclear localization (Fig. 4f,g) and increased FOXO3 occupancy at TGTTT-core FOXO consensus-like motifs in the Phgdh promoter (P1 and P2), but not at a distal gene-body region that was used as a negative control (P3)¹⁸ (Fig. 4h). Of note, this regulatory relationship was also evident in human liver-metastasis samples: hepatocytes with strong PHGDH expression exhibited predominantly non-nuclear FOXO3, whereas hepatocytes with weak or absent PHGDH expression exhibited predominantly nuclear FOXO3 (Extended Data Fig. 5h). Consistent with this, inhibiting CXCR2 in mice bearing sgPhgdh PDAC liver metastases promoted FOXO3 nuclear localization in hepatocytes (Extended Data Fig. 5i). Together, these findings establish FOXO3 as a transcriptional repressor of PHGDH, whose activity is relieved by PI3K–AKT signalling downstream of the CXCL5–CXCR2 axis.

FOXO3A is phosphorylated by AKT at Thr32, Ser253 and Ser315, which promotes its cytoplasmic retention¹⁹. To investigate the functional importance of these phosphorylation sites, we used a non-phosphorylatable FOXO3A mutant (T32A/S253A/S315A; FOXO3A(AAA)) by replacing these three residues with alanine¹⁹. Endogenous FOXO3A was deleted in a mouse hepatocyte cell line (AML12) and replaced with either wild-type (WT) or mutant (AAA) constructs (Extended Data Fig. 6a). After inhibition of PI3K, FOXO3A(WT) translocated to the nucleus, whereas the non-phosphorylatable FOXO3A(AAA) mutant remained constitutively nuclear (Extended Data Fig. 6b,c). Consistent with its nuclear localization, FOXO3A(AAA) expression significantly reduced the activity of the Phgdh promoter, mimicking the effect of PI3K inhibition (Extended Data Fig. 6d,e). In co-culture with exSer-dependent PDAC cells, FOXO3A(AAA)-expressing hepatocytes showed enhanced FOXO3A binding to the Phgdh promoter (Extended Data Fig. 6f) and impaired serine production, as evidenced by reduced incorporation of ¹³C₆ into co-cultured PDAC cells (Extended Data Fig. 6g,h). Moreover, both Cxcr2 deletion and FOXO3A(AAA) expression in AML12 cells similarly suppressed their capacity to support the growth of exSer-dependent PDAC cells under −SG conditions, supporting the functional relevance of FOXO3A-mediated PHGDH regulation downstream of CXCR2–PI3K–AKT signalling (Extended Data Fig. 6i–k). Together, these findings identify FOXO3A as a transcriptional repressor of Phgdh that is inactivated by CXCL5–CXCR2–PI3K–AKT signalling in hepatocytes, representing an important downstream component of the hepatocyte serine-supplying program that supports exSer-dependent PDAC cells during liver metastasis.

Our findings thus far show that genetic ablation of Cxcl5 in PDAC cells, or targeted deletion of Phgdh or Cxcr2 in hepatocytes (Figs. 1 and 3), effectively suppresses the growth of exSer-dependent PDAC under dietary serine restriction. Several CXCR2 (or CXCR1 and CXCR2; CXCR1/2) inhibitors have entered clinical testing for PDAC, including SX-682 (NCT04477343). Furthermore, the safety and feasibility of a −SG diet have been demonstrated in a phase I clinical trial²⁰. To evaluate the translational potential of our findings, we next tested the therapeutic efficacy of combining a pharmacological CXCR2 inhibitor (SB225002) with dietary serine restriction in preclinical models of exSer-dependent PDAC.

To this end, mice were intrasplenically injected with sgTOM or sgPhgdh HY19636 cells and subsequently maintained on either an AA control diet or a –SG diet, with concurrent administration of vehicle or a CXCR2 inhibitor (CXCRi; SB225002) (Fig. 5a). Notably, the combination of CXCR2 inhibition and a −SG diet did not affect liver weights in mice that underwent sham surgery (Extended Data Fig. 7a,b), suggesting the potential safety and translational relevance of this strategy for future clinical applications. CXCR2 inhibition significantly reduced the liver metastatic burden of sgPhgdh PDAC cells in mice that were fed a –SG diet, as measured by liver weight and metastatic area (Fig. 5b–d). This effect was not observed in mice injected with sgTOM cells or in those maintained on a complete AA control diet (Fig. 5b–d), suggesting that the therapeutic effect of CXCR2 inhibition is mediated mainly by the disruption of hepatocyte-derived serine supply rather than by immune modulation. Consistently, PHGDH expression was robustly upregulated in hepatocytes adjacent to sgPhgdh metastases, and this effect was diminished after CXCR2 inhibition (Extended Data Fig. 7c,d), further supporting a model in which CXCL5–CXCR2 signalling induces hepatocyte-derived serine synthesis to support the metastatic outgrowth of exSer-dependent PDAC when dietary serine is restricted. To assess therapeutic benefits beyond tumour burden, we repeated the intrasplenic injection experiment using a lower number of injected cells to monitor survival. Mice bearing sgPhgdh metastases and treated with SB225002 survived significantly longer than did untreated controls (Fig. 5e).

Although the intrasplenic injection model is well suited to the study of later stages of metastasis, including tumour-cell engraftment and expansion in the liver, it bypasses early steps of the metastatic cascade. To address this limitation and better recapitulate clinical progression after primary tumour resection, we used a spontaneous metastasis model²¹. sgTOM or sgPhgdh HY19636 cells were orthotopically implanted into the pancreas. Three weeks later, primary tumours were surgically resected, and mice were euthanized at week 7 (Fig. 5f). To assess the stage-dependent therapeutic effects of CXCR2 inhibition, SB225002 was administered either 2 days after (CXCR2i (during)) or 2 weeks after (CXCR2i (after)) tumour implantation. CXCR2 inhibition had no effect on primary tumour growth regardless of Phgdh status (Fig. 5g), but significantly reduced the liver metastasis of sgPhgdh PDAC cells when administered at either time point (Fig. 5h,i). These data show that the CXCL5–CXCR2 axis has an essential role in sustaining the liver metastatic outgrowth of exSer-dependent PDAC under serine restriction, but that it has little effect on primary tumour growth.

How exSer-dependent PDAC cells secure serine within liver metastases, and whether this involves organ-specific stromal metabolic responses, has remained unclear. Here we show that loss of PHGDH and dependence on exSer reprogram AA metabolism in hepatocytes through a CXCL5–CXCR2–PI3K–AKT–FOXO3A–PHGDH axis, thereby supporting the outgrowth of PDAC in the liver metastatic niche (Fig. 5j).

We observed a loss of PHGDH in 31.3% of liver metastases—a level comparable with that reported in primary PDAC tumours (37.2%)⁶—which suggests that SSP deficiency does not prevent liver metastatic colonization (Fig. 1a,b). These SSP-deficient cells might compensate by acquiring extracellular serine, and might even gain a selective advantage from this dependency. As reported previously in brain tumours, the uptake of extracellular serine can spare glucose-derived carbon by reducing cells’ reliance on de novo serine synthesis, and can redirect it towards other biosynthetic processes²²; a similar mechanism might support PDAC growth if glucose is constrained in a similar manner in liver metastases, as has been described in primary PDAC²³.

Our study also indicates that serine acquisition is organ-dependent in this PDAC subtype. Whereas primary tumours rely on NGF-driven tumour innervation and neuronal serine supply⁶, liver metastases show no clear increase in innervation but instead engage a CXCL5-associated, hepatocyte-dependent serine supply program (Fig. 3b,c and Extended Data Fig. 4a–f). Consistently, dietary serine restriction did not induce the expression of Cxcl5 in pancreatic primary tumours (Extended Data Fig. 4g). Notably, hepatocyte co-culture attenuated NGF induction while enhancing CXCL5 expression in Phgdh-knockout PDAC cells (Fig. 3a and Extended Data Fig. 4d), further supporting the idea that distinct microenvironments favour different paracrine programs for serine acquisition. These findings also raise the possibility of bidirectional PDAC–hepatocyte communication, although the hepatocyte-derived factor(s) responsible remain unidentified. Together, these observations suggest that exSer-dependent PDAC cells use different organ-specific stromal circuits to satisfy the same metabolic need. Thus, CXCL5–CXCR2 signalling should not be viewed as a universal mechanism that supports all PHGDH-deficient lesions, but rather as a pathway directed at the liver metastatic niche, with particular therapeutic relevance in peri-operative or post-resection settings, in which suppression of liver metastatic recurrence is a major goal.

Because the CXCL5–CXCR2 axis is also implicated in immune-cell recruitment¹³, whether immune mechanisms contribute to the anti-metastatic effects observed here remains an important question. However, the selective reduction of Phgdh-deficient PDAC metastases in hepatocyte-specific Cxcr2-knockout mice, together with the loss of CXCR2 inhibition efficacy under a complete AA control diet, supports hepatocyte CXCR2-dependent metabolic responses as a major contributor in this model (Figs. 3j–l and 5b–d). Future studies will be needed to define how this metabolic axis interacts with immune responses.

Our data support combining dietary SG restriction with CXCR2 inhibition as a liver-metastasis-directed strategy in this PDAC subtype. Although this combination showed no clear efficacy against the pancreatic primary tumour within the time frame of our study, it suppressed liver metastatic burden, prolonged survival and reduced liver metastatic recurrence after primary tumour resection (Fig. 5g–i). Because CXCR1/2 inhibitors are in clinical testing for PDAC, including SX-682 (NCT04477343), and because an –SG diet has shown safety and feasibility in a phase I trial²⁰, this approach might be practical as a time-limited peri-operative intervention for patients with PHGDH-deficient, exSer-dependent PDAC, given the physiological roles of serine in normal tissues^24,25. PHGDH immunostaining of resected specimens, or of biopsy samples obtained pre-operatively by endoscopic ultrasound-guided fine-needle aspiration (EUS-FNA), could be used to help to identify patients who are most likely to benefit. Clinical implementation will require careful optimization of treatment duration, nutritional support and biomarker-guided monitoring.

Direct inhibition of PHGDH represents a key alternative strategy for targeting serine metabolism, and pharmacological PHGDH inhibition has been shown to cooperate with dietary SG restriction to suppress tumour growth in vivo²⁶. However, PHGDH inhibitors that are currently available act systemically rather than selectively in hepatocytes, and might therefore suppress compensatory de novo serine biosynthesis in normal tissues under extracellular serine limitation. By contrast, our rationale for CXCR2 inhibition is to block the tumour-induced activation of a hepatocyte-derived serine supply while potentially preserving this compensatory pathway in normal tissues, thereby exploiting the differential vulnerability of exSer-dependent PDAC cells. Nevertheless, pharmacological inhibition of CXCR2 is not hepatocyte-specific and is likely to exert pleiotropic effects throughout the metastatic microenvironment, including on immune and stromal cells. Accordingly, the anti-metastatic effects observed with systemic CXCR2 inhibition in this study cannot be attributed solely to the blockade of CXCR2 signalling in hepatocytes. Thus, whether CXCR2 inhibition provides a therapeutic advantage over systemic PHGDH inhibition remains unresolved and will require direct comparisons of their efficacy and tolerability under dietary SG restriction.

Several limitations should be noted. First, because the human analyses used rapid-autopsy specimens, which reflect end-stage disease, tumour serine demand and hepatocyte serine supply might be accentuated by reduced food intake and cachexia-associated declines in circulating serine in the terminal phase of disease. Thus, the liver metastatic niche during early colonization warrants separate investigation, as does the conservation of PHGDH status between matched primary tumours and liver metastases. Second, the mechanisms that drive CXCL5 induction in PDAC cells and serine efflux from hepatocytes remain unresolved, although both represent processes that might be targetable. Third, the extent to which hepatocyte-derived serine supports tumour growth in human disease, and whether this dependency is even greater under standard dietary conditions than it is in mouse models, remains unclear. If the latter is the case, dietary serine restriction could prove dispensable in patients, and CXCR2 inhibition alone might suffice in selected settings. Although our absolute quantification experiments support the feasibility of a hepatocyte-derived serine supply under the tested culture conditions (Fig. 2i–l and Extended Data Fig. 3h), they do not directly compare hepatocyte serine export with tumour-cell uptake demand in vivo. Addressing these questions will require future metabolic flux analyses, including in vivo ¹³C-glucose tracing.

In summary, we identify a CXCL5–CXCR2–PI3K–AKT–FOXO3A–PHGDH axis through which PHGDH-deficient, exSer-dependent PDAC rewires hepatocyte AA metabolism in the liver metastatic niche to support metastatic outgrowth. These findings support a model in which tumour-intrinsic metabolic states can elicit organ-specific stromal programs, providing a basis for tailoring therapy according to both tumour metabolism and anatomical site.

Human samples

Snap-frozen and formalin-fixed paraffin-embedded (FFPE) liver tissue sections from de-identified patients with PDAC and individuals without PDAC were obtained from the Rapid Autopsy Program at the University of Nebraska Medical Center (UNMC), in accordance with institutional review board (IRB) approval (IRB 091-01). Written informed consent for the research use of these specimens was obtained from all participants in accordance with the approved protocol. To ensure specimen quality, organs were collected within 3 h post-mortem, after which tissues were either flash-frozen in liquid nitrogen or immediately fixed in formalin. For histological analyses, FFPE tissues were sectioned at a thickness of 4 μm and mounted onto charged glass slides. Tissue homogenates were prepared from snap-frozen samples using Lysis Buffer (R&D Systems, 895347) and a bead-based tissue lyser (TissueLyser LT; QIAGEN), according to the manufacturers’ protocols.

Immunohistochemistry and immunofluorescence (manual staining)

FFPE sections of mouse and human liver tissue (5 µm) were baked at 65 °C for 1 h, cooled to room temperature, deparaffinized in xylene and rehydrated through a graded ethanol series. Antigen retrieval was performed by heating sections in 0.05 M sodium citrate buffer (pH 6.0) for 20 min using a microwave oven. Endogenous peroxidase activity was quenched by incubation in 3% hydrogen peroxide in methanol for 10 min. Sections were blocked for 1 h at room temperature with 2.5% horse serum (Vector Laboratories, MP-7800-15) for immunohistochemistry (IHC) or with 10% normal goat serum (NGS; Vector Laboratories, S-1000-20) for immunofluorescence. Sections were incubated overnight at 4 °C in a humidified chamber with primary antibodies: anti-PHGDH (Sigma, HPA021241, 1:100), anti-albumin (Abcam, ab106582, 1:200), anti-CK19 (DSHB, TROMA-III, 1:100), anti-Ki67 (Abcam, ab15580, 1:400), and anti-FOXO3A (Cell Signaling Technology, 2497, 1:200).

For chromogenic IHC, signal detection was performed using the VECTASTAIN Elite ABC-HRP kit (Vector Laboratories, PK-6100) according to the manufacturer’s instructions, followed by development with the DAB Substrate Kit (Vector Laboratories, SK-4100). Sections were counterstained with haematoxylin (Vector Laboratories, H-3401), dehydrated through graded ethanol, cleared in xylene and mounted with Permount mounting medium (Thermo Fisher Scientific, SP15-100). Bright-field images were acquired using an Aperio digital slide scanner (Leica Biosystems) using a 20× objective lens. For immunofluorescence, sections were washed and incubated with fluorescently conjugated secondary antibodies (1:1,000, Supplementary Table 2) at room temperature for 1 h, followed by nuclear counterstaining with DAPI. Some samples that were included in analyses of liver weight or whole-section tumour burden were not included in immunostaining quantification when extensive necrosis, tissue disruption, poor section quality or staining artefacts prevented reliable assessment of marker-positive cells or staining intensity.

For quantification of PHGDH IHC intensity, chromogenic PHGDH IHC images were analysed using ImageScope (Leica Biosystems), as previously described⁶. Tumour and hepatocyte regions were annotated manually, and staining intensity was quantified using the Positive Pixel Count v.9 algorithm. Pixels were classified as negative (0–100), weak (100–175) or strong (175–200) on the basis of intensity thresholds. For quantification of PHGDH immunofluorescence intensity, images were analysed using ZEN lite (ZEISS, v.3.8.99) and Fiji (ImageJ 1.54g; Java 1.8.0_345) to measure fluorescence intensities. Colocalization analysis is described in ‘Colocalization analysis’ below. For mouse samples, five to eight randomly selected tissue sections per group were analysed for IHC. For human samples, all available patient specimens were included in the IHC analysis (n = 32 for normal liver samples and n = 32 for liver-metastasis samples). For immunofluorescence-based correlation analyses, three randomly selected individuals per group were analysed, with three fields per individual.

PGP9.5 immunohistochemistry (automated staining)

For PGP9.5 staining, 5-μm-thick sections were prepared and immunostained on a Leica BondRX automated stainer, according to the manufacturer’s instructions. In brief, tissues underwent deparaffinization online, followed by epitope retrieval for 20 min at 100 °C with Leica Biosystems ER1 (pH 6) solution (Leica, AR9961) and endogenous peroxidase activity blocking with H₂O₂. Sections were then incubated with primary antibodies against PGP9.5 (Abcam, ab108986) at a 1:350 dilution for 15 min at ambient temperature. Primary antibodies were detected with the anti-rabbit HRP-conjugated polymer and 3,3′-diaminobenzidine (DAB) substrate that are provided in the Leica BOND Polymer Refine Detection System (DS9800). Finally, samples were counterstained with haematoxylin, dehydrated and coverslipped with Permount. Slides were scanned at 40× on a Hamamatsu Nanozoomer (2.0-HT) whole-slide scanner using NDP.scan (v.3.4.2). For quantification of PGP9.5 IHC, 4–13 randomly selected fields per mouse at 40× magnification were analysed, depending on the size of the metastatic lesions, to determine the PGP9.5-positive area. Regions containing necrosis or overt non-specific staining were excluded from the analysis. Quantification of DAB-positive areas was performed using Fiji (ImageJ). Colour deconvolution (H-DAB) was applied to separate the DAB signal, followed by background subtraction and fixed thresholding applied uniformly across all images. The percentage of DAB-positive area relative to the total image area was calculated automatically. For each mouse, values are presented as the mean of the analysed fields. Statistical analysis was performed by one-way ANOVA followed by Holm–Šídák’s post-hoc multiple-comparison tests.

Colocalization analysis

Fluorescence imaging was performed on a ZEISS spinning-disk confocal microscope. DAPI was excited at 353 nm and emission was collected at 465 nm; albumin was excited at 488 nm and emission was collected at 509 nm; and FOXO3A was excited at 587 nm and emission was collected at 610 nm. Acquisition settings were kept constant across experimental conditions to enable direct comparison of signal intensity and colocalization. Colocalization between FOXO3A and nuclear DAPI staining was quantified using the Coloc 2 plug-in in Fiji (ImageJ, v.2.1.0). To minimize diffuse background signal, manual background subtraction was applied uniformly across images by subtracting an intensity value of 112 from the DAPI channel and 107 from the FOXO3A channel in human samples. In mouse samples, 73 was subtracted from the DAPI channel and 80 was subtracted from the FOXO3A channel across both images to reduce background and improve statistical colocalization. Regions of interest (ROIs) were drawn over nuclei with a strong DAPI signal to ensure analysis was restricted to areas with clear nuclear staining. Within each ROI, Coloc 2 was run to calculate colocalization statistics, including Pearson’s correlation coefficient and Spearman’s rank correlation. Default Coloc 2 settings were used unless otherwise noted. For mouse liver-metastasis samples treated with the CXCR2 inhibitor, DAPI staining occasionally appeared more diffuse than it did in vehicle-treated controls, which can reduce the dynamic range of pixel-intensity-based colocalization metrics (particularly Pearson’s correlation). Pearson’s and Spearman’s coefficients were therefore interpreted in conjunction with a qualitative assessment of FOXO3A signal distribution. In these samples, the FOXO3A signal in hepatocytes showed a round, punctate pattern consistent with nuclear localization.

Quantification of metastatic burden in the liver

Metastatic burden in the liver was quantified using H&E-stained sections. For each mouse, H&E-stained sections from multiple liver lobes were scanned at 20× magnification using an Aperio digital slide scanner (Leica Biosystems). Whole-slide H&E images were analysed using MetFinder, a deep-learning-based tool for automated quantification of metastatic tumour burden in mouse organs (https://metfinder.org/)²⁷. MetFinder automatically annotated each whole-slide image into normal liver and metastatic tumour regions. The resulting heat maps were carefully inspected to verify appropriate annotation, and metastatic burden was calculated as the percentage of metastatic tumour area relative to the total liver section area.

Cell culture

The human pancreatic cancer cell lines PaTu8902 and MiaPaCa2 and the mouse hepatocyte cell line AML12 were obtained from the American Type Culture Collection (ATCC) or DSMZ. PaTu8902 and MiaPaCa2 cells were cultured in DMEM (Corning) supplemented with 10% fetal bovine serum (FBS; Atlanta Biologicals, S11550H). AML12 cells were maintained in DMEM/F12 (Corning) supplemented with 10% FBS, 10 µg ml⁻¹ insulin, 5.5 µg ml⁻¹ transferrin, 5 ng ml⁻¹ selenium and 40 ng ml⁻¹ dexamethasone (Gibco). All media were supplemented with 1% penicillin–streptomycin (Gibco). The primary mouse PDAC cell line HY19636 was established from pancreatic tumours derived from KPC genetically engineered mouse models (LSL-Kras^G12D/+; Trp53^flox/+; p48-Cre), as described previously⁹. HY19636 cells were genetically modified to silence or overexpress genes of interest using CRISPR–Cas9-based approaches. All mouse PDAC cell lines were cultured in DMEM (Corning) supplemented with 10% FBS and 1% penicillin–streptomycin (Gibco). Primary mouse hepatocytes were isolated from C57BL/6 mice as described elsewhere in the Methods. Isolated primary mouse hepatocytes were plated on 0.01% collagen-coated culture plates and maintained for 6–10 days in Williams’ E medium (Corning) supplemented with 1% GlutaMAX (Gibco) and 1% penicillin–streptomycin (Gibco). All cells were cultured at 37 °C in a humidified incubator with 5% CO₂. Cultures were routinely confirmed to be mycoplasma free using the MycoAlert Detection Kit (Lonza) or PCR-based assays. Cell lines were periodically authenticated by short tandem repeat profiling, and low-passage cultures were maintained in a centralized laboratory cell bank. For experiments using media containing different concentrations of serine and glycine, basal DMEM containing 2.78 mM (0.5 g l⁻¹) glucose, 4 mM glutamine and 1 mM sodium pyruvate was supplemented as indicated with serine and glycine at 400 µM (high SG), 150 µM (low SG), or 0 µM (−SG), and with 10% dialysed FBS, unless otherwise noted.

Reagents

For pharmacological inhibition of CXCR2 and PI3K, SB225002 (Selleck, S7651) and BKM120 (buparlisib; Selleck, S2247) were used, respectively. SB225002 was prepared as a 20 mg ml⁻¹ stock solution in DMSO, and BKM120 was prepared in DMSO. For in vitro experiments, SB225002 and BKM120 were used at final concentrations of 100 nM and 50 nM, respectively, unless otherwise indicated. Vehicle control samples received an equivalent volume of DMSO. Recombinant mouse CXCL5 (LIX) (rCXCL5; R&D Systems, 433-MC-025) and a CXCL5-neutralizing antibody (R&D Systems, MAB433-100) were used at final concentrations of 500 ng ml⁻¹ and 2,000 ng ml⁻¹, respectively.

Isolation of mouse hepatocytes

Primary mouse hepatocytes were isolated using a two-step collagenase digestion method as described previously²⁸. Mice were anaesthetized, and a midline abdominal incision was made. After placement of a medium-sized ligation clip (Teleflex, 002200) on the inferior vena cava (IVC) immediately below the diaphragm and above the liver, the IVC was cannulated with a 20-gauge infusion catheter. The liver was first perfused with 20 ml pre-warmed perfusion buffer consisting of Hank’s balanced salt solution (HBSS) without Ca²⁺, Mg²⁺ or phenol red, supplemented with 10 mM HEPES and 0.5 mM EGTA (pH 7.4), at 37 °C and a flow rate of 7 ml min⁻¹. The portal vein was severed to allow drainage of blood and perfusion buffer. Subsequently, the liver was perfused with 30 ml of digestion buffer composed of HBSS containing Ca²⁺, Mg²⁺ and phenol red, supplemented with 10 mM HEPES and 25 µg ml⁻¹ Liberase TM (Sigma-Aldrich, 5401127001), at a flow rate of 5 ml min⁻¹ for 6 min. After digestion, the liver was carefully excised and transferred to a 10-cm culture dish containing HBSS without Ca²⁺ and Mg²⁺. The liver capsule was gently torn, and hepatocytes were released into the buffer by gentle agitation. The resulting cell suspension was collected using a 25-ml serological pipette, filtered through a 70-µm cell strainer and centrifuged at 40g for 1 min at 4 °C without deceleration. The supernatant was aspirated, and the cell pellet was resuspended in HBSS. This low-speed centrifugation step was repeated several times to remove debris and non-parenchymal cells. The final pellet was resuspended in 10 ml Williams’ E medium (Corning) and mixed with an equal volume of Percoll solution, followed by centrifugation at 200g for 10 min at 4 °C. The resulting hepatocyte-enriched pellet was used for primary culture or downstream cell-sorting experiments.

Plasmids and generation of stable cell lines

To generate control (sgTOM), Phgdh-knockout or Cxcl5-knockout PDAC cell lines, the lentiviral CRISPR–Cas9 vector lentiCRISPRv2 (a gift from F. Zhang; Addgene plasmid 52961) was used. Single-guide RNAs (sgRNAs) targeting the indicated genes were cloned into the lentiCRISPRv2 backbone according to the standard protocol. The sgRNA target sequences were as follows:

sgTOM: 5′-GGCCACGAGTTCGAGATCGA-3′

sgPhgdh_1: 5′-TGAGCCCGGAATACGAGCAG-3′

sgPhgdh_2: 5′-AGGTGCTCCCTACCAAGCCG-3′

sgCxcl5_1: 5′-ATGGCGAGATGGAACCGCTG-3′

sgCxcl5_2: 5′-TTCCTCAGTCATAGCCGCAA-3′.

After lentiviral transduction, single-cell-derived clones were generated using a MoFlo XDP cell sorter (Beckman Coulter) and subsequently screened for efficient gene knockout. To minimize potential off-target effects and clonal variability, multiple independently derived clones with validated knockout were pooled and used for downstream experiments. For Phgdh or Cxcl5 knockout, clones generated using two independent sgRNAs were pooled before analysis. Rescue experiments were performed using the doxycycline-inducible lentiviral expression vector pCW57.1 carrying either wild-type or catalytically inactive human PHGDH (Addgene plasmids 154916 and 154903), which were gifts from R. Possemato and M. Pacold. Lentiviral transduction was performed according to standard protocols, and transduced cells were selected with puromycin (2 µg ml⁻¹). For the generation of GFP-expressing sgTOM and mCherry-expressing sgPhgdh PDAC cell lines, MIGR1 (Addgene, 27490) and pMSCV-IRES-mCherry FP (Addgene, 52114; a gift from W. Pear) were used. AML12 cells with stable knockout of Cxcr2 or Foxo3a were generated using sgRNAs cloned into the pSpCas9(BB)-2A-GFP (PX458) vector (Addgene plasmid 48138). The sgRNA target sequences were as follows: sgCxcr2, 5′-TACGCAGTACGACCCTCAAA-3′; sgFoxo3a-sg1, 5′-GGACTGTCGTCTGCCGACTC-3′; and sgFoxo3a-sg2, 5′-TCTCGATGGCGCGGGTGATC-3′. For Foxo3a knockout, clones generated using the two independent sgRNAs were pooled before analysis. For ectopic expression of human FOXO3A, AML12 cells were transfected with pcDNA3-Flag-FOXO3A/FKHRL1 (wild type; Addgene plasmid 10708) or pcDNA3-Flag-FOXO3A/FKHRL1-AAA, a non-phosphorylatable mutant (Addgene plasmid 10709).

For transfection of retroviral plasmids, HEK293FT cells were co-transfected with the retroviral vector and the packaging plasmids pHit60 and VSVG at a 0.5:0.25:0.25 ratio. For transfection of lentiviral vectors, HEK293FT cells were co-transfected with the lentiviral vector and the packaging plasmids psPAX2 (Addgene plasmid 12260) and pMD2.G (Addgene plasmid 12259) at a 0.5:0.25:0.25 ratio. Lipofectamine 3000 (Thermo Fisher Scientific) was used as the transfection reagent according to the manufacturer’s instructions. Viral supernatants were collected 48 h and 72 h after transfection, filtered through a 0.45-µm filter and used for infection in the presence of polybrene (8 µg ml⁻¹; EMD Millipore).

Mouse experiments

Intrasplenic and orthotopic injections of PDAC cells were performed as described previously²⁹. In brief, mice were anaesthetized by intraperitoneal administration of ketamine and xylazine. After anaesthesia, a small incision was made in the top-left region of the abdomen, and either the pancreas or the spleen was gently exteriorized. For intrasplenic injection, PDAC cells were suspended in 100 µl PBS and loaded into an insulin syringe (28-gauge needle; BD, 329461) pre-filled with 100 µl PBS. The externalized spleen was divided using ligating clips (Teleflex, 002200), and cells were injected into the distal hemispleen. After injection, the splenic vein was ligated at the splenic hilum using ligating clips (Teleflex, 001200), and the injected hemispleen was surgically removed. For orthotopic pancreatic implantation, PDAC cells were resuspended in 20 µl of a 1:1 mixture of Matrigel (Corning, 356231) and HBSS and injected into the pancreatic tail using insulin syringes fitted with a 29-gauge needle (BD, 324702). Unless otherwise specified, 5 × 10⁵ cells (for end-point assays) or 2.5 × 10⁵ cells (for survival analysis) were used for intrasplenic injections, whereas 5 × 10⁴ cells were injected for orthotopic pancreatic implantation.

For the orthotopic implantation followed by primary tumour resection (PTR) model (Fig. 5f–i), primary tumours were surgically resected 3 weeks after orthotopic implantation by distal pancreatectomy with splenectomy, essentially as described previously²¹. In brief, under anaesthesia, a sharp midline laparotomy was performed, with an optional left subcostal ‘hockey-stick’ extension as needed. The peritoneum was entered with electrocautery, and the pancreatic tail tumour was elevated without direct manipulation of the tumour mass by handling non-tumour pancreatic tissue (no-touch technique). Adhesions to surrounding structures were divided sharply or with electrocautery. Normal pancreas was identified and separated from gastric tissue, and two medium-sized ligating clips (Horizon, 002200) were applied proximal to the gross tumour edge to include draining vasculature. The pancreas was transected between the clips, and dissection was continued along the connective tissue plane between the tumour or spleen and the stomach until the superior splenic pole vasculature was encountered. Small-sized clips (Horizon 001200) were applied to the superior attachment between spleen and stomach, and the specimen was removed. The surgical field was confirmed to be haemostatic before closing the peritoneum and skin. Mice were euthanized at week 7 (4 weeks after PTR), and livers were collected for downstream analyses.

When the procedure was complete, the peritoneum was closed with 3-0 VICRYL violet sutures (Ethicon, J311H), and the skin incision was closed using the BD AutoClip wound closure system (BD). Mice were euthanized at the experimental end point, and livers or pancreatic tumours were collected for downstream analyses. For survival studies, mice were monitored daily and euthanized after reaching predefined humane end points in accordance with the Institutional Animal Care and Use Committee (IACUC) of New York University (NYU) Grossman School of Medicine.

For mouse experiments, C57BL/6J and NCr nude mice aged 8–10 weeks were used. Female mice were used in all experiments unless otherwise specified. Mice were fed either an AA control diet (ENVIGO, TD.01084) or a −SG diet (ENVIGO, TD.180296) for 15 days before tumour-cell implantation and were maintained on the same diet until the experimental end point. For pharmacological inhibition of CXCR2, mice were treated with the CXCR2 antagonist SB225002 (Selleck, S7651) or vehicle control by oral gavage at a dose of 1 mg kg⁻¹. The vehicle consisted of 2% DMSO, 30% polyethylene glycol (PEG) and 5% Tween 20. Drug administration began 48 h after tumour-cell implantation and continued throughout the study period until euthanasia, with dosing performed every other day.

The experiments in this study were performed in compliance with the NYU IACUC, under protocols IA16-00507 and IA16-01331. All mouse experiments were performed in a specific-pathogen-free conventional animal facility. Mice were maintained in pre-sterilized, disposable irradiated cages supplied with irradiated chow and acidified drinking water. Microisolator cages were housed on ventilated rack systems. Sample sizes were determined empirically on the basis of preliminary experiments, and no formal statistical power calculations were performed. Investigators were not blinded to group allocation, because knowledge of treatment groups was required for proper execution of the experimental procedures. Tumour weight, an objective end-point measurement, was assessed only at the experimental end point after euthanasia and tumour collection. In accordance with the approved IACUC protocol, mice bearing palpable tumours were regularly examined, and tumour dimensions were assessed by palpation. Mice were euthanized if a tumour exceeded 2 cm in any dimension. No pancreatic tumours reached this limit. For the liver-metastasis models, the IACUC protocol did not specify a maximum liver size or weight; mice were instead monitored for general condition and body-weight loss and were euthanized if their body weight decreased by more than 20% from baseline. This body-weight limit was not exceeded in any of the liver-metastasis experiments.

In vivo hepatocyte-specific gene knockout using AAV-based CRISPR

In vivo hepatocyte-specific knockout of Phgdh or Cxcr2 was performed as described previously¹² (Extended Data Fig. 2a). AAV vectors expressing sgRNAs were generated using the AAV2/8-U6-sgRNA-TBG-iCre-WPRE plasmid as the backbone. Plasmid construction and sgRNA cloning were performed by GenScript. The sgRNA target sequences were as follows:

sgScr (control): 5′-TACACGTCGCTAGGTTGCCC-3′

sgPhgdh: 5′-AGGTGCTCCCTACCAAGCCG-3′

sgCxcr2: 5′- TACGCAGTACGACCCTCAAA-3′.

Plasmids were submitted to the Viral Vector Core at the University of Iowa for AAV production. Rosa26-LSL-Cas9-EGFP mice on a C57BL/6J background were obtained from the Jackson Laboratory (strain 026175). Male and female homozygous Rosa26-LSL-Cas9-EGFP mice aged 6–8 weeks were used. Male mice were used only in the experiment shown in Extended Data Fig. 2e; female mice were used in all other experiments. Mice were injected retro-orbitally with AAV particles carrying either control sgRNA (sgScr) or sgRNAs targeting Phgdh or Cxcr2 at a dose of 5 × 10¹¹ genome copies per mouse. One week after AAV administration, Cas9-expressing mice were injected intrasplenically with PDAC cells for downstream analyses.

For validation of knockout efficiency, hepatocytes were isolated from mice seven days after AAV injection, and EGFP-positive hepatocytes were sorted using a MoFlo cell sorter (Beckman Coulter). Sorted cells were subsequently subjected to western blot analysis.

RNA extraction

Total RNA was extracted from cells cultured in vitro or isolated by cell sorting using the PureLink RNA Mini Kit (Thermo Fisher Scientific, 12183025), according to the manufacturer’s instructions. For transwell co-culture experiments, total RNA was extracted from hepatocytes and cancer cells plated in the bottom chamber of a six-well Boyden co-culture plate (Greiner Bio-One, 657610) using the same kit.

Immunoblotting

Whole-cell protein lysates were prepared from hepatocytes or cancer cells cultured under monoculture or co-culture conditions using radioimmunoprecipitation assay (RIPA) buffer (Sigma-Aldrich, 20-188) supplemented with protease and phosphatase inhibitor cocktails (Roche). Protein lysates were separated on 4–20% gradient gels (Bio-Rad, 4561096) and transferred to polyvinylidene fluoride (PVDF) membranes (Merck Millipore, IPVH00010) using Tris–glycine transfer buffer containing 10% methanol. Membranes were blocked with 3% bovine serum albumin (BSA; Sigma-Aldrich, A2058) for at least 1 h at room temperature and incubated with primary antibodies overnight at 4 °C at the following dilutions: PHGDH (1:1,000; Sigma-Aldrich, HPA021241), AKT (1:1,000; Cell Signaling Technology (CST), 4691), phospho-AKT (1:1,000; CST, 9271), CXCL5 (LIX) (mouse) (1:300; R&D Systems, MAB433-100), CXCR2 (1:1,000; Invitrogen, PA1-31217), FOXO3A (1:1,000; CST, FOXO3A (75D8) rabbit monoclonal antibody, 2497S), phospho-FOXO3A (Ser315) (1:500; Proteintech, 28755-1-AP) and phospho-FOXO3A (Ser253) (1:500; Invitrogen, PA5-36816) and β-actin (1:1,000; Sigma-Aldrich, A5441). Membranes were washed at least three times (15 min each) with Tris-buffered saline containing 0.1% Tween 20 (TBST) and incubated with horseradish peroxidase (HRP)-conjugated secondary antibodies (anti-mouse IgG (1:5,000; CST, 7076S) or anti-rabbit IgG (1:5,000; CST, 7074S)) for 1 h at room temperature. Signals were detected using an enhanced chemiluminescence (ECL) detection system (Bio-Rad, 1705061), and images were acquired using a ChemiDoc imaging system (Image Lab Touch Software, BIO-RAD, v.2.3.0.07). When reliable signals could not be obtained after membrane stripping and reprobing, equal amounts of protein lysate were loaded onto separate gels and processed in parallel. These instances are indicated in the legends to Supplementary Fig. 1, which contains the uncropped gel and blot source images.

qPCR

Total RNA was reverse-transcribed using SuperScript VILO IV Reverse Transcriptase (Thermo Fisher Scientific) with oligo(dT) primers to generate complementary DNA (cDNA). qPCR was performed using SYBR Green Supermix (Bio-Rad) using the CFX Manager software (Bio-Rad, v.3.1.1517.0823). Relative mRNA expression levels were calculated using the ΔCt method and normalized to Actb mRNA expression. The sequences of qPCR primers were as follows: Phgdh forward, 5′-GACGTGAACTTGGTGAACGC-3′; Phgdh reverse, 5′-AGACAGCTCCGTTGAGCATC-3′; Psat1 forward, 5′-GCCTGAGACAGCGAACCAATG-3′; Psat1 reverse, 5′-CATGGTGCTAAGGCGACAGC-3′; Psph forward, 5′-AACTGGTTCTCCCGTCATCG-3′; Psph reverse, 5′-CTCTTAAAAGCGCCGAACCG-3′; Actb forward, 5′-CGATATCGCTGCGCTGGTC-3′; Actb reverse, 5′-CCACGATGGAGGGGAATACAG-3′; Cxcl5 forward, 5′-TAAAAGGGGTGCAGTGGGTT-3′; Cxcl5 reverse, 5′-GAGCACCAGCTCGGGATATG-3′.

All primers were specific for mouse genes.

Bulk RNA-seq

Bulk RNA-seq library preparation, FASTQ generation, read alignment to the reference genome using HISAT2 (v.2.2.1), read mapping and differential gene-expression analysis using DESeq2 (v.1.44.0) were performed by Novogene. Differentially regulated genes (log₂-transformed fold change ≤ −2 and P < 0.05) were subjected to GSEA for pathway analysis. Enrichment scores and corresponding P values are reported in the figures.

GSEA

Differential expression results were converted into ranked gene lists for GSEA by selecting all genes with available log₂(fold change) values and assigning each a ranking metric equal to the log₂(fold change) from the DE analysis. The ranked list was saved as a two-column, tab-delimited text file with the first column containing gene symbols and the second containing the ranking metric, with no header line and unique ranking values.

The KEGG pathway ‘FoxO signalling pathway - Mus musculus (house mouse)’ (mmu04068) gene set was obtained from the KEGG database (https://www.genome.jp/dbget-bin/www_bget?path:mmu04068). The list of Mus musculus gene symbols was extracted from the ‘Genes’ section of the pathway entry and formatted into Gene Matrix Transposed (GMT) format, consisting of a single line with the pathway ID in the first column, a brief description in the second column and the gene symbols separated by tab characters in subsequent columns.

GSEA was performed in R (v.4.4.2) using the clusterProfiler package (v.4.14.6). Ranked gene lists were read into R and converted into named numeric vectors sorted in decreasing order of the ranking metric. The mmu04068 GMT file was read using read.gmt() and supplied as the TERM2GENE argument to the GSEA() function. All available genes were tested (pvalueCutoff = 1) to enable subsequent filtering. Enrichment plots were generated with the gseaplot2() function from the enrichplot package (v.1.26.6).

ORA

Raw gene-expression data were processed and differential expression analysis was performed using DESeq2. Genes were annotated with Entrez Gene IDs using the bitr function from the clusterProfiler R package and the org.Mm.eg.db mouse genome annotation database (v.3.20.0). Genes with P_adj < 0.05 and absolute log₂(fold change) > 1 were considered significantly differentially expressed.

To identify enriched biological pathways, ORA was performed using the enrichKEGG function from clusterProfiler, specifying the mouse organism (mmu). Resulting enriched pathways were filtered for significance at an FDR threshold of 5% ( P_adj < 0.05). Pathway annotations were converted to human-readable gene symbols using setReadable. Visualization of the top enriched KEGG pathways was performed with the dotplot function.

Cell sorting

For isolation of hepatocytes from mouse livers with or without liver metastases, crude hepatocyte suspensions obtained as described elsewhere in the Methods were further purified by fluorescence-activated cell sorting using a MoFlo cell sorter (Beckman Coulter). Dead cells were excluded by DAPI staining. In in vivo hepatocyte-specific knockout experiments, EGFP-positive hepatocytes that are indicative of successful Cas9-mediated gene editing were isolated by cell sorting for downstream analyses. For the isolation of GFP- or mCherry-expressing PDAC cells from mouse liver metastases, metastatic liver nodules were excised, mechanically minced with scissors and enzymatically digested in DMEM containing 1 mg ml⁻¹ collagenase IV (Gibco), 100 µg ml⁻¹ DNase I (Roche), 1% FBS, 10 mM HEPES and 2% antibiotic–antimycotic (Thermo Fisher Scientific) for 40 min at 37 °C in the dark, with gentle agitation every 10 min. Digested tissues were washed twice with DMEM containing 10% FBS and filtered through a 40-µm nylon mesh strainer (Corning). Cell suspensions were treated with ACK lysis buffer (Thermo Fisher Scientific) for 10 min at 4 °C in the dark to remove red blood cells, followed by sorting of GFP- or mCherry-positive PDAC cells using a MoFlo cell sorter (Beckman Coulter). In all sorting experiments, dead cells were excluded by DAPI staining. Sorted cells were used for protein extraction, RNA extraction or in vitro culture for the collection of CM.

Flow cytometry

EdU incorporation in PDAC cells grown in direct co-culture with mouse hepatocytes was assessed using the Click-iT Plus EdU Alexa Fluor 647 Flow Cytometry Assay Kit (Thermo Fisher Scientific, C10634), according to the manufacturer’s instructions. In brief, mCherry-labelled Phgdh-knockout HY19636 cells were mixed with mouse hepatocytes at a 1:1 ratio (1,500 cells each per well) and seeded into 12-well plates. After overnight incubation in complete DMEM, cells were washed with PBS and cultured in basal DMEM, with or without SG supplementation, for 48 h. Cells were then treated with 10 μM EdU for 60 min, collected by trypsinization and fixed. EdU incorporation was detected following the manufacturer’s protocol. Flow cytometric analysis was performed using a BD LSR II UV flow cytometer, and EdU uptake was quantified specifically in mCherry-positive PDAC cells.

Transwell co-culture experiments

For transwell co-culture experiments, primary mouse hepatocytes or AML12 cells were seeded onto collagen-coated plates at a density of 12,000 cells per well for 6-well plates or 3,000 cells per well for 24-well plates. PDAC cells (15,000 cells per well) were seeded separately onto 1.0-µm pore-size transwell inserts (Greiner Bio-One, 657610). In selected experiments, the configuration was reversed, with PDAC cells seeded in the bottom wells and hepatocytes seeded onto the transwell inserts.

After overnight attachment, cells were gently washed with PBS, and the culture medium was replaced with basal DMEM containing 2.78 mM glucose, 4 mM glutamine and 1 mM sodium pyruvate, supplemented with 10% dialysed FBS and either 400 µM (high), 150 µM (low) or 0 µM (null) serine and glycine. Co-cultures were maintained for the indicated time periods and subsequently underwent protein extraction, RNA extraction or intracellular metabolite analysis, as described below. For glucose tracing experiments, co-cultures were processed as described in ‘Metabolomics’ below.

Metabolomics

For ¹³C₆-glucose tracing experiments, PDAC cells and hepatocytes were plated either as monocultures or in transwell co-culture configurations, as described previously^30,31,32. After two washes with PBS, cells were cultured in basal DMEM without glucose for 12 h and subsequently incubated in basal DMEM supplemented with 25 mM ¹³C₆-glucose (Cambridge Isotope Laboratories, CLM-1396) for an additional 24 h.

To remove residual medium and contaminants, cells were rinsed with 0.9% NaCl prepared in high-performance-liquid-chromatography-grade water. Metabolites were extracted using 80% methanol containing 1 µg norvaline as an internal standard. AA standards (Cambridge Isotope Laboratories, MSK-A2-1.2) were included for metabolite identification and quantification. Cell extracts were vortexed for 15 min at 4 °C and centrifuged at maximum speed for 10 min to separate insoluble material. Polar metabolites were collected from the aqueous phase, and 300 µl of each sample was transferred into polypropylene vials (Agilent Technologies, 5190-2243) and dried under vacuum using a SpeedVac concentrator (Thermo Fisher Scientific Savant, SPD111V). Dried metabolites were derivatized by incubation with 20 µl methoxyamine hydrochloride (20 mg ml⁻¹ in pyridine; freshly prepared) for 60 min at 37 °C, followed by incubation with 20 µl MTBSTFA containing 1% tert-butyldimethylchlorosilane (t-BDMSC) for 30 min at 37 °C. Samples were analysed with a 7890B gas chromatograph coupled to a 5977B mass spectrometer (Agilent Technologies) equipped with a DB-35ms Ultra Inert column (Agilent Technologies, 122-3832UI) using Agilent MassHunter GC/MS Acquisition (v.B.07.05.2479). The GC–MS parameters, quantification and correction for natural isotope abundances were performed as described previously^31,33,34.

Growth assays

PDAC cells were seeded either as monocultures or in transwell co-culture with hepatocytes in 24-well plates as described in the previous section. Cells were cultured for 5 days or for the indicated time periods in medium containing high or low SG concentrations. At the end of the culture period, PDAC cells were washed with PBS, fixed with ice-cold methanol (pre-chilled at −20 °C) and stained with crystal violet solution (0.5% w/v in methanol). After incubation for 20 min at room temperature with gentle rocking, plates were washed thoroughly under running tap water and air-dried for 2 h or overnight. Crystal violet staining was then solubilized with 300 µl methanol per well and transferred to 96-well plates, and absorbance was measured at 570 nm using a SpectraMax M5 plate reader (Agilent) with SoftMax Pro (Molecular Devices, v.7.0).

Chromatin immunoprecipitation

Chromatin immunoprecipitation (ChIP) was performed using the SimpleChIP Enzymatic Chromatin IP Kit (Magnetic Beads) (CST, 9003) according to the manufacturer’s instructions, with minor modifications. AML12 hepatocytes were cultured alone or co-cultured with PDAC cells (Phgdh-KO HY19636 cells) using transwell assays (see ‘Transwell co-culture experiments’ for details) and treated as indicated. Experimental conditions included AML12 monoculture, AML12–PDAC co-culture and co-culture in the presence of CXCR2 or PI3K inhibitors. After treatment, cells were cross-linked, nuclei were isolated and chromatin was digested as per the kit protocol. After resuspension, the samples were sonicated in Diagenode Bioruptor for 15 cycles with 60 s on and 60 s off at 4 °C. For each immunoprecipitation, 5–10 μg of digested chromatin was incubated overnight at 4 °C with rotation with an anti-FOXO3 antibody (1:200, CST, 2497). Normal IgG supplied with the kit was used as a negative control. A 2% aliquot of chromatin was reserved as input control. Immunocomplexes were captured and washed sequentially with low- and high-salt buffers according to the manufacturer’s instructions. Chromatin was eluted, cross-links were reversed and DNA was purified using the spin columns provided with the kit. Purified DNA was analysed by qPCR using primers targeting putative FOXO-binding regions (TGTTT-core FOXO consensus-like motifs) in the Phgdh promoter (P1 and P2) and a distal gene-body region (P3; negative control)¹⁸ (Fig. 4h). Primer sequences: P1, Fw CTGGGCCAGAGAAGGGAAAG, Rv GCCTGTGCTGTTACCTCCAT; P2, Fw TGAGATTTAATTCCCTCGTGGAG, Rv GTGAGCTTTAACACGCACGAT; and P3, Fw ACAGTAAGGCGCTCAGTCAC, Rv TGGCTGGATTCAGTAACGGC). ChIP–qPCR signals were normalized to the 2% input sample and enrichment was expressed as the FOXO3 immunoprecipitation relative to IgG control (FOXO3/IgG).

Dual-reporter analysis

PHGDH promoter activity was assessed using the Secrete-Pair Gaussia luciferase/secreted alkaline phosphatase (GLuc/SEAP) dual luminescence assay system (GeneCopoeia, LF031) together with a GLuc reporter construct containing the mouse Phgdh promoter (GeneCopoeia, MPRM60654). Mouse hepatocyte AML12 cells, in which endogenous Foxo3a was knocked out and reconstituted with human wild-type FOXO3A or a FOXO3A(AAA) mutant, were transfected with the GLuc/SEAP reporter vectors using Lipofectamine 3000 transfection reagent (Thermo Fisher Scientific) according to the manufacturer’s instructions. Forty-eight hours after transfection, cells were co-cultured with PDAC cells for an additional 48 h in 12-well plates under conditions described elsewhere. CM (200 µl per well) was collected and transferred to 96-well plates for luminescence measurements. GLuc and SEAP activities were measured using a SpectraMax M5 plate reader (Agilent) with SoftMax Pro (Molecular Devices, v.7.0) following the manufacturer’s instructions. Each experimental condition was analysed using three biological replicates per group. GLuc luminescence values were normalized to the corresponding SEAP signals to control for transfection efficiency, and results were expressed as fold change relative to the wild-type group. The experiment was independently repeated twice.

D-serine and L-serine assay

For Fig. 3e, concentrations of D-serine and L-serine were measured using a DL-Serine Assay kit (Abcam, 241027) according to the manufacturer’s instructions. In brief, PDAC cells cultured under monoculture or co-culture conditions were collected and centrifuged, followed by pretreatment with the sample clean-up mix provided in the kit. Samples were then deproteinized and clarified by centrifugation through 3-kDa molecular weight cut-off filters (Amicon Ultra-0.5 Centrifugal Filter; MilliporeSigma, UFC5003). Processed samples, together with d-serine standards supplied with the kit, were subjected to fluorescence-based quantification using a SpectraMax M5 plate reader (Agilent) with SoftMax Pro (Molecular Devices, v.7.0) (excitation/emission = 535/587 nm). Concentrations of d-serine and l-serine were calculated according to the manufacturer’s protocol.

Isolation of TIF from mouse livers

Isolation of TIF was adapted from previously established protocols^35,36,37 with minor modifications. Freshly isolated mouse livers were cut into small pieces (1–3 mm³) in ice-cold PBS supplemented with a protease inhibitor cocktail. Tissue fragments were carefully transferred into 15-ml tubes using pre-chilled glass pipettes, washed briefly with PBS and resuspended in an equal volume of PBS relative to tissue volume. Samples were incubated at 37 °C in a CO₂ incubator for 1 h with occasional gentle tapping, followed by centrifugation at 1,000g for 8 min at 4 °C. The supernatant was transferred to a pre-chilled microcentrifuge tube using a pre-chilled glass pipette and centrifuged at 2,000g for 15 min at 4 °C. The resulting supernatant was further clarified by centrifugation at 20,000g for 30 min at 4 °C. The clarified supernatant was concentrated using a vacuum concentrator for 1 h to obtain TIF at a final protein concentration of approximately 2–5 mg ml⁻¹ per liver from C57BL/6 mice. TIF samples were isolated from three to five mice per group. Sample purity was assessed using a lactate dehydrogenase (LDH) assay kit (Abcam, ab102526) according to the manufacturer’s instructions. Samples exhibiting less than 10% cytotoxicity were snap-frozen, stored at −80 °C and used for subsequent analyses.

Enzyme-linked immunosorbent assay

The concentrations of CXCL5 and NGF in CM from PDAC cells, TIF derived from mouse tissues and tissue homogenates of human tissues were measured using ELISA kits for human CXCL5 (R&D Systems, DY254), mouse CXCL5 (R&D Systems, MX000) and mouse NGF (LSBio, LS-F5156), according to the manufacturers’ instructions. In brief, CM from mouse PDAC cells cultured with or without hepatocyte co-culture was collected 24 h after medium replacement and centrifuged at low speed to remove cellular debris. The resulting clarified supernatants were then subjected to enzyme-linked immunosorbent assay (ELISA). The preparation of TIF from mouse liver tissue and tissue homogenates from frozen human liver tissue is described elsewhere in the Methods.

For CXCL5 ELISA using homogenates of snap-frozen liver-metastasis tissues from autopsy samples (Fig. 3c), PHGDH status was assigned for each ELISA sample based on PHGDH immunostaining of corresponding metastatic lesions in available sections, and samples were grouped as PHGDH-strong versus PHGDH-weak/negative on the basis of PHGDH staining intensity in metastatic tumour cells. Because the ELISA samples were prepared from frozen tissue pieces, the set of samples used for ELISA did not necessarily match one-to-one with the FFPE specimens used for quantitative PHGDH evaluation in both tumour cells and hepatocytes (Fig. 1a,b). In some cases, although PHGDH staining was performed to categorize the frozen-tissue ELISA samples, the corresponding FFPE sections lacked sufficient hepatocytes to score hepatocyte PHGDH, and those cases were therefore excluded from the analyses in Fig. 1a,b.

Statistical analysis and reproducibility

Statistical analyses were performed using GraphPad Prism (v.9.0 and v.10.5.0). For comparisons between two groups, two-tailed unpaired Welch’s t-tests were used. For experiments involving more than two groups, one-way ANOVA followed by Tukey’s multiple-comparisons or Holm–Šídák’s post-hoc multiple-comparison tests was applied. P < 0.05 was considered statistically significant. For stacked isotopologue distribution plots, statistical significance was assessed for each isotopologue by two-way ANOVA followed by Tukey’s multiple-comparisons test. Exact P values are provided in the corresponding source data files. Data are presented as mean ± s.d. or as individual data points, as indicated in the figure legends. For bar graphs, each symbol represents one biologically independent sample unless otherwise stated. Immunoblot experiments were independently repeated at least twice with similar results. For in vivo studies, each intrasplenic or intrapancreatic injection cohort represents one independent experiment from multiple repeats; the number of biologically independent mice per group (n) is indicated in the corresponding figure legends, and each symbol represents one mouse.

Reporting summary

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

1. Inzunza, J. & Del Valle, A. C. Deciphering the liver’s role in pancreatic cancer metastasis: pathways and therapeutic approaches. npj Precis. Oncol. 9, 395 (2025).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

2. Link, J. M. et al. Ongoing replication stress tolerance and clonal T cell responses distinguish liver and lung recurrence and outcomes in pancreatic cancer. Nat. Cancer 6, 123–144 (2025).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

3. Ganguly, K. & Kimmelman, A. C. Reprogramming of tissue metabolism during cancer metastasis. Trends Cancer 9, 461–471 (2023).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

4. Krieg, S., Fernandes, S. I., Kolliopoulos, C., Liu, M. & Fendt, S. M. Metabolic signaling in cancer metastasis. Cancer Discov. 14, 934–952 (2024).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

5. Demicco, M., Liu, X. Z., Leithner, K. & Fendt, S. M. Metabolic heterogeneity in cancer. Nat. Metab. 6, 18–38 (2024).

   Article  PubMed  Google Scholar 

6. Banh, R. S. et al. Neurons release serine to support mRNA translation in pancreatic cancer. Cell 183, 1202–1218 (2020).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

7. Zhang, S. et al. Single cell transcriptomic analyses implicate an immunosuppressive tumor microenvironment in pancreatic cancer liver metastasis. Nat. Commun. 14, 5123 (2023).

   Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

8. Law, H. C. et al. The proteomic landscape of pancreatic ductal adenocarcinoma liver metastases identifies molecular subtypes and associations with clinical response. Clin. Cancer Res. 26, 1065–1076 (2020).

   Article  CAS  PubMed  Google Scholar 

9. Bardeesy, N. et al. Both p16^Ink4a and the p19^Arf-p53 pathway constrain progression of pancreatic adenocarcinoma in the mouse. Proc. Natl Acad. Sci. USA 103, 5947–5952 (2006).

   Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

10. Maddocks, O. D. K. et al. Modulating the therapeutic response of tumours to dietary serine and glycine starvation. Nature 544, 372–376 (2017).

    Article  ADS  CAS  PubMed  Google Scholar 

11. Platt, R. J. et al. CRISPR–Cas9 knockin mice for genome editing and cancer modeling. Cell 159, 440–455 (2014).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

12. Katsuda, T. et al. Rapid in vivo multiplexed editing (RIME) of the adult mouse liver. Hepatology 78, 486–502 (2023).

    Article  PubMed  Google Scholar 

13. Steele, C. W. et al. CXCR2 inhibition profoundly suppresses metastases and augments immunotherapy in pancreatic ductal adenocarcinoma. Cancer Cell 29, 832–845 (2016).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

14. Huang, H. & Tindall, D. J. Regulation of FOXO protein stability via ubiquitination and proteasome degradation. Biochim. Biophys. Acta 1813, 1961–1964 (2011).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

15. Santo, E. E. et al. FOXO3a is a major target of inactivation by PI3K/AKT signaling in aggressive neuroblastoma. Cancer Res. 73, 2189–2198 (2013).

    Article  CAS  PubMed  Google Scholar 

16. Ferber, E. C. et al. FOXO3a regulates reactive oxygen metabolism by inhibiting mitochondrial gene expression. Cell Death Differ. 19, 968–979 (2012).

    Article  CAS  PubMed  Google Scholar 

17. Litvak, V. et al. A FOXO3–IRF7 gene regulatory circuit limits inflammatory sequelae of antiviral responses. Nature 490, 421–425 (2012).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

18. Tao, R., Xiong, X., DePinho, R. A., Deng, C. X. & Dong, X. C. FoxO3 transcription factor and Sirt6 deacetylase regulate low density lipoprotein (LDL)-cholesterol homeostasis via control of the proprotein convertase subtilisin/kexin type 9 (Pcsk9) gene expression. J. Biol. Chem. 288, 29252–29259 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

19. Brunet, A. et al. Akt promotes cell survival by phosphorylating and inhibiting a Forkhead transcription factor. Cell 96, 857–868 (1999).

    Article  ADS  CAS  PubMed  Google Scholar 

20. Tong, H. et al. Dual impacts of serine/glycine-free diet in enhancing antitumor immunity and promoting evasion via PD-L1 lactylation. Cell Metab. 36, 2493–2510 (2024).

    Article  CAS  PubMed  Google Scholar 

21. Sorrentino, A. et al. Surgical resection of murine PDAC alters hepatic metastases and immune microenvironment. Pancreas 54, e831–e841 (2025).

    Article  CAS  Google Scholar 

22. Scott, A. J. et al. Rewiring of cortical glucose metabolism fuels human brain cancer growth. Nature 646, 413–422 (2025).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

23. Kamphorst, J. J. et al. Human pancreatic cancer tumors are nutrient poor and tumor cells actively scavenge extracellular protein. Cancer Res. 75, 544–553 (2015).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

24. Gheller, B. J. et al. Extracellular serine and glycine are required for mouse and human skeletal muscle stem and progenitor cell function. Mol. Metab. 43, 101106 (2021).

    Article  CAS  PubMed  Google Scholar 

25. Lim, E. W. et al. Serine and glycine physiology reversibly modulate retinal and peripheral nerve function. Cell Metab. 36, 2315–2328 (2024).

    Article  CAS  PubMed  Google Scholar 

26. Tajan, M. et al. Serine synthesis pathway inhibition cooperates with dietary serine and glycine limitation for cancer therapy. Nat. Commun. 12, 366 (2021).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

27. Karz, A. et al. MetFinder: a tool for automated quantitation of metastatic burden in histological sections from preclinical models. Pigment Cell Melanoma Res. 38, e13195 (2025).

    Article  CAS  PubMed  Google Scholar 

28. Charni-Natan, M. & Goldstein, I. Protocol for primary mouse hepatocyte isolation. STAR Protoc. 1, 100086 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

29. Yamamoto, K. et al. Autophagy promotes immune evasion of pancreatic cancer by degrading MHC-I. Nature 581, 100–105 (2020).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

30. Sousa, C. M. et al. Pancreatic stellate cells support tumour metabolism through autophagic alanine secretion. Nature 536, 479–483 (2016).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

31. Parker, S. J. et al. Selective alanine transporter utilization creates a targetable metabolic niche in pancreatic cancer. Cancer Discov. 10, 1018–1037 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

32. Encarnación-Rosado, J. et al. Targeting pancreatic cancer metabolic dependencies through glutamine antagonism. Nat. Cancer 5, 85–99 (2024).

    Article  PubMed  Google Scholar 

33. Hollinshead, K. E. R. et al. Respiratory supercomplexes promote mitochondrial efficiency and growth in severely hypoxic pancreatic cancer. Cell Rep. 33, 108231 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

34. Parker, S. J. et al. Spontaneous hydrolysis and spurious metabolic properties of α-ketoglutarate esters. Nat. Commun. 12, 4905 (2021).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

35. Celis, J. E. et al. Proteomic characterization of the interstitial fluid perfusing the breast tumor microenvironment. Mol. Cell. Proteomics 3, 327–344 (2004).

    Article  CAS  PubMed  Google Scholar 

36. Sun, W. et al. Characterization of the liver tissue interstitial fluid (TIF) proteome indicates potential for application in liver disease biomarker discovery. J. Proteome Res. 9, 1020–1031 (2010).

    Article  ADS  CAS  PubMed  Google Scholar 

37. Sun, W., Jiang, Y. & He, F. in Serum/Plasma Proteomics: Methods in Molecular Biology Vol. 728 (eds Simpson, R. J. & Greening, D. W.) 247–257 (2011).

Download references

Author notes

1. These authors contributed equally: Koelina Ganguly, Keisuke Yamamoto, Jason Rodencal

Authors and Affiliations

1. Department of Radiation Oncology, New York University Grossman School of Medicine, New York, NY, USA

   Koelina Ganguly, Keisuke Yamamoto, Jason Rodencal, Joel Encarnacion-Rosado, Albert S. W. Sohn, Douglas E. Biancur, Elaine Y. Lin, Ruohong Wang, Elshaddai Z. White, Richard Possemato & Alec C. Kimmelman

2. Laura and Isaac Perlmutter Cancer Center, New York University Grossman School of Medicine, New York, NY, USA

   Koelina Ganguly, Keisuke Yamamoto, Jason Rodencal, Joel Encarnacion-Rosado, Albert S. W. Sohn, Douglas E. Biancur, Elaine Y. Lin, Ruohong Wang, Elshaddai Z. White & Alec C. Kimmelman

3. Department of Gastroenterology, Graduate School of Medicine, University of Tokyo, Tokyo, Japan

   Keisuke Yamamoto & Mitsuhiro Fujishiro

4. Department of Biochemistry and Molecular Pharmacology, New York University Grossman School of Medicine, New York, NY, USA

   Carolina Alcantara Hirsch, Anthony Sorrentino & Dafna Bar-Sagi

5. Department of Surgery, New York University (NYU) Grossman School of Medicine and NYU Langone Health, New York, NY, USA

   Anthony Sorrentino

6. Eppley Institute for Research in Cancer and Allied Diseases, Fred & Pamela Buffett Cancer Center, University of Nebraska Medical Center, Omaha, NE, USA

   Paul M. Grandgenett & Michael A. Hollingsworth

7. Department of Preventive Medicine, Graduate School of Medicine, University of Tokyo, Tokyo, Japan

   Miwako Kakiuchi

8. Department of Pathology, New York University Grossman School of Medicine, New York, NY, USA

   Richard Possemato

Contributions

K.G., K.Y. and A.C.K. conceived the project and designed the experiments. K.G. performed most of the experiments and wrote the manuscript. J.R. performed additional experiments for the revision and contributed to writing the manuscript. M.K. and A.S.W.S. assisted with the isolation and culture of primary hepatocytes. J.E.-R. and R.W. assisted with mass-spectrometry experiments. J.E.-R., A.S.W.S., D.E.B., E.Y.L., C.A.H., A.S. and D.B.-S. assisted with mouse studies. E.Z.W. helped to generate Phgdh-knockout PDAC lines. P.M.G. and M.A.H. provided human PDAC liver-metastasis specimens. R.P. provided PHGDH constructs. M.F. provided intellectual support. K.Y. and A.C.K. supervised the research, provided intellectual feedback and wrote and critically edited the paper. All authors reviewed and approved the final manuscript.

Corresponding authors

Correspondence to Keisuke Yamamoto or Alec C. Kimmelman.

Competing interests

A.C.K. is an inventor on patents pertaining to KRAS-regulated metabolic pathways; redox control pathways in pancreatic cancer; targeting GOT1 as a therapeutic approach; alanine import inhibition; and the autophagic control of iron metabolism. A.C.K. is on the scientific advisory board of Cornerstone Pharmaceuticals (formerly Rafael Pharmaceuticals). E.Z.W. completed this work during his postdoctoral fellowship at NYU Grossman School of Medicine and is currently an employee of Merck Sharp & Dohme, a subsidiary of Merck; the opinions and perspectives expressed herein are those of the author (E.Z.W.) and do not represent the opinions or perspectives of his current employer. The remaining authors declare no competing interests.

Peer review information

Nature thanks the anonymous reviewers for their contribution to the peer review of this work.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions