Abstract Due to the rise in overnutrition, the incidence of obesity-induced hepatocellular carcinoma (HCC) will continue to escalate; however, our understanding of the obesity to HCC developmental axis is limited. We constructed a single-cell atlas to interrogate the dynamic transcriptomic changes during hepatocarcinogenesis in mice. Here we identify fatty acid binding protein 5 (FABP5) as a driver of obesity-induced HCC. Analysis of transformed cells reveals that FABP5 inhibition and silencing predispose cancer cells to lipid peroxidation and ferroptosis-induced cell death. Pharmacological inhibition and genetic ablation of FABP5 ameliorates the HCC burden in male mice, corresponding to enhanced ferroptosis in the tumour. Moreover, FABP5 inhibition induces a pro-inflammatory tumour microenvironment characterized by tumour-associated macrophages with increased expression of the co-stimulatory molecules CD80 and CD86 and increased CD8^+ T cell activation. Our work unravels the dual functional role of FABP5 in diet-induced HCC, inducing the transformation of hepatocytes and an immunosuppressive phenotype of tumour-associated macrophages and illustrates FABP5 inhibition as a potential therapeutic approach. __________________________________________________________________ HCC accounts for 90% of primary liver tumours and is the second leading cause of cancer-related death worldwide^[49]1,[50]2. In recent years, the elevated prevalence of obesity and metabolic dysregulation in industrialized societies has made obesity-induced HCC one of the most rapidly increasing cancers in the United States. Obesity can lead to HCC development through the chronic induction of non-alcoholic fatty liver disease (NAFLD), non-alcoholic steatohepatitis (NASH), fibrosis and HCC^[51]3. Conventionally, a ‘two-hit’ hypothesis has been proposed for NAFLD progression to HCC. First, excess dietary lipids and increased fatty acid (FA) synthesis lead to lipid accumulation in hepatocytes^[52]4. This imbalance in lipid accumulation triggers endoplasmic reticulum (ER) stress and necro-inflammatory responses, promoting the recruitment of adaptive immune cells (T cells, B cells and natural killer (NK) cells) by liver-resident macrophages^[53]5,[54]6. Continuous hepatic stress signals and cell death propagate chronic inflammation in the liver, producing mitogenic cytokines, such as interleukin (IL)-6 (refs. [55]7,[56]8). Thus, obesity-induced HCC arises in a unique environment of perpetual metabolic and inflammatory dysregulation, further amplifying the need to identify key specific drivers of HCC progression. Enhanced FA metabolism supported by exogenous lipid uptake and de novo lipogenesis is essential for maintaining cancer cell homoeostasis^[57]9. Using FAs, cancer cells demonstrate increased FA oxidation (FAO), which generates reducing power to alleviate oxidative stress and fuel their bioenergetic needs^[58]9,[59]10. Furthermore, cancer cells also frequently upregulate essential enzymes for lipid synthesis, such as FA synthase (FASN) or stearoyl-coenzyme A desaturase 1 (SCD1), which are crucial for generating phospholipids as building blocks for proliferation^[60]11,[61]12. Beyond bioenergetics, FA uptake and de novo lipogenesis can also potentiate oncogenic signalling during malignant transformation^[62]13. These multifaceted functions of FAs in cancer highlight the importance of generating obesity-induced HCC in animal models that can recapitulate metabolic syndromes observed in human pathophysiology. In recent years, dietary-based models have emerged as a promising alternative to studying the progression of obesity-driven HCC in mice^[63]14. In patients with NAFLD, choline deficiency exacerbates disease progression by inhibiting lipoprotein secretion and inducing oxidative stress^[64]15. In line with this, several reports have used the choline-deficient high-fat diet (CD-HFD) feeding to chronologically induce metabolic syndromes, NAFLD, NASH and HCC^[65]16–[66]18. Thus, the chronicity and recapitulative nature of this model make it an optimal approach to study the progression of obesity-induced HCC. Single-cell transcriptomics has emerged as a toolkit to dissect molecular alterations in diverse hepatic cell types^[67]19,[68]20. Much effort has been spent on delineating heterogeneous cell types, which have improved our understanding of their functions in homoeostasis and disease^[69]21. Despite this, none has attempted to construct an HCC progression atlas to delineate different cellular transcriptional signatures and potentially identify the genes associated with disease development. To this end, we generated a transcriptomic atlas for obesity-driven HCC using single-cell RNA sequencing (scRNA-seq) to find gene signatures and markers that are important during its progression. Through our approach, we were able to map the discrete transcriptional alterations in diverse cell types (such as hepatocytes, Kupffer cells (KCs), T cells and B cells) relevant for driving HCC progression. This transcriptomic atlas enabled the identification of distinctive patterns of gene expression induction through pseudotime inference^[70]22. By combining this approach with intrahepatic metabolic analysis, we also aimed to distinguish specific drivers of transformation across both parenchymal cell (PC) and non-parenchymal cell (NPC) types during HCC progression, paving the groundwork for identifying critical therapeutic targets. Here, using a detailed single-cell transcriptomic analysis of hepatic cells, we found FABP5 to be a central molecule involved in obesity-driven HCC in mice. Our findings reveal a bifunctional mechanism of FABP5 in HCC progression. In cancer cells, small molecule inhibition and genetic ablation of FABP5 predisposes transformed hepatocytes to cell death by enhancing lipid peroxidation-induced ferroptosis. In tumour-associated macrophages (TAMs), small molecule inhibition of FABP5 promotes a pro-inflammatory tumour microenvironment (TME) by stimulating the expression of the co-stimulatory receptors CD80 and CD86, thereby enhancing intratumoral T cell proliferation and cytotoxicity. Our work demonstrates the dual-targeting mechanisms of FABP5 inhibition, paving the foundation for future studies using this therapeutic approach. Results CD-HFD feeding promotes NAFLD–NASH progression in mice To study obesity-induced HCC, we used the CD-HFD dietary model that accelerates HCC development by inducing hepatic lipid accumulation, thereby exacerbating lipotoxicity. Based on previous studies, this dietary feeding strategy chronologically induces NAFLD, NASH and HCC in mice after feeding CD-HFD for 3, 6 and 12 months, respectively^[71]16–[72]18. We further characterized this model by studying the effect of chronic CD-HFD on metabolism (body weight, glucose metabolism and lipid metabolism) and inflammation (circulating and hepatic leucocytes, and fibrosis) ([73]Extended Data Fig. 1a). We found a persistent increase in body weight and fat mass in mice fed with CD-HFD compared with those fed a normal diet (ND) ([74]Extended Data Fig. 1b,[75]c). Histological analysis of haematoxylin and eosin (H&E) staining showed Mallory–Denk bodies (MDBs) and ballooned hepatocytes in mice fed with CD-HFD for 3 months, indicating severe chronic hepatitis caused by NAFLD ([76]Extended Data Fig. 1d). Oil Red O staining of liver sections revealed a marked accumulation of neutral lipids (triglycerides (TGs) and/or cholesterol esters (CEs)) after feeding mice with CD-HFD for 3 months ([77]Extended Data Fig. 1e). This was further confirmed by the observed elevation in intrahepatic TGs and CEs ([78]Extended Data Fig. 1f,[79]g). Long-term CD-HFD feeding also elevated total blood cholesterol ([80]Extended Data Fig. 1h). Detailed analysis of cholesterol distribution in different plasma lipoproteins using fast protein liquid chromatography showed cholesterol enrichment in the very-low-density lipoprotein (VLDL), low-density lipoprotein (LDL) and high-density lipoprotein (HDL) fractions under CD-HFD feeding conditions ([81]Extended Data Fig. 1i). We next evaluated whether hepatic lipid accumulation influenced glucose homoeostasis in CD-HFD-fed mice by measuring fasting glucose and performing a glucose tolerance test (GTT) and insulin tolerance test (ITT). CD-HFD feeding resulted in elevated fasting glucose ([82]Extended Data Fig. 1j) and impaired glucose tolerance after 3 or 6 months on the diet, respectively ([83]Extended Data Fig. 1k). CD-HFD-fed mice also developed an abnormal insulin response ([84]Extended Data Fig. 1l), indicating that chronic CD-HFD feeding causes glucose intolerance and insulin resistance. To further evaluate NAFLD–NASH progression, we first analysed hepatic inflammation by quantifying liver immune cells after 6 months of CD-HFD feeding. CD-HFD feeding induced a significant accumulation of leucocytes (CD45^+ cells) ([85]Extended Data Fig. 1m). Although we did not find differences in CD11b^medF4/80^high KC abundance, CD-HFD feeding led to an increase in CD11b^highF4/80^med monocyte-derived macrophages and activated (CD44^+CD62L^−) CD8^+ T cells ([86]Extended Data Fig. 1n). Furthermore, the severe hepatic steatosis and inflammation observed in mice fed with CD-HFD resulted in a substantial increase in hepatic fibrosis as indicated by the elevated collagen content assessed by Picrosirius red staining ([87]Extended Data Fig. 1o). Taken together, our data demonstrate that CD-HFD induces metabolic, inflammatory and fibrotic alterations associated with human NAFLD–NASH progression. CD-HFD induces transcriptomic alterations in the immune landscape during HCC progression We next characterized the molecular and cellular mechanism by which long-term CD-HFD can induce HCC ([88]Fig. 1a). To this end, we first assessed alpha fetal protein (AFP) circulating levels, commonly used as a prognostic marker of tumour development^[89]23. We found that ~50% of the mice on CD-HFD displayed high plasma AFP levels (>100 ng dl^−1) after 15 months of CD-HFD feeding. In contrast, none of the mice on ND demonstrated high circulating AFP ([90]Fig. 1b). In line with this result, mice with high AFP exhibited both macro- and microscopic evidence of tumour growth ([91]Fig. 1b (left) and [92]Extended Data Fig. 1p). Further analysis revealed different histological patterns among the tumours isolated from CD-HFD-fed mice, with distinctive lipid patterns ranging from steatotic to non-steatotic, irrespective of circulating AFP levels ([93]Extended Data Fig. 1q). Fig. 1 |. scRNA-seq uncovers dynamic changes in TCA cycle metabolic flux in obesity-induced HCC. Fig. 1 | [94]Open in a new tab a, Developmental timeline of obesity-induced HCC in the CD-HFD model. scRNA-seq was performed at the 15-month time point by (1) separating PC and NPC fractions; (2) enriching for live cells; and (3) combining at a 1:1 ratio and submitting for sequencing. b, Circulating AFP in C57BL/6 mice at 12 and 15 months of CD-HFD feeding. High AFP is defined as >100 ng dl^−1 and low AFP is defined as <100 ng dl^−1. n = 39 at 12 months and n = 32 at 15 months after CD-HFD feeding. c, UMAP representation of 13 distinctive cell types from 15-month CD-HFD or ND-fed mice. EC, endothelial cell; DC, dendritic cell. d, UMAP representation of four distinctive experimental conditions from 15-month CD-HFD or ND-fed mice. Low AFP corresponds to HCC-negative animals and high AFP corresponds to HCC-positive mice (liver and HCC were collected from the same high AFP mice). e, UMAP representation of five distinctive cell clusters from hepatocytes and cancer cells (left). The top five differentially expressed marker genes are shown for each cell cluster (right). hep, hepatocytes. f, Expression of AFP in hepatocytes and cancer cells by feature plot and violin plot (top right). g, Pseudotime value visualization by Monocle3 across hepatocytes and cancer cells from 15-month CD-HFD and ND-fed mice. Violin plot of pseudotime value in each experimental condition (top left). h, Differential genes as a function of pseudotime value were identified using Monocle3. Genes corresponding to mitochondrial metabolism are highlighted in red. i, GO analysis of upregulated and downregulated DEGs identified as a function of Monocle3 pseudotime progression. j, Schematic demonstrating metabolic flux measured by PINTA and ex vivo NMR. V[cs] flux and V[pc] to V[cs] ratio were evaluated in ND-fed control livers, high AFP CD-HFD-fed tumour-adjacent livers and high AFP CD-HFD-fed HCC. n = 3 for each experimental group. k, V[fao] flux analysed by PINTA and ex vivo NMR in ND-fed control livers, high AFP CD-HFD-fed tumour-adjacent livers and high AFP CD-HFD-fed HCC. n = 3 for each experimental group. Statistical analysis was conducted by non-parametric two-sided t-tests (j,k). P < 0.05 considered statistically significant. For b,j,k, each dot represents an individual animal and bar height indicates mean and s.e.m. Data represent two or more independent experiments (b,j,k). V[pdh], pyruvate dehydrogenase flux. By using scRNA-seq, we then sought to leverage this toolkit by sequencing the liver cells of three distinctive groups of mice (each group pooled from the liver region of four animals): C57BL6 mice fed with ND (group 1), HCC-negative mice fed with CD-HFD (group 2) and HCC-positive mice on CD-HFD (group 3) for 15 months. Our pre-sequencing process included isolating PCs and NPCs, followed by flow-sorting live cells before combining with PCs and NPCs at a 1:1 ratio for subsequent sequencing. After filtering out low-viability cells, we captured a total of 29,066 cells that clustered into 29 discrete cell populations ([95]Extended Data Fig. 2a,[96]b), from which we identified 13 unique cell types. ([97]Extended Data Fig. 2c,[98]d). Identified cell types predominantly included hepatocytes/transformed cells, endothelial cells, macrophages, KCs, T/NK cells and B cells ([99]Fig. 1c and differentially expressed genes (DEGs) available in [100]Supplementary Table 1). Delineating sequenced cells by disease progression revealed a distinctive transcriptional signature of PCs and NPCs in HCC, compared with non-tumour-bearing livers ([101]Fig. 1d). Notably, we found that PTPRC (CD45^+)-positive immune cells (T cells and macrophages) were greatly enriched intratumorally ([102]Extended Data Fig. 2e), compelling us to evaluate intratumoral immune phenotypes. An in-depth analysis of macrophages revealed six distinctive cell populations corresponding to monocyte-derived macrophages (MoMPs; MoMP1, MoMP2 and MoMP3) and KCs (KC1, KC2 and KC3) ([103]Extended Data Fig. 2f and DEGs shown in [104]Supplementary Table 2). MoMPs were characterized by high expression of monocytic infiltration markers such as CHIL3, LY6C and S100A4. On the contrary, the KC populations were marked by high CLEF4F, CD5L and PTGS1 expression. Comparison among experimental conditions found that TAMs were enriched for MoMP marker expression (predominately MoMP1) with few displaying KC signatures ([105]Extended Data Fig. 2g). We further validated the TAM gene signature by adapting pro- and anti-inflammatory macrophage profiles from recent work by Li and colleagues ([106]Extended Data Fig. 2h)^[107]20. Notably, HCC progression led to the continuous induction of pro-inflammatory and reduction of anti-inflammatory gene signatures. Validation by flow cytometry showed enrichment of CD11b^+Ly6C^+ TAMs, confirming the pro-inflammatory signature of intratumoral macrophages compared with those in the livers of ND-fed and tumour-adjacent livers of CD-HFD fed mice ([108]Extended Data Fig. 2i). Analysis of T/NK cells uncovered nine distinctive clusters (CD8_1, CD8_2, CD8_3, CD8_4, CD4_1, CD4_2, NK_1, NK_2 and NKT_1) ([109]Extended Data Fig. 2j and DEGs shown in [110]Supplementary Table 3). Additionally, tumour-infiltrating T cells displayed hallmarks of T cell exhaustion and dysfunction with elevated expression of established markers NR4A2, TOX and PDCD1 ([111]Extended Data Fig. 2k)^[112]24–[113]26. We verified these findings through flow cytometry, finding CD8^+ T cell enrichment ([114]Extended Data Fig. 2l) and elevated expression of TOX and PD1 in tumour-infiltrating CD8^+ T cells ([115]Extended Data Fig. 2m). Dynamic metabolic rewiring during hepatocyte–HCC progression Hepatocyte lipid metabolic dysfunction is a key driver of obesity-induced HCC^[116]27. Our transcriptomic dataset also allowed us to delineate key alterations during the hepatocyte–cancer transition. We found four clusters of non-transformed hepatocytes (Hep_1, Hep_2, Hep_3 and Hep_4) and 1 cluster of transformed hepatocytes (HepT_1) ([117]Fig. 1e and DEGs shown in [118]Supplementary Table 4). The HepT_1 cluster was enriched for cells from HCC ([119]Extended Data Fig. 3a,[120]b), as confirmed by high AFP expression ([121]Fig. 1f). Through evaluating hepatocyte periportal (ALB, CYP2F2, HPX, APOE and HSD17B13) and central (GLUL, AKR1C6, OAT, GSTM1 and ALDH3A2) signatures^[122]28, we found that Hep_2 closely resembles periportal hepatocytes, whereas Hep_4 more resembles central hepatocytes ([123]Extended Data Fig. 3c). Of note, this suggests that HepT_1 is more similar to central hepatocytes, with a detectable expression of central and minimal expression of periportal signatures. To infer transcriptomic changes during the hepatocyte–cancer transition, we performed in silico pseudotime trajectory analysis (Monocle3) by ordering cells along a continuous spectrum from hepatocytes to cancer-transformed hepatocytes ([124]Fig. 1g)^[125]22,[126]29. We validated our strategy by inferring the pseudotime value (PTV) of livers from ND, CD-HFD and HCC. In line with expectations, we observed a continuous increase in PTV from ND to CD-HFD and HCC, with transformed cells displaying the highest PTV. We further checked for upregulation of glycolytic markers (ENO1, FBP1 and GAPDH). Cells with higher PTV demonstrated higher expression of these signatures associated with HCC development ([127]Extended Data Fig. 3c). We also found that mitochondrial-related genes (MT-CO1, MT-CYTB and MT-NO3) were downregulated, suggesting mitochondrial respiration abnormalities in HCC ([128]Extended Data Fig. 3d). To screen for markers of HCC progression, we next identified transcripts that change in expression as a function of PTV ([129]Fig. 1h). We found a total of 378 upregulated and 142 downregulated genes ([130]Supplementary Table 5). Gene Ontology (GO) analysis of DEGs revealed several pathways that were altered during HCC progression ([131]Fig. 1i). Notably, we uncovered that FA uptake and electron transport-coupled (ETC) proton transport were within the metabolic pathways that were downregulated. These changes suggest that transformed cells shift away from mitochondrial respiration towards alternative means of ATP generation. We validated our PTV analysis using the Slingshot package to recluster sequenced hepatocytes before constructing a smooth lineage to calculate one-dimensional PTV^[132]30. Slingshot analysis reorganized non-transformed and transformed hepatocytes into a single trajectory with five distinctive clusters, each with its distinctive PTV ([133]Extended Data Fig. 3e,[134]f). As expected from previous Monocle analysis, we found that transformed cancer cells reside almost exclusively in cluster 5, corresponding to high PTV ([135]Extended Data Fig. 3g). We then performed differential analysis for transcripts that change as a function of Slingshot PTV to identify the top 250 upregulated and downregulated genes ([136]Supplementary Table 6). GO analysis of DEGs found that peptide biosynthetic processes and cellular component biogenesis were upregulated ([137]Extended Data Fig. 3h) and oxidative phosphorylation and lipid metabolic pathways were downregulated ([138]Extended Data Fig. 3i) during HCC progression. To further assess the changes in respiration, we utilized a tracer analysis method to evaluate the in vivo HCC metabolic flux^[139]31. A [3-^13C] lactate and [1,2,3,4,5,6,6-^2H[7]] glucose tracer was infused in tumour-bearing mice, followed by gas chromatography–mass spectrometry (GC–MS) and liquid chromatography–tandem MS (LC–MS/MS)-based analysis of tissue samples. Hepatic anaplerosis, citrate synthase flux (V[cs]), and glucose production were measured. This enabled the non-invasive measurement of enzymes that regulate the citric acid (TCA) cycle and informed the rate of TCA influx (V[cs]) and TCA efflux ([pyruvate carboxylase flux]/[citrate synthase flux], V[pc]/V[cs]). We found that HCC demonstrated reduced TCA influx (V[cs]) and increased TCA efflux (V[pc]/V[cs]) as compared with ND-fed control liver ([140]Fig. 1j). We further extrapolated the contribution of FAs to TCA cycle flux (V[fao]) and found a similar reduction in FAO rate after HCC development ([141]Fig. 1k). Together, our transcriptomic atlas illustrates the dynamic metabolic perturbations that accompany HCC development and outlines a shift away from mitochondrial-dependent energy generation in HCC. FABP5 is specifically upregulated and associated with oncogenes in obesity-induced HCC We then used our transcriptomic atlas to delineate cancer-specific related genes in HCC. By comparing AFP^+ transformed cells to AFP^− hepatocytes across all sequenced samples, we found 829 upregulated and 433 downregulated genes ([142]Fig. 2a and [143]Supplementary Table 7). Transcripts previously described to be induced upon cancer development, such as AKR1C18, GLUL and TSPAN8 (refs. [144]32,[145]33), were found among the upregulated genes. Notably, AKR1C18 was also highly expressed in tumours of the diethylnitrosamine (DEN)-induced HCC model, demonstrating overlap in oncogenic induction across different HCC animal models. Conversely, transcripts of hepatocyte-specific gene signatures, such as CPS1, SERPINA3K and APOA5, were downregulated in transformed cells, suggesting de-differentiation of the hepatocyte lineage^[146]34. We then performed Gene Set Enrichment Analysis (GSEA) of differential gene signatures in AFP^+-transformed cells ([147]Fig. 2b). Among the upregulated pathways, we found the non-canonical NF-κB pathway, which has been established to be an important driver of HCC^[148]35. FA metabolism was suppressed in transformed cells within the downregulated pathways, further supporting a switch away from mitochondrial respiration. Fig. 2 |. FABP5 is specifically upregulated and associated with worse survival in HCC. Fig. 2 | [149]Open in a new tab a, Volcano plot of DEGs in AFP-positive cancer cells compared with AFP-negative hepatocytes after 15 months of CD-HFD feeding. Select upregulated and downregulated genes are indicated. b, GSEA of significantly upregulated and downregulated genes in AFP-positive cancer cells from 15-month CD-HFD fed mice. GAP, GTPase-activating-protein; IQGAP, Ras GTPase-activating-like protein. c, FABP5 mRNA expression in healthy human liver and liver HCC from TCGA database. Data points indicate human patients per group; n = 369 for the HCC group and n = 160 for the liver group. HCC group: minima, 1.499; maxima, 8.807; centre, 4.18; box, 3.402–4.829; whisker, 1.5–6.919. Control group: minima, 0.2667; maxima, 7.077; centre, 1.946; box, 1.528–2.767; whisker, 0.2667–3.964. d, Correlation of FABP5 expression with overall survival in patients with HCC from TCGA database. Data extracted from 369 patients with HCC with a 50% cutoff for high and low FABP5 expression. e, IF imaging of FABP5 and stratification of FABP5 expression patterns in 11 HCC tumours from 15-month CD-HFD fed mice. f, IHC imaging of FABP5 in 11 HCC tumours from 15-month CD-HFD fed mice. Scale bars, 57 μm (e) and 20 μm (f). Statistical analyses used were Wilcox test, no adjustments made (a); a two-sided Mann–Whitney U-test (c); and a log-rank Mantel–Cox test (d). P < 0.05 was considered statistically significant. Data in e,f represent two or more independent experiments. Notably, the FABP5 transcript was significantly upregulated in AFP^+-transformed cells ([150]Fig. 2a), whereas it was minimally expressed in non-transformed hepatocytes. FABP5 messenger RNA was only detectable upon cancer progression and consistently coexpressed with AFP. Next, we evaluated the top 20 upstream transcriptional regulators of FABP5^+ cancer cells ([151]Supplementary Table 8). According to ingenuity pathway analysis (IPA), c-Myc, which has been described to transactivate FABP5 expression upon hypomethylation of its promoter region in prostate cancer^[152]36, was predicted to be activated in FABP5^+AFP^+ transformed cells. To further evaluate FABP5 upregulation in HCC, we reanalysed single-cell transcriptomic data generated from a carcinogen-induced DEN^+ carbon tetrachloride (CCL[4]) HCC model^[153]37 at 3, 10 and 30 days after tumour induction^[154]38. Like our CD-HFD-induced HCC, DEN^+CCL[4]-induced HCC demonstrates elevated expression of AFP and FABP5 upon tumour initiation ([155]Extended Data Fig. 3j–[156]l). Of interest, the expression pattern of FABP5 is more ubiquitous in DEN^+CCL[4] than in CD-HFD-induced HCC, demonstrating increased heterogeneity among transformed cancer cells in a diet-induced HCC model. We also observed a reduction in the expression of mitochondrial genes in DEN^+CCL[4]-induced HCC ([157]Extended Data Fig. 3m,[158]n), aligning with our findings of reduced mitochondrial gene expression during HCC development. Next, we examined the correlation of FABP5 expression with survival in human HCC, using Gene Expression Profiling Interactive Analysis (GEPIA)^[159]39. The Cancer Genome Atlas (TCGA) data analysis demonstrated FABP5 overexpression in liver hepatocellular carcinoma compared with non-cancerous livers ([160]Fig. 2c). In addition, patients with FABP5^high expression showed a significantly lower 5-year overall survival rate compared with FABP5^low patients ([161]Fig. 2d), thus supporting a potential oncogenic function of FABP5 in both, mouse and human HCC development. FABPs are a family of lipid chaperones that reversibly bind saturated and unsaturated long-chain FAs to shuttle them within different cell compartments, altering cellular FA distribution and metabolism^[162]40. The specific FABP5 cellular localization is particularly important to dictate its potential oncogenic role as different functions for FABP5 have been reported when localized in the nucleus, mitochondria or plasma membrane^[163]41,[164]42. Therefore, we assessed the expression and localization of FABP5 by combined scRNA-seq, immunofluorescence (IF) and immunohistochemistry (IHC) analysis of HCC samples. The scRNA-seq analysis revealed detectable FABP5 expression in AFP^+ cancer cells and a subpopulation of CSF1R^+ mononuclear phagocytes ([165]Extended Data Fig. 4a,[166]b). This was confirmed by IF analysis, where we found a marked co-staining of FABP5 with AFP^+ cancer cells ([167]Extended Data Fig. 4c) and with CD68^+ mononuclear phagocytes ([168]Extended Data Fig. 4d). The detection of FABP5 in mononuclear phagocytes is consistent with its high expression in KCs and lipid-rich atherosclerotic macrophages^[169]43,[170]44. It is particularly noteworthy as its function in macrophages is not yet elucidated. By IF analysis, we identified four cytoplasmic expression patterns of FABP5 across sections from 11 tumour samples ([171]Fig. 2e), ranging from minimal to ubiquitous expression in transformed cancer cells and immune cells. Of interest, ubiquitous FABP5 expression in HCC parallels expression patterns from the fetal liver, suggesting a potential correlation between FABP5 expression and HCC de-differentiation. We corroborated these expression patterns by IHC and phalloidin staining ([172]Fig. 2f and [173]Extended Data Fig. 4e), which further supports the cytoplasmic expression of FABP5 in cancer and immune cells. We next sought to evaluate FABP5 expression relative to the boundaries and centre of HCC by using scanning microscopy to visualize the whole tumour ([174]Extended Data Fig. 4f–[175]h). At tumour boundaries, we found minimal FABP expression in the tumour-adjacent liver, further supporting FABP5’s tumour-specific upregulation. Within the tumour, we did not observe changes in FABP5 expression relative to the tumour centre, demonstrating the ubiquitous nature of FABP5 expression in HCC. FABP5 suppression sensitizes cancer cells to lipid peroxidation, ER stress and ferroptosis FABP5 regulates different molecular processes that can contribute toward HCC development, including activation of PPAR signalling, regulation of mitochondrial metabolism and altering intracellular lipid composition^[176]4,[177]41. All these distinctive mechanisms could contribute to the progression of HCC. To dissect the precise molecular processes FABP5 may regulate in HCC development, we performed RNA sequencing on a human hepatocyte-derived carcinoma cell line (Huh7) treated with either FABP5 inhibitor or FABP5 siRNA ([178]Fig. 3a). We used a competitive small molecule inhibitor (SBFI-103), which exhibits a high affinity for FABP5 compared with other FABP family members^[179]45, at a concentration of 5 μM for 48 h. Fig. 3 |. FABP5 inhibition sensitizes cancer cells to lipid peroxidation, ER stress and ferroptosis. Fig. 3 | [180]Open in a new tab a, Schematic outlining SBFI-103 and FABP5 siRNA treatment regimen on Huh7 cells before RNA sequencing. b, Overlap of 467 significantly upregulated genes and 524 downregulated genes from inhibitor-treated and siRNA-treated Huh7 cells. c, GO analysis of overlapping upregulated and downregulated DEGs identified by RNA sequencing of inhibitor-treated and siRNA-treated Huh7 cells. d, Heatmap showing cell-cycle phase transition genes in 5 μM SBFI-103-treated Huh7 cells. e, Heatmap showing PERK-mediated UPR genes in 5 μM SBFI-103-treated Huh7 cells. f, MDA concentration in SBFI-103-treated and vehicle-treated Huh7 cell lines as measured through colorimetric assay. n = 3 for both experimental groups. g, Flow cytometry analysis of BODIPY 581/591 C11 mean fluorescent intensity (MFI) in 5 μM SBFI-103-treated Huh7 cells. n = 3 for both experimental groups. h, Flow cytometry analysis of CellRox MFI in 5 μM SBFI-103-treated Huh7 cells. n = 3 for both experimental groups. i, Western blot of PERK, ATF4, IRE1A and BIP in 5 μM SBFI-103-treated and control Huh7 cells. n = 3 for both experimental groups. Statistical analysis was conducted with non-parametric two-sided t-tests (f–h); P < 0.05 was considered statistically significant. For f–h, each dot represents an individual animal and the bar height indicates the mean and s.e.m. Data (f–i) represent two or more independent experiments. Notably, SBFI-103 treatment did not affect FABP5 mRNA levels ([181]Extended Data Fig. 5a). Principal-component analysis (PCA) revealed a distinctive grouping of inhibitor- and control-treated cells ([182]Extended Data Fig. 5b). GSEA showed downregulation of chromosomal segregation and cell division processes in inhibitor-treated cells ([183]Extended Data Fig. 5c,[184]d). As expected, siRNA treatment significantly reduced FABP5 mRNA levels ([185]Extended Data Fig. 5e) and PCA of siRNA-treated samples also demonstrated clear separation based on experimental conditions ([186]Extended Data Fig. 5f). GSEA of DEGs from siRNA-treated cells revealed that FABP5 silencing is an important regulator of cellular de-differentiation ([187]Extended Data Fig. 5g,[188]h). Our dataset identified 467 overlapping upregulated genes and 524 overlapping downregulated genes in SBFI-103 inhibitor- and FABP5 siRNA-treated cells ([189]Fig. 3b and [190]Supplementary Table 9). We next performed GO analysis on these overlapping regulated genes. Pathways such as long-chain FA biosynthesis and PERK-mediated unfolded protein response were upregulated, whereas biological processes related to cell-cycle proliferation and differentiation were downregulated ([191]Fig. 3c–[192]e). Thus, we evaluated whether further FA metabolism alterations induced by FABP5 modulation could lead to ER stress and/or cell death. Ferroptosis, a form of cell death mediated by iron-dependent lipid peroxidation, connects the many differential pathways in our RNA sequencing analysis. As ferroptosis occurs upon extensive oxidation of polyunsaturated phospholipids, leading to the accumulation of oxidatively damaged lipid peroxides^[193]46,[194]47, we examined the effect of FABP5 inhibition or silencing on FA uptake and peroxidation in Huh7 cells. Fluorescent lipid (BODIPY FL C16) uptake was increased upon SBFI-103 or FABP5 siRNA treatment ([195]Extended Data Fig. 5i), whereas glucose uptake was not affected ([196]Extended Data Fig. 5j). We then measured malondialdehyde (MDA) concentration, one of the final products of polyunsaturated FA peroxidation. As shown in [197]Fig. 3f and [198]Extended Data Fig. 5k, Huh7 cells treated with SBFI-103 or FABP5 siRNA exhibited increased overall MDA concentration compared with their respective controls. Next, we quantified lipid peroxidation in FABP5 inhibition or silencing conditions using BODIPY 581/591 C11 ([199]Fig. 3g and [200]Extended Data Fig. 5l). We found that SBFI-103 or FABP5 siRNA treatment enhanced BODIPY 581/591 C11 intensity, indicating increased lipid peroxidation upon treatment. Additionally, we found elevated reactive oxygen species (ROS) production in Huh7 cell lines by CellROX assay ([201]Fig. 3h and [202]Extended Data Fig. 5m) upon treatment of FABP5 inhibitor or siRNA. Given that a hallmark of lipid peroxidation and ROS accumulation is the induction of the unfolded protein response (UPR), we determined the activation of signalling proteins that result in distinctive signalling pathways leading to cell death^[203]48, such as inositol requiring protein-1 (IRE1) and protein kinase RNA-like ER kinase (PERK). Western blot analysis of SBFI-103- and FABP5 siRNA-treated Huh7 cells found higher protein expression of PERK, ATF4, IRE1A and BIP, all central regulators of ER stress ([204]Fig. 3i and [205]Extended Data Fig. 5n). Taken together, our results indicate that FABP5 inhibition can regulate cancer viability by potentiating lipid peroxidation-induced ferroptosis and UPR. FABP5 inhibition ameliorates obesity-induced HCC progression To dissect whether FABP5 contributes to HCC progression, we next inhibited FABP5 function using SBFI-103 and assessed HCC burden ([206]Fig. 4a). After 12 months of CD-HFD feeding, we administered SBFI-103 once every 2 days for 3 months. Using circulating AFP levels as a prognostic marker before and after treatment, we found that vehicle-treated mice maintained high AFP levels after 15 months, whereas SBFI-103-treated mice showed a marked reduction in circulating AFP ([207]Fig. 4b). Notably, we found that ~50% (10 of 19) of control mice developed HCC, whereas only ~30% (6 of 20) of SBFI-103-treated animals exhibited signs of HCC, as indicated by the evaluation of visible tumour nodules ([208]Fig. 4c). Furthermore, SBFI-103 treatment also reduced the number of tumour nodules ([209]Fig. 4d) as well as the total tumour volume ([210]Fig. 4e). We found that inhibitor treatment significantly reduced the abundance of FABP5^+ cells in the tumour and concurrently increased terminal deoxynucleotidyl transferase dUTP nick end labelling (TUNEL)^+ cells, suggesting that SBFI-103 treatment led to the specific ablation of FABP5^+ cancer cells ([211]Fig. 4f). We next assessed the impact of SBFI-103 treatment on the expression of established oncogenes. Notably and in line with results observed in plasma, FABP5 inhibition reduced hepatic AFP levels ([212]Fig. 4g) and restored the expression of tumour suppressor genes, such as PTEN and RB^[213]49. We further examined the effect of SBFI-103 on the Hippo pathway, a well-established driver of HCC progression and regulator of de-differentiation^[214]50,[215]51. We focused on Yes-associated protein (YAP), a potent modulator of gene expression by binding to other transcription factors. Depending on its binding partner and subcellular localization, YAP might act as an oncogene or tumour suppressor. In this regard, translocation to the nucleus is critical for pathway activation and oncogenic activity^[216]52. We found that SBFI-103 treatment reduced total YAP protein levels, while increasing its phosphorylation (S127P YAP), which is a marker of cytoplasmic retention ([217]Fig. 4g). In agreement with this finding, the expression of bona fide YAP response genes assessed by scRNA-seq analysis, including CTGF, ANKRD1, IGFBP3, F3, TGFB2, PTPN14 and MYOF^[218]53 were markedly downregulated in cancer cells isolated from SBFI-103 mice when compared with transformed cells of control mice ([219]Fig. 4h). Fig. 4 |. FABP5 inhibition ameliorates HCC progression. Fig. 4 | [220]Open in a new tab a, Schematic outlining inhibitor constant (K[i]; μM) of SBFI-103 in FABPs and treatment regimen for FABP5 inhibition in CD-HFD induced HCC. i.p., intraperitoneal. b, Circulating AFP in CD-HFD fed mice before (12 months) and after (15 months) treatment with SBFI-103. Lines connect before- and after-treatment AFP values from the same animal. n = 13 for both experimental groups. c, Quantification of HCC-positive mice after 3 months of SBFI-103 treatment. n = 19 for vehicle-treated mice and n = 20 for SBFI-103-treated mice. d, Quantification of HCC nodules per liver in tumour-bearing mice after 3 months of SBFI-103 treatment. n = 10 for vehicle-treated mice and n = 6 for SBFI-103-treated mice. e, Quantification of HCC volume in tumour-bearing mice after 3 months of SBFI-103 treatment. n = 10 for vehicle-treated mice and n = 6 for SBFI-103-treated mice. f, IF imaging and quantification of FABP5 and TUNEL staining in control and SBFI-103-treated HCC. n = 4 for both experimental groups. g, Western blot analysis of AFP, YAP, P-YAP (Ser127), PTEN and RB in SBFI-103-treated HCC, control HCC and control liver region. n = 3 for each experimental group. h, Dot-plot expression of AFP, YAP1, FABP5 and YAP response genes from single-cell analysis from SBFI-103-treated and control HCC. Scale bars, 100 μm (f). Statistical analysis was carried out by non-parametric two-sided t-tests (d–f); P < 0.05 was considered statistically significant. For d–f, each dot represents an individual animal and the bar height indicates the mean and s.e.m. Data (f,g) represent two or more independent experiments. SBFI-103 treatment reduces HCC by promoting FAO, lipid peroxidation and ferroptosis We next examined how pharmacological inhibition of FABP5 affects the transcriptional profile of tumour cells using scRNA-seq of liver tumours isolated from 15-month CD-HFD-fed mice (each group pooled from liver regions of three animals). A total of 1,358 transformed hepatocytes were sequenced, corresponding to four unique clusters (HepT_1, HepT_2, HepT_3 and HepT_4) ([221]Fig. 5a). Transformed cells from control-treated mice correspond to HepT_4, a unique population that exhibits both high FABP5 and AFP expression. Of note, transformed cells from SBFI-103-treated mice were absent from the HepT_4 cluster and resided exclusively in HepT_1 ([222]Fig. 5b). We next performed a differential transcriptional comparison analysis between transformed cells from SBFI-103 and control-treated mice ([223]Fig. 5c and [224]Supplementary Table 10). We noticed that hepatocyte-derived cancer cells from inhibitor-treated mice had higher expression of genes associated with energy expenditure, such as Oat, Glul and Rgn. SBFI-103 treatment also induced the expression of FABP1 and FABP2, suggesting a compensatory rescue of FABP family proteins upon inhibiting FABP5 activity ([225]Extended Data Fig. 6a). We then performed a GSEA comparison of DEGs and found that pathways associated with mitochondrial respiration were upregulated in SBFI-103-treated conditions ([226]Fig. 5d). Notably, mitochondrial FAO and respiratory electron transport were both upregulated ([227]Extended Data Fig. 6b). In contrast, DNA replication-related pathways were downregulated in the inhibitor treatment condition, suggesting reduced cell division. Fig. 5 |. FABP5 inhibition restores HCC FAO and promotes lipid peroxidation-induced ferroptosis. Fig. 5 | [228]Open in a new tab a, UMAP representation of four distinctive cancer clusters from SBFI-103- or vehicle-treated HCC (left). The top five signature genes from each cluster are shown via heatmap (right). b, UMAP representation showing sequenced cancer cells belonging to SBFI-103- or vehicle-treated conditions. c, Volcano plot of DEGs in SBFI-103-treated cancer cells compared with vehicle-treated cancer cells. Select upregulated and downregulated genes are indicated. d, GSEA of significantly upregulated and downregulated genes in SBFI-103-treated cancer cells. e, EM analysis and quantification of mitochondria from the vehicle-treated liver, vehicle-treated HCC and SBFI-103-treated HCC (left). Probability curves depict the distribution of mitochondrial aspect ratio (right). n = 3 for each experimental group. f, V[fao] flux analysed by PINTA and ex vivo NMR in vehicle-treated liver, vehicle-treated HCC and SBFI-103-treated HCC. n = 3 for each experimental group. g, OCR was measured after 8 h of treatment with 5 μM SBFI-103 in Huh7 cells with and without 20 μm etoximir (Eto) by Seahorse extracellular flux analyzer. Basal and spare respiratory capacity were quantified after adding 1 μm oligomycin, 2 μm FCCP and 2.5 μm rotenone + antimycin. n = 8 for each experimental condition. h, MDA concentration in SBFI-103-treated and vehicle-treated HCC as measured through colorimetric assay. n = 3 for each experimental group. i, Western blot of ACSL4, TFR1 and CHOP in SBFI-103-treated HCC, vehicle-treated HCC and vehicle-treated liver. n = 3 for each experimental group. j, RT–PCR analysis of PTGS2 and ASNS in SBFI-103-treated HCC, vehicle-treated HCC and vehicle-treated liver. n = 3 for each experimental group. k, RT–PCR analysis of unspliced (us) and spliced (s) forms of XBP1 in SBFI-103-treated and vehicle-treated HCC. n = 3 for each experimental group. l, Representative Perls Prussian blue staining of iron content in SBFI-103-treated HCC and vehicle-treated HCC. m, DHE staining (left) and quantification (right) of ROS content in SBFI-103-treated and vehicle-treated HCC. n = 3 for each experimental group. Scale bars, 2 μm (e), 40 μm (l) and 20 μm (m). Statistical analyses were non-parametric two-sided t-tests (e–h,j,k,m) and a Wilcox test with no adjustments made (c); P < 0.05 was considered statistically significant. For e–h,j,k, each dot represents an individual animal and the bar height indicates the mean and s.e.m. Data (g–m,l) represent two or more independent experiments. Next, we examined the functional consequences of altered mitochondria respiration gene signatures in mice treated with FABP5 inhibitor. Mitochondrial morphology strongly influences tumour metabolism and proliferation, with spherical mitochondria associated with a rapidly dividing cell phenotype^[229]54. FABP5 can regulate mitochondrial integrity and FA respiration, which is responsible for tumour metabolic changes and anti-tumoural immune activation^[230]55. We used electron microscopy (EM) to examine mitochondrial size and sphericity in the tumour and liver region of mice treated with or without SBFI-103. Of note, we observed no differences in mitochondrial size across treatment conditions ([231]Extended Data Fig. 6c). In control-treated mice, we found that the tumour region demonstrated increased sphericity in mitochondrial morphology compared with the adjacent liver, indicative of mitochondrial dysfunction. Of note, FABP5 inhibition resulted in more tubulated mitochondrial in the tumour region, with a strong resemblance to mitochondria from the adjacent liver ([232]Fig. 5e). To assess whether changes in their activity accompanied these mitochondrial changes, we again used positional isotopomer NMR tracer analysis (PINTA) as a non-invasive assessment of hepatic mitochondrial flux. SBFI-103 treatment increased TCA cycle influx, as measured through V[cs], and reduced TCA cycle efflux, measured by V[pc]/V[cs] ([233]Extended Data Fig. 6d). SBFI-103 treatment also increased the rate of total FAO flux ([234]Fig. 5f), which is consistent with the reduction of electron transport complexes (complex V, complex III, complex IV and complex II) ([235]Extended Data Fig. 6e) and the transcriptional changes observed in transformed cells from SBFI-103-treated mice by scRNA-seq analysis. We next validated the effects of FABP5 inhibition or silencing in human hepatocarcinoma cells (Huh7) ([236]Extended Data Fig. 6f). Both FABP5 pharmacological inhibition and silencing led to induction in overall ATP production in Huh7 cells ([237]Extended Data Fig. 6g,[238]h), which was concurrent with increased mitochondrial respiration and elevated FAO ([239]Fig. 5g and [240]Extended Data Fig. 6i) as measured by a Seahorse XF Analyzer. These data suggest that FABP5 pharmacological inhibition restores the dynamic metabolic alterations that we previously observed in mitochondrial metabolism during HCC development. Next, we sought to verify whether SBFI-103 treatment promotes lipid peroxidation and ferroptosis in vivo, as observed from FABP5 inhibition in the Huh7 cell line. As predicted, we found that SBFI-103-treated HCCs showed a significant increase in intratumoral MDA content ([241]Fig. 5h). Next, we measured protein and mRNA levels of specific biomarkers of ferroptosis and associated cell death mechanisms, including acyl-CoA synthetase long-chain family member 4 (ACSL4), transferrin receptor 1 (TFR1), C/EBP homologous protein (CHOP), prostaglandin-endoperoxide synthase 2 (PTGS2) and asparagine synthetase (ASNS)^[242]56. We found that SBFI-103 treatment resulted in induced protein levels of ACSL4, TFR1 and CHOP ([243]Fig. 5i) and increased mRNA levels of PTGS2 and ASNS ([244]Fig. 5j). Moreover, we also found that FABP5 inhibition increased the presence of spliced X-box binding protein-1 (sXBP1) ([245]Fig. 5k), an important mediator of the UPR response^[246]57. In HCC samples from SBFI-103-treated mice, we found iron accumulation, assessed by Perls Prussian blue staining ([247]Fig. 5l), a hallmark of ferroptosis that mediates the conversion of lipid peroxides^[248]47. Furthermore, we also observed an increased intratumoral ROS, captured by dihydroethidium (DHE) staining ([249]Fig. 5m). Our findings here demonstrate that FABP5 inhibition induces metabolic rewiring through promoting FAO and enhances lipid peroxidation-induced ferroptosis in cancer cells leading to a reduction in HCC burden. Genetic ablation of FABP5 in hepatocytes reduces tumour incidence by sensitizing transformed cells to ferroptosis Upon FABP5 inhibition, we observed a remarkable decrease in circulating AFP biomarkers and reduced tumour incidence in CD-HFD-induced HCC; however, the promiscuous nature of a small molecule approach would affect a wide range of cell types, including cancer, immune and stromal cells, which regulate the progression of HCC. Therefore, to directly address the function of FABP5 in transformed hepatocytes, we generated a hepatocyte-specific FABP5-deficient mouse model (hereupon referred to as Fabp5^HKO) by crossing Fabp5^flox/flox mice^[250]58 with mice expressing the hepatocyte-specific tamoxifen-inducible AlbCre^ERT promoter^[251]59. To verify the successful deletion of FABP5, we separated hepatocytes from NPCs after tamoxifen administration to Fabp5^HKO mice at 4 weeks old. We found that hepatocytes exhibited a significant reduction in FABP5 expression (~75%), whereas NPCs’ FABP5 expression was unchanged ([252]Extended Data Fig. 7a), thus demonstrating the specificity of the Fabp5^HKO model. We next sought to evaluate whether FABP5 ablation in hepatocytes would be protected from HCC by feeding Fabp5^HKO mice a CD-HFD for 15 months ([253]Fig. 6a). We observed no significant difference in body weight ([254]Extended Data Fig. 7b), fasting glucose ([255]Extended Data Fig. 7c) and circulating cholesterol ([256]Extended Data Fig. 7d) in Fabp5^HKO compared with control mice after 15 months on CD-HFD. Notably, while ~35% of wild-type (WT) mice exhibit high AFP, Fabp5^HKO mice are protected from elevated AFP, in line with our previous observations from SBFI-103 treatment ([257]Fig. 4b). Additionally, AFP high WT mice correlated with elevated alanine transaminase (ALT) and aspartate aminotransferase (AST) levels. In contrast, no Fabp5^HKO mice exhibited high ALT or AST ([258]Extended Data Fig. 7e,[259]f). Our results demonstrate that FABP5 ablation in hepatocytes reduces circulating HCC and liver damage biomarkers. Fig. 6 |. Genetic ablation of FABP5 reduces HCC burden by inducing lipid peroxidation and ferroptosis. Fig. 6 | [260]Open in a new tab a, Schematic outlining induction of HCC by 15 months of CD-HFD feeding in tamoxifen-treated Fabp5^HKO mice. b, Circulating AFP in CD-HFD-fed mice at 15 months after CD-HFD feeding in Fabp5^HKO mice. n = 11 for WT mice and n = 13 for Fabp5^HKO mice. c, Quantification of HCC-positive mice after 15 months of CD-HFD feeding in Fabp5^HKO mice. n = 11 for WT mice and n = 13 for Fabp5^HKO mice. d, Quantification of HCC nodules per liver in tumour-bearing mice after 15 months of CD-HFD feeding in Fabp5^HKO mice. n = 4 for WT mice and n = 2 for Fabp5^HKO mice. e, Quantification of HCC volume in tumour-bearing mice after 15 months of CD-HFD feeding in Fabp5^HKO mice. n = 4 for WT mice and n = 2 for Fabp5^HKO mice. f, UMAP representation of four distinctive cancer clusters Fabp5^HKO or WT HCC (left). Top five signature genes from each cluster are shown via heatmap (right). g, UMAP representation showing sequenced cancer cells belonging to Fabp5^HKO or WT conditions. h, Expression of AFP and FABP5 in hepatocytes and cancer cells treated with vehicle or SBFI-103 by Feature plot. i, Volcano plot of DEGs in Fabp5^HKO cancer cells compared with WT cancer cells. Select upregulated and downregulated genes are indicated. KO, knockout. j, GO analysis of significantly upregulated and downregulated genes in Fabp5^HKO cancer cells. k, MDA concentration in Fabp5^HKO and WT HCC as measured through colorimetric assay. n = 3 for WT mice and n = 2 for Fabp5^HKO mice. l, Western blot of ACSL4, TFR1, AFP, BIP and CHOP in Fabp5^HKO HCC, WT HCC and WT liver. n = 3 for WT mice and n = 2 for Fabp5^HKO mice. Statistical analysis was conducted by a Wilcox test with no adjustments made (i). For d,e,k, each dot represents an individual animal and the bar height indicates the mean and s.e.m. We next evaluated tumour burden in Fabp5^HKO mice, finding that ~45% (5 of 11) of control mice and ~15% (2 of 13) of Fabp5^HKO mice developed HCC ([261]Fig. 6c). Fabp5^HKO animals showed a trend for reduction in the number of tumour nodules ([262]Fig. 6d) and total tumour volume ([263]Fig. 6e); significance could not be calculated due to few Fabp5^HKO animals bearing HCC. To evaluate the genetic signature of Fabp5^HKO tumours, we performed scRNA-seq on single-cell suspensions from the HCC region of Fabp5^HKO mice fed CD-HFD for 15 months (WT conditions pooled from the HCC region of three animals and Fabp5^HKO conditions pooled from HCC region of two animals). A total of 8,898 cells from WT mice and 9,829 cells from Fabp5^HKO mice were identified for subsequent analysis. Upon filtering for transformed cells, a total of 2,116 cells were sequenced corresponding to four distinctive clusters (HepT_1, HepT_2, HepT_3 and HepT_4) ([264]Fig. 6f). Notably, FABP5 and AFP expression corresponds to cluster HepT_3, which are exclusively transformed cells sequenced from control mice ([265]Fig. 6g). This is further exemplified in [266]Fig. 6h, where we observe the overlap of AFP and FABP5, absent from Fabp5^HKO HCC. We next performed a differential gene expression analysis between transformed cells from Fabp5^HKO and control mice ([267]Fig. 6i and [268]Supplementary Table 11). Notably, we found HCC biomarkers such as AFP and AKR1C18 to be downregulated in Fabp5^HKO mice, further demonstrating the distinctive transcriptomics of FABP5 absent HCCs. GO analysis of DEGs identified FA metabolic pathways such as very long-chain FA metabolism, mitochondrial ETC and oxidative phosphorylation to be upregulated upon FABP5 ablation ([269]Fig. 6j), similar to our observations upon SBFI-103 treatment ([270]Fig. 5d). To evaluate lipid peroxidation and ferroptosis in Fabp5^HKO HCCs, we measured MDA content and we also performed a western blot of relevant proteins in available HCC tumours. For Fabp5^HKO mice, two distinctive tumour nodules from a single mouse (labelled with turquoise dots) were used for analysis as only two Fabp5^HKO mice with observable HCCs were available. We observed a significant increase in MDA content in Fabp5^HKO HCCs ([271]Fig. 6k). We then measured protein levels of ferroptosis and associated stress mediators and found higher expression of ACSL4, TFR1, BIP and CHOP in transformed cells with absent FABP5 ([272]Fig. 6l). Thus, genetic ablation of FABP5 protects from HCC progression by promoting lipid peroxidation and ferroptosis, consistent with our observations in CD-HFD-fed mice treated with the FABP5 inhibitor SBFI-103. FABP5 pharmacological inhibition induces a pro-inflammatory tumour microenvironment Beyond cancer cells, FABP5 is an important regulator of macrophage inflammation and polarization and is highly expressed in CD68^+ TAMs of tumours isolated from CD-HFD-fed mice ([273]Fig. 2d). In homoeostasis and infection, FABP5 controls immune tolerance by sustaining macrophage immunosuppressive signatures to dampen the host immune response^[274]44,[275]60; however, the contribution of FABP5 in regulating the TAM immunosuppressive phenotype and activation during obesity-driven HCC progression remains poorly understood. Thus, we investigated the effect of FABP5 inhibition in reshaping the TME using our scRNA-seq transcriptomic atlas. We identified different immune subsets in the TME, including mononuclear phagocytes (MPs), CD8^+ T cells, CD4^+ T cells, B cells, neutrophils and dendritic cells ([276]Fig. 7a). Given the high expression of FABP5 in MPs, we focused on this cell type and identified four unique clusters, corresponding to infiltrating MPs (Infiltrating_1 and Infiltrating_2) and resident MPs (Resident_1 and Resident_2) ([277]Fig. 7b and [278]Supplementary Table 12). FABP5 inhibition greatly enriched the Infiltrating_1 population, which showed high expression of monocyte markers such as fibronectin 1 (Fn1), chitinase-like 3 (Chil3) and S100 calcium-binding protein A10 (S100a10) ([279]Fig. 7c). Notably, we found no difference in immune cell distribution and MPs distribution in the tumour microenvironment of Fabp5^HKO animals by scRNA-seq and flow cytometry ([280]Extended Data Fig. 7g–[281]o), suggesting a distinctive effect of FABP5 inhibition independent of transformed hepatocytes. Fig. 7 |. SBFI-103 promotes the accumulation of pro-inflammatory macrophages and cytotoxic T cells. Fig. 7 | [282]Open in a new tab a, UMAP representation of six distinctive immune cell types from SBFI-103 or vehicle-treated HCC. b, UMAP representation of four distinctive mononuclear phagocyte clusters from SBFI-103 or vehicle-treated HCC (left). Top four signature genes from each cluster are shown via heatmap (right). c, UMAP representation showing sequenced mononuclear phagocyte belonging to SBFI-103 or vehicle-treated conditions (left). Quantification of percentage composition from each mononuclear phagocyte cluster is shown (right). d, RNAscope representation (left) and quantification (right) of CX3CR1 (green) and CLEC4F (red) mRNA expression in SBFI-103 or vehicle-treated HCC. n = 4 for both experimental conditions. e, Flow cytometry analysis of CD86 expression in CD11b^+F4/80^+ TAMs upon SBFI-103 treatment. n = 4 for vehicle-treated and n = 5 for SBFI-103-treated HCC. f, Flow cytometry analysis of CD80 expression in CD11b^+F4/80^+ TAMs upon SBFI-103 treatment. n = 4 for vehicle-treated and n = 5 for SBFI-103-treated HCC. g, Flow cytometry quantification of intratumoral CD8^+ T cells as a percentage of CD45^+ immune cells upon SBFI-103 treatment. n = 4 for vehicle-treated and n = 5 for SBFI-103-treated HCC. h, Flow cytometry quantification of Ki67^+ intratumoral CD8^+ T cells upon SBFI-103 treatment. n = 4 for vehicle-treated and n = 5 for SBFI-103-treated HCC. i, Heatmap of upregulated secreted factors and downstream ligands from MPs and CD8^+ T cells in SBFI-103-treated conditions identified by NicheNet analysis. Shown are (i) predicted regulatory ligands from MPs and CD8^+ T cells based on DEGs; (ii) predicted ligand activity of each secreted factor based on the expression of downstream target; and (iii) the relative expression of secreted ligands in SBFI-103-treated as compared with control conditions. Scale bars, 12 μm (d). Statistical analysis was conducted by non-parametric two-sided t-tests (d–h); P < 0.05 was considered statistically significant. For d–h, each dot represents an individual animal and bar height indicates mean and s.e.m. Data (e–h) represent two or more independent experiments. Trem2^+CD9^+ macrophages have been identified to rewire the tumour myeloid landscape to create an immunosuppressive environment^[283]61,[284]62. We next evaluated the expression of TREM2 and CD9 in MPs treated with SBFI-103. We observed a reduction in mRNA expression in both markers ([285]Extended Data Fig. 8a,[286]b), further suggesting a shift towards a pro-inflammatory microenvironment. We next sought to evaluate the spatial distribution of transcripts associated with infiltrating (CX3CR1) and resident MPs (CLEC4F) to further support our scRNA-seq analysis^[287]63. In line with our expectations, we found an increase in CX3CR1 infiltrating MP signal and a reduction in CLEC4F resident MP signal by RNAscope upon SBFI-103 treatment ([288]Fig. 7d and [289]Extended Data Fig. 8c). To further evaluate immune rewiring upon SBFI-103 treatment, we examined TME alterations through flow cytometry analysis (gating strategy shown in [290]Extended Data Fig. 8d). To this end, we first quantified the percentage of CD45^+CD11b^highF4/80^high TAMs. While the percentage of TAMs remained consistent across treatments ([291]Extended Data Fig. 8d), we found an increased percentage of CD11b-^highLy6C^highLy6G^neg infiltrating MPs ([292]Extended Data Fig. 8e). We also found increased expression of monocytic marker Ly6C in the TAM population ([293]Extended Data Fig. 8f). Moreover, we observed an accumulation of infiltrating macrophages that expressed higher surface levels of the co-stimulatory molecules, CD86 and CD80, in SBFI-103-treated mice ([294]Fig. 7e,[295]f). CD86 and CD80 induce the activation and proliferation of T cells by binding to the CD28 receptor, inducing the production of IL-2 by T cells^[296]64. Thus, we examined whether FABP5 inhibition induces T cell activation and proliferation in tumour-infiltrating T cells. Notably, we observed an increase in tumour-infiltrating CD8^+ T cells and a greater percentage of CD44^+CD62L^− effector T cells in tumours from mice treated with FABP5 inhibitor compared with untreated mice ([297]Fig. 7g and [298]Extended Data Fig. 8h). The higher T cell accumulation in tumours correlated with an increase in T cell proliferation, as measured by intracellular Ki67 staining ([299]Fig. 7h). Consistent with this finding, we found an increase in CD8^+ T cells by IHC staining upon FABP inhibition ([300]Extended Data Fig. 8i). While FABP5 inhibition did not have a meaningful effect on CD8^+ PD-1 expression, we found a trend in reduced PD-1 expression in CD4^+ T cells. ([301]Extended Data Fig. 8j). We next examined alterations in the secretome of MPs and CD8^+ T cells after SBFI-103 treatment using NicheNet to define ligand–target interactions^[302]65,[303]66. Notably, we observed the upregulation of transforming growth factor-β (Tgfb1), IL-15 (Il15) and C-C motif chemokine ligand 2 (Ccl2) in MPs from inhibitor-treated mice ([304]Fig. 7i and [305]Extended Data Fig. 8k). These secreted factors are predicted to induce cytotoxic gene networks in tumour-infiltrating CD8^+ T cells, such as granzyme B (Gzmb), B cell lymphoma 2 (Bcl2) and C–C motif chemokine ligand 3 (Ccl3). Additionally, we also observed autocrine interferon-γ (Ifnγ) signalling in CD8^+ T cells, which further amplifies T cell activation and trafficking gene networks. These findings support that FABP5 inhibition skews the TME to a pro-inflammatory phenotype, characterized by inflammatory TAMs that express high co-stimulatory molecules and increased activity and proliferation of tumour-infiltrating CD8+ T cells. We then ascertained whether the effects of FABP5 inhibition on the phenotype of tumour immune-associated cells were restricted to HCC or shared within other tumour models. To this end, we administrated SBFI-103 in a highly immunogenic colon adenocarcinoma (MC38) subcutaneous murine cancer model^[306]67. One week after tumour cell implantation, mice were treated with SBFI-103 or vehicle control daily for 7 days and then tumours were isolated and processed for flow cytometry analysis ([307]Extended Data Fig. 9a). FABP5 inhibition reduced overall tumour growth and weight ([308]Extended Data Fig. 9b,[309]c), promoting the accumulation of myeloid Ly6C^+ cells and CD45^+CD11b^highF4/80^high TAMs with higher CD86 expression ([310]Extended Data Fig. 9d,[311]e). Additionally, we observed increased tumour-infiltrating T cells ([312]Extended Data Fig. 9f) with higher levels of intracellular Ki67 ([313]Extended Data Fig. 9g). Our findings in the MC38 tumour model are consistent with observed TME reprogramming in HCC, demonstrating the significance of FABP5 in maintaining an immunosuppressive TME. FABP5 inhibition in macrophages leads to co-stimulation of CD8^+ T cells To directly examine whether FABP5 inhibition or silencing in macrophages can alter T cell proliferation and cytotoxicity, we set up a three-stage antigen-presenting co-culture experiment with bone-marrow-derived macrophages (BMDMs), transgenic ovalbumin antigen recognizing OT-I (T cells) and ovalbumin-presenting B16 cancer cells ([314]Fig. 8a). In stage 1, we examined the effect of FABP5 inhibition and silencing in non-polarized or activated, via IL-4 stimulation, BMDMs. We found that FABP5 mRNA and protein levels were induced after IL-4 stimulation ([315]Fig. 8b,[316]c). Notably, FABP5 inhibition or silencing reduced the immunosuppressive phenotype of IL-4-treated macrophages by diminishing the expression of ARG1, YM1, RETNLA1 and MRC1 ([317]Fig. 8d and [318]Extended Data Fig. 10a), as well as decreasing arginase activity ([319]Fig. 8e and [320]Extended Data Fig. 10b). Next, we used flow cytometry analysis to evaluate BMDMs and co-cultured T cell phenotypes upon FABP5 inhibition and silencing (gating strategy shown in [321]Extended Data Fig. 10c, [322]d). SBFI-103 treatment and FABP5 silencing also led to an increased expression level of CD80 and CD86 ([323]Fig. 8f,[324]g and [325]Extended Data Fig. 10e,[326]f), consistent with our finding in tumours from SBFI-103-treated mice ([327]Fig. 7e,[328]f). Fig. 8 |. FABP5 inhibition in macrophages enhances CD8^+ T cell co-stimulation to promote CD8 proliferation and cytotoxicity. Fig. 8 | [329]Open in a new tab a, Schematic outlining treatment co-culture setup of SBFI-103- or FABP5 siRNA-treated BMDMs with OT-I^+ T cells or B16-OVA cancer cells. b, RT–PCR analysis of FABP5 mRNA expression in 15 ng ml^−1 IL-4-treated and untreated BMDMs. n = 3 for both experimental conditions. Unpol, unpolarized. c, Western blot analysis (left) and quantification (right) of FABP5 in 15 ng ml^−1 IL-4-treated and untreated BMDMs. n = 3 for both experimental conditions. d, RT–PCR analysis of ARG1, MRC1, RETNLA and YM1 mRNA expression in 10 μM SBFI-103-treated IL-4 polarized BMDMs. n = 3 for both experimental conditions. e, Arginase activity in 10 μM SBFI-103-treated IL-4 polarized BMDMs. n = 3 for both experimental conditions. f, Flow cytometry analysis of CD86 MFI in 10 μM SBFI-103-treated BMDMs in non-polarized and IL-4-treated conditions. n = 3 for each experimental condition. g, Flow cytometry analysis of CD80 MFI in 10 μM SBFI-103-treated BMDMs in non-polarized and IL-4-treated conditions. n = 3 for each experimental condition. h, Flow cytometry analysis of CD69^+, CD25^+ OT-I^+CD8^+ T cells after 48 h of co-culture with 10 μM SBFI-103-treated BMDMs in non-polarized and IL-4-treated conditions. n = 3 for each experimental condition. i, Flow cytometry analysis of CTV in OT-I^+CD8^+ T cells after 48 h of co-culture with 10 μM SBFI-103-treated BMDMs in non-polarized and IL-4-treated conditions. n = 3 for each experimental condition. j, Flow cytometry analysis of IFNγ OT-I^+CD8^+ T cells after 48 h of co-culture with 10 μM SBFI-103-treated BMDMs in non-polarized and IL-4-treated conditions. n = 4 for each experimental condition. k, Flow cytometry analysis of caspase 3^highLIVE/DEAD Aqua^high B16-OVA cancer cells after 24 h of co-culture with OT-I^+CD8^+ T cells previously activated in the presence of SBFI-103- or vehicle-treated BMDMs. n = 4 for each experimental condition. Statistical analysis was carried out by non-parametric two-sided t-tests (b–k); P < 0.05 considered statistically significant. For b–k, each dot represents an individual animal and bar height indicates mean and s.e.m. Data (b–k) represent two or more independent experiments. In stage 2, we examined whether FABP5-modulated macrophages can directly enhance T cell proliferation by setting up a co-culture system with SBFI-103 or FABP5 siRNA-treated BMDMs and antigen-specific OT-I^+CD8^+ T cells. Naive (CD44^− and CD62L^+) CD8^+ T cells were isolated and pretreated with CellTrace Violet (CTV) before co-culturing with BMDMs pulsed with ovalbumin SIINFEKL peptides. After 48 h in co-culture, we examined T cell activation and found an increase in the percentage of CD69^+- and CD25^+-activated CD8^+ T cells when co-cultured with BMDMs pretreated with FABP5 inhibitor or siRNA ([330]Fig. 8h and [331]Extended Data Fig. 10g). Additionally, CTV analysis verified that CD8^+ T cells undergo accelerated cell division and proliferation when co-cultured with SBFI-103 pretreated BMDMs ([332]Fig. 8i and [333]Extended Data Fig. 10h). In stage 3, we investigated the cytotoxicity of co-cultured T cells by measuring the expression of IFNγ and induction of apoptosis in T cells when co-cultured with ovalbumin-presenting B16 melanoma cells. As expected, CD8^+ T cells co-cultured with inhibitor- or siRNA-treated macrophages produced higher levels of IFNγ ([334]Fig. 8j and [335]Extended Data Fig. 10i). When co-cultured with ovalbumin-presenting B16 cancer cells, we found that T cells previously co-cultured with FABP5 inhibited or silenced BMDMs induced enhanced cell death in B16 cancer cells, as measured through the apoptotic caspase 3^active, LIVE/DEAD^high population ([336]Fig. 8k and [337]Extended Data Fig. 10j). Our data demonstrate that inhibiting and silencing FABP5 in BMDMs can promote the proliferation and cytotoxicity of CD8^+ T cells. Discussion In western societies, increased incidences of obesity and metabolic syndrome are the most rapidly growing cause of cancer^[338]68. Of these obesity-induced cancers, HCC is preceded by NAFLD and NASH, two of the most common hepatic abnormalities^[339]2. While a substantial number of efforts have been dedicated to studying NAFLD and NASH, aiming to develop effective therapeutic interventions, there have been considerably fewer efforts in studying the NAFLD–NASH–HCC transition axis. Building upon recent scRNA-seq studies to examine hepatic changes in homoeostasis and disease^[340]19,[341]21,[342]28, we generated a single-cell transcriptomic atlas to uncover the unique metabolic and transcriptional alterations that accompany the NAFLD–NASH–HCC transition axis, using a mouse model that uniquely mimics human disease^[343]17. Our work delineated distinctive changes in metabolic and inflammatory signatures across diverse cell types, which help to identify potential therapeutic targets. In this regard, we uncovered that FABP5 was abundantly expressed in different cell types, including hepatic cancerous cells and anti-inflammatory macrophages. Here we describe a dual role for FABP5 in diet-induced HCC development, on the one hand participating in the transformation of hepatocytes and, on the other, restraining the immunosuppressive phenotype of TAMs. We also illustrate FABP5 inhibition as a potent bifunctional therapeutic approach. Recent work has shown that HFDs, especially when rich in saturated fats, can hyperactivate transcriptional programmes, such as oncogenic MYC, which regulate FABP5 transcription^[344]36. In line with this, our scRNA-seq analysis predicts c-MYC to be hyperactivated in FABP5^+ hepatic cancerous cells, suggesting that the upstream regulation of FABP5 could be associated with well-established oncogenic networks. Other transcription factors have been associated with the upregulation of FABP5 in different tumours, including FOXA1, PPARγ and NF-κB^[345]41,[346]69,[347]70. These transcription factors that regulate FA metabolism and inflammation may contribute to the specific upregulation of FABP5 observed in TAMs and hepatocyte-derived cancerous cells in HCC. FABP5 function is highly context- and cell-type-dependent. In prostate cancer, FABP5 has been implicated as a central mechanistic link between cytosolic lipid metabolism and metastatic nuclear receptor signalling, demonstrating that FABP5 expression is critical for the abilities of FA synthase (FASN) and monoacylglycerol lipase (MAGL) to promote nuclear receptor activation^[348]45. In line with this, FABP5 nuclear translocation potentiates cell survival and growth by promoting PPARβ/δ transcriptional signalling, thus engaging a positive-feedback loop to further amplify the FABP5–PPARβ/δ axis^[349]69. Building upon these studies, we found FABP5 to be an important regulator of mitochondrial biogenesis and FAO. By inhibiting FABP5, we restored mitochondrial sphericity and increased mitochondrial respiration in transformed cancer cells. Beyond liver cancer, FABP5 is upregulated in numerous cancers, particularly in breast, lung and prostate cancer^[350]71–[351]73. Additionally, the expression of FABP5 is associated with poor prognosis and high-tumour grade in triple-negative breast cancer and HCC^[352]74,[353]75. Despite these studies, there has been no association between FABP5 expression and clinical biomarkers. Our finding expands upon the prognostic function of FABP5 by demonstrating a strong correlation between FABP5 and AFP, an established cancer biomarker^[354]23. Not only did FABP5 expression co-reside on a scRNA-seq level with AFP^+ cancer cells, but it also predicts the detection of circulating AFP. Notably, the coexpression of AFP and FABP5 is not limited to HCC but could also be readily detected in the fetal liver, upon which both genes are abundantly expressed^[355]76. Of the many oncogenic pathways, activation of the Hippo pathway in HCC is an independent predictor of HCC survival and is associated with high AFP levels^[356]50. Hyperactivation of YAP1, a transcriptional regulator in the Hippo pathway, plays a key role in regulating HCC differentiation and growth^[357]51. In our study, we found that FABP5 inhibition can simultaneously reduce the total expression of YAP1 and increase YAP1 phosphorylation, promoting translocation from the nucleus to the cytoplasm. We demonstrate that FABP5 inhibition may indirectly affect the Hippo pathway to improve HCC differentiation and prognosis. Of note, a recent study has found FABP5 to be a binding partner of hypoxia-inducible factor-1α (HIF-1α) in HCC^[358]77. This FABP5–HIF-1α axis is important for regulating lipid accumulation in HCC cells, thereby enhancing cell cycle and proliferation signatures. In line with this study, we also found FABP5 to be a central regulator of lipid homoeostasis in HCC cancer cells. Our finding builds upon the multifaceted functions of FABP5 by demonstrating that FABP5 inhibition leads to lipid peroxidation and ER stress, providing a mechanism by which selective inhibition of FABP5 induces cellular ferroptosis. Ferroptosis is an intracellular form of cell death dependent upon iron accumulation and oxidative stress^[359]47. Targeting ferroptosis through inhibiting glutathione peroxidase 4 (GPX4) or suppressing upstream dihydroorotate dehydrogenase induces extensive mitochondrial lipid peroxidation and synergizes with US Food and Drug Administration-approved chemotherapy to suppress tumour growth^[360]78. We demonstrate that FABP5 protects HCC from ferroptosis in the highly oxidative tumour microenvironment and that FABP5 inhibition suppresses the HCC burden by potentiating lipid peroxidation and ER stress. Mechanistically, FABP5 inhibition can potentiate lipid peroxidation by regulating intracellular polyunsaturated FA (PUFA) to be esterified by ACSL4 and LPCAT3 into part of the plasma membrane, which reacts with ROS to form lipid peroxides and perturbs membrane integrity^[361]79. Of note, FABP5 demonstrates a strong binding affinity for PUFA^[362]80. It may also regulate the intracellular ratio of PUFA compared with other FAs, which we intend to evaluate through high-throughput lipidomic analysis. In addition, we show that FABP5 inhibition induces upregulation of lipid esterification enzymes (ACSL4 and ASNS) and enhances ROS accumulation in the TME to potentiate ferroptosis. In macrophages, FABP5 has been associated with anti-inflammatory lipid-laden macrophages in the atherosclerotic plaque^[363]43,[364]81. Functionally, FABP5 activation enhances the accumulation of lipids in monocytes, enhancing their secretion of IL-10 and promoting regulatory T cell activation^[365]82. Of note, FABP5 is also abundantly expressed in activated T cells. FABP5 is a central regulator of mitochondrial lipid trafficking and integrity in regulatory T cells, where inhibition and silencing of FABP5 lead to mitochondrial sphericity and reduction in the ETC complex^[366]55. HCC arises in chronic inflammatory conditions and many immune cells can curtail or propagate disease progression. Auto-aggressive CXCR6^+CD8^+ T cells in NASH are susceptible to metabolic stimuli to trigger MHC class-I-independent self-cytotoxicity^[367]83. Consistent with this finding, administration of anti-PD-1 immunotherapy during NASH accelerated HCC development^[368]18, highlighting the prominent role of CD8^+ T cells in pathogenesis. HCC development is also accompanied by the accumulation of IgA^+ cells, which express programmed death ligand 1 (PD-L1) and IL-10 to suppress liver cytotoxic CD8^+ T cells^[369]84. In contrast, anti-PD-1 immunotherapy in advanced HCC can significantly improve overall and progression-free survival in human patients, becoming one of the only US Food and Drug Administration-approved immunotherapies in HCC^[370]85. Building on these findings, we identified that inhibition of FABP5 leads to a pro-inflammatory TME, characterized by an increase in Ly6C^+ MPs and activated CD8^+ T cells. These findings correlate with recent observations showing enhanced FABP5-mediated FA metabolism in component 1q^+ TAMs associated with immunosuppression in the TME^[371]86. We further demonstrate that FABP5 is a critical regulator of macrophage-induced T cell activation by controlling the surface expression of co-stimulatory ligands CD80 and CD86. These ligands trigger CD28 receptor activation in T cells, thereby enhancing the proliferation and cytotoxicity of tumour-infiltrating T cells. On the mechanistic level, the induction of a macrophage pro-inflammatory signature by FABP5 inhibition could be associated with changes in macrophage mitochondrial metabolism. It is established that macrophage FAO is important for alternative (M2) activation of macrophages and suppression of this pathway through mitochondrial carnitine palmitoyl-transferase 1 (CPT1a) by etomoxir inhibits macrophage M2 polarization^[372]87. Together, our findings strongly suggest that inhibition of FABP5 promotes tumour cell death and macrophage immune rewiring to a more anti-tumoural phenotype; however, the results showing the marked protection to CD-HFD-induced HCC in mice lacking FABP5 in hepatocytes suggest a main role of FABP5 in hepatocytes during the progression of the disease. Further studies using a mouse lacking FABP5 in myeloid cells will be important to dissect the specific contribution of FABP5 in macrophages during obesity-driven HCC. Methods Mice Animals were maintained under pathogenic-free conditions and all experiments were approved by the Institutional Animal Care Use Committee (IACUC) of Yale University School of Medicine. C57BL/6J (RRID IMSR_JAX:000664) and OT-I transgenic mice (RRID IMSR_JAX:003831) were purchased from The Jackson Laboratory. FABP5 conditional knockout (Fabp5^flox/flox) mice were obtained from the Kaczocha laboratory^[373]58 and P. Chambon and D. Metzger (University of Strasbourg, IGBMC, France) provided the AlbCRE^ERT2 mice^[374]59. To induce obesity-driven HCC, 6-week-old C57BL/6J mice and Fabp5^HKO were fed with CD-HFD (Research Diets; D05010402) or control diet (Research Diets; D12450H) for up to 15 months. T cells were isolated from 6–10-week-old OT-I transgenic mice for co-culture experiments with SIINFEKL pulsed BMDMs. No statistical methods were used to predetermine sample sizes, but our sample sizes are similar to those reported in previous publications^[375]17,[376]18. Murine syngeneic tumour model A heterotopic model of MC38 (ATCC; RRID CVCL_B288) murine colon adenocarcinoma cell was used. A total of 2.5 × 10^5 cells were injected subcutaneously into the dorsal flank of 10-week-old mice. When tumours became palpable, they were monitored for growth by measuring the length and width of the tumour using a calliper and tumour volume was determined using the following formula: volume = 0.52 × (width)^2 × (length). Mice with no palpable tumour at day 10 were removed from the study. The maximal tumour size/burden approved by IACUC is 2,000 mm^3 for mice. All analyses were performed following the IACUC recommendations and the maximal tumour size/burden was not exceeded in any case. HCC burden and volume After killing, the livers of animals fed with CD-HFD were collected and images were taken with a ruler from the front and back angles. Visible HCC nodules were identified and the tumour volume was calculated using the following formula: volume = 0.52 × (width)^2 × (length). A blind analysis was performed by an independent researcher. All analyses were performed following the IACUC recommendations and the maximal tumour size/burden of 2,000 mm^3 was not exceeded in any case. Histology and IHC Following killing, livers were fixed in 10% formalin (Sigma) overnight at 4 °C. Tissues were then incubated in 70% ethanol before embedding in paraffin sections and serial liver sections were cut at 10-μm thickness. Consecutive sections were stained with H&E, Picrosirius red and Pearls Prussian blue to analyse hepatocyte morphology, fibrosis and iron content, respectively. For IHC, sections were washed with PBS, incubated with 1% hydrogen peroxidase (Sigma; H1009) in PBS for 30 min and blocked with donkey serum (Sigma; D9663) in PBS for 30 min. Liver sections were incubated overnight at 4 °C with primary FABP5 antibodies (R&D Systems; AF1476). The following day, sections were incubated for 30 min at room temperature with biotinylated rabbit anti-goat IgG secondary antibodies (Jackson; 305-065-046). Slides were washed with PBS and incubated at room temperature with streptavidin peroxidase-conjugates (Rockland; S000–03) for 30 min at 1:500 dilution in PBS. Finally, sections were incubated with 3,3′-diaminobenzidine (DAB) substrate (Rockland; DAB-10) for 3–8 min and counterstained with haematoxylin. We used the TUNEL Assay kit (QIAGEN; 74104) for apoptosis staining according to the manufacturer’s instructions. Images were taken with an EVOS XL Core microscope (Thermo Scientific; AMEX1000) or Motic EasyScan Infinity 60. Quantification of stained areas was performed with ImageJ (Fiji) software. IF Following killing, the livers were fixed in 10% formalin (Sigma) overnight at 4 °C. Tissues were incubated in 15% and 30% sucrose (Sigma; S0389) gradients overnight and embedded in an optical cutting temperature compound. Livers were cryosectioned (10 μm) in the transverse axis and sections were washed with 0.1% Triton X-100 (Sigma; T8787) in phosphate-buffered saline (PBS-T). Sections were incubated at 4 °C overnight with FABP5 (R&D systems; AF1476), AFP (Protein-tech; 14550–1-AP), CD68 (Bio-Rad; MCA1957) primary antibodies or Phalloidin-Alexa Fluor 647 membrane dye (Thermo Fisher; A22287) (1:1,000 dilution) after blocking with blocking buffer (5% donkey serum and 0.5% BSA in PBS-T) for 1 h at room temperature (RT). Following overnight incubation, sections were stained with Alexa Fluor secondary antibodies (Jackson) for 1 h at RT and immersed in a mounting medium with nuclear fluorescent dye 4′,6-diamidino-2-phenylindole dihydrochloride (DAPI) (Vector Laboratories; H-1200). Sections were captured using the SELLARIS confocal microscope (Leica). For image processing, ImageJ (Fiji) software was used. RNAscope in situ hybridization To detect single mRNA molecules, an RNAscope was performed on fresh-frozen HCC slices. Slices of 10 μm were cut from fresh-frozen tissue, dried for 10 min at RT and kept at −80 °C. In situ hybridization was performed according to the protocol of RNAscope Multiplex Fluorescent Reagent kit v2 (ACD Bio; 320293). In brief, slides were dried at RT for 5 min before incubation in cold 4% PFA for 30 min. Slides were then dehydrated in 50%, 70% and 100% ethanol for 5 min each at RT. For antigen accessibility, slides were treated with Protease IV for 20 min at RT. Sections were washed twice with PBS and C2 probes for CX3CR1 and CLEC4F were added. Subsequent amplifications were performed according to the manufacturer’s protocol. Before mounting the slides, DAPI was added to label the nuclei and images were captured immediately using the SELLARIS confocal microscope (Leica) to preserve fluorescence quality. Real-time PCR Total RNA from FABP5 siRNA or SBFI-103-treated Huh7 cells/BMDMs was extracted using RNeasy mini kits and reverse-transcribed into cDNA with iScript cDNA synthesis kit. Real-time PCR was conducted with specific oligonucleotides described in [377]Supplementary Table 13. qPCR was performed with SsoFast EvaGreen Supermix using the same thermal profile conditions for all primers set: 40 amplification cycles (95 °C for 5 s and 60 °C for 10 s). All samples were analysed in triplicate and 36b4 was used for normalization. Cell lines and primary cell culture The human Huh7 (01042712), HepG2 (HB-8065), mouse MC38 (CVCL_B288) and mouse B16.OVA (CRL-6322) cell line was purchased from Sigma and ATCC. Cells were cultured for three passages before being subjected to experimental treatment or implantation in the flank of uninfected mice. Cells were maintained in DMEM (Huh7, HepG2) or RPMI 1640 (MC38, B16.OVA) supplemented with 10% fetal bovine serum (R&D Systems), 4 mM l-glutamine (Sigma) and 1% penicillin/streptomycin (Gibco) under 5% CO[2] at 37 °C in a humidified incubator. BMDMs were isolated from the femur of 6-week-old C57BL/6J mice and maintained in RPMI 1640 supplemented with 30% conditional medium (see below), 20% fetal bovine serum, 4 mM l-glutamine and 1% penicillin/streptomycin before subjecting to experimental treatments. Conditional medium was generated through culturing L929 cell lines in RPMI 1640 supplemented with 10% fetal bovine serum, 4 mM l-glutamine and 1% penicillin/streptomycin for 10 days. Naive CD8^+ T cells were obtained from total splenocytes isolated from 6-week-old OT-I mice using the naive CD8 T cell kit (Stem Cell Technologies; 19858) according to the manufacturer’s instructions and all co-culture experiments were maintained under RPMI 1640 supplemented with 10% fetal bovine serum, 4 mM l-glutamine and 1% penicillin/streptomycin. Preparation of single-cell suspension Hepatic single-cell suspensions were prepared for submission for 10x single-cell RNA sequencing and flow cytometry. Livers were perfused first with a digestion cocktail at 37 °C containing 1.5 mg ml^−1 Collagenase I (Sigma; SCR103) and 0.5 mg ml^−1 DNase I (Roche; 4716728001) from the portal vein until tissue softens (approximately 15 min) and then perfused with DMEM containing 10% FBS. Tissues were removed and placed in cold medium before shaking to dissociate hepatic single-cell suspensions. Isolated suspensions were centrifuged at 60g at RT for 2 min to separate the PC (hepatocytes and cholangiocytes) and NPC (endothelial and immune cells) fractions. For the PC fractions, a 35% Percoll (Sigma; P1644) gradient was used to enrich live cells after spinning at 500g at 4 °C for 5 min. NPC fractions can be directly processed for flow cytometry or enriched for scRNA-seq. To enrich for live NPCs, cells were stained with LIVE/DEAD Fixable Aqua Dead Cell Stain kit (Invitrogen; [378]L34957) before sorting on the FACSAria II cell sorter. Purified PC and NPC fractions were mixed at a 1:1 ratio before submission for single-cell sequencing. Flow cytometry The antibodies used for flow cytometry analysis are described in [379]Supplementary Table 14 and were used at a 1:500 dilution in PBS with 2% FBS. For blood, plasma samples were treated with Red Blood Cell lysis buffer (Roche; 11814389001) for 5 min in RT before staining. Staining was performed in 1% FBS in PBS for 30 min on ice. Dead cells were excluded with the LIVE/DEAD Fixable Aqua Dead Cell Stain kit (Invitrogen; [380]L34957). For intracellular staining, cells were stimulated using a PMA/ionomycin cell stimulation cocktail (Invitrogen; 00-4970-93). For CTV staining, isolated CD8^+ T cells were incubated in 5 μM CTV (Invitrogen; [381]C34557) at RT for 10 min and washed three times with PBS before co-culturing with BMDMs. An LSR II flow cytometer (BD) and FlowJo were used for acquisition and data analysis. All the gating strategies for the flow cytometry analysis are shown in [382]Supplementary Figs. 1–[383]5. SBFI-103 administration in animals, cell lines and primary cells SBFI-103 (Stony Brooks University) was injected into C57BL/6 mice at 15 mg kg^−1. To prepare SBFI-103 for injection, 5 mg SBFI-103 was dissolved in 50 μl DMSO (Sigma) and mixed well with 50 μl Kolliphor EL (Sigma; 61791-12-6). Then, 900 μl warm PBS was added to create a working concentration of 5 mg ml^−1 for intraperitoneal injection. For SBFI-103 treatment in Huh7 cells and BMDMs, a working stock concentration of 50 mM SBFI-103 in DMSO was made, followed by a 1:10,000 dilution in medium (DMEM or RPMI 1640) containing 1% FBS. Seahorse analysis of mitochondrial FAO Huh7 cells were seeded at 20,000 cells per well on XF96 V3-PS cell culture microplates (Agilent; 101085–004) in complete medium (DMEM, 0.5 mM glucose, 1.0 mM glutamine, 1% FBS and 0.5 mM carnitine). The next day, the medium was supplemented with 100 μM BSA-conjugated palmitate (Sigma; P9767) and 100 μM l-carnitine (Sigma; A6706). Mitochondrial bioenergetics were assessed in the absence or presence of 20 μM etomoxir (Cayman Chemical; 828934-41-4) through measurement of COR changes in response to the following compounds: 1 μM oligomycin (Sigma; 75351), 2 μM FCCP (Sigma; C2920) and 2.5 μM rotenone (Sigma; R8875)/antimycin A (Sigma; A8674) mixture. Oxygen consumption rate (OCR) data were normalized to protein levels after lysing cells with RIPA buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1% Nonidet P-40, 0.5% sodium deoxycholate and 0.1% sodium dodecyl sulfate (SDS)) and quantified through a Pierce BCA protein assay kit (Thermo; 23227). Western blot Cells were lysed in ice-cold radioimmunoprecipitation assay buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1% Nonidet P-40, 0.5% sodium deoxycholate and 0.1% SDS) supplemented with protease inhibitor cocktail (Abcam; ab271306). Cell lysates were rotated at 4 °C for 1 h before the insoluble material was removed by centrifugation at 12,000g for 10 min. After normalizing for equal protein concentrations, cell lysates were resuspended in SDS sample buffer before separation by SDS–PAGE and transferred onto 0.45-μm nitrocellulose membranes. Blots were then probed separately with primary antibodies as described in [384]Supplementary Table 14. Following incubation with Alexa Fluor 680 or Alexa Fluor 800 secondary antibodies (Jackson), images were collected using the Odyssey Infrared imaging system (Licor). Densitometry analysis of the gels was carried out using ImageJ software. Body composition, fasting glucose, GTT and ITT Fat and lean mass were measured using non-invasive magnetic resonance technology (EchoMRI) 6 months after initiation of CD-HFD feeding. Glucose levels were determined in the blood collected from the tail vein using a Contour Ultra blood glucometer. For fasting glucose measurements and glucose tolerance tests, mice were fasted for 16 h overnight before intraperitoneal injection of glucose (1 g kg^−1) (Thermo; A16828.0C). Similarly, for the insulin tolerance tests, 6-h fasting mice were injected with insulin (2.0 U kg^−1) (Gibco; 12585014) intraperitoneally and glucose levels in the blood were analysed immediately before and 15, 30, 60 and 120 min after the glucose/insulin injection. ALT, AST and AFP measurements ALT activity (Abcam; ab105134), AST activity (Sigma; MAK055) and AFP concentration (R&D Systems; MAFP00) were determined in the serum with commercial assay kits following the manufacturers’ recommendations. Lipid peroxidation measurements MDA concentration, BODIPY 581/591 C11 intensity and CellROX flow cytometry assay were used to measure lipid peroxidation in HCC and cell lines. In brief, for the MDA measurement (Abcam; [385]Ab118970), cells or tissue samples were lysed using the MDA lysis buffer according to the manufacturer’s protocol. TBA reagent was added to lysed samples or standards following incubation at 95 °C for 60 min and absorbance was measured on a plate reader at OD 532 nm. For BODIPY 581/591 C11 (Invitrogen; D3861) and CellROX assay (Invitrogen; [386]C10444), SBFI-103 or FABP5 siRNA (Dharmacon; L-008710-00-0010) treated Huh7 cells were washed with PBS and stained with respective dyes for 30 min at 37 °C. Subsequently, cells were analysed on an LSR II flow cytometer (BD) for staining intensity and processed by FlowJo. Oil Red O staining A 35-ml Oil Red O (Sigma; O0625) stock solution (0.2% wt/vol in methanol) was mixed with 10 ml 1 M NaOH and filtered. Liver sections from OCT were briefly rinsed with 78% methanol, stained with 0.16% ORO solution for 50 min and then destained in 78% methanol for 5 min. The stained sections were captured with an EVOS XL Core microscope and quantified with ImageJ. EM and mitochondrial analysis HCC and liver nodules were fixed with 2.5% glutaraldehyde (Sigma; 340855) and 2% PFA in 0.1 M sodium cacodylate (pH 7.4) for 2 h at RT. Cells were postfixed in 1% OsO[4] in the same buffer for 1 h, then stained en bloc with 2% aqueous uranyl acetate (Fisher Scientific; 18-607-646) for 30 min, dehydrated in a graded series of ethanol to 100% and embedded in Poly/bed 812 (Polysciences; 08792–1) for 24 h. Thin sections (60 nm) were cut with a Leica ultramicrotome and poststained with uranyl acetate and lead citrate. Digital images were taken using a Morada charge-coupled device camera fitted with iTEM imaging software (Olympus). Mitochondrial analysis was performed as described previously^[387]88. A blinded investigator manually traced mitochondrial profiles using ImageJ software. The mitochondrial cross-sectional area and mitochondrial aspect ratio (major axis divided by minor axis, minimum value of 1.0) were calculated to measure mitochondrial size and shape, respectively. Probability plots were utilized to estimate changes in mitochondrial size and shape and statistical differences were tested using the Kolmogorov–Smirnov test. Metabolic flux by PINTA One week before the studies, mice underwent surgery under isoflurane anaesthesia to have a catheter placed in the right internal jugular vein, tunnelled under the skin to the back of the neck. Analgesics were provided in surgery (bupivacaine and carprofen) and for 72 h after that (carprofen). On the day of the terminal study, following a glycogen-depleting overnight (14 h) fast, mice were placed in clear plastic restrainers with their tails taped to the restrainer, catheters were exposed and rinsed with normal saline and the catheters were protected in an inaccessible space behind the animals’ backs. Mice underwent a primed (3×) continuous infusion of [1,2,3,4,5,6,6-^2H[7]] glucose (continuous infusion rate 0.5 mg kg^−1 min^−1) and [3-^13C] sodium lactate (continuous infusion rate 3.4 mg kg^−1 min^−1). Blood was collected from the tail vein at 100, 110 and 120 min of infusion. After 120 min, mice were killed with intravenous pentobarbital and livers were excised and transferred to a pre-chilled dish on dry ice. Tumours were visualized and cut apart from the non-tumor liver and both the tumour and liver were weighed. Tissues were then freeze-clamped in tongs and pre-chilled in liquid nitrogen (within 30 s of removal of the liver). [^2H[7]], [^13C[1]] and [^13C[2]] glucose enrichment was measured by gas chromatography–mass spectrometry (GC–MS) as described previously^[388]89. Whole-body glucose turnover was measured by tracer dilution: [MATH: Turnover=(tr acerenrichment< mtext>plasmaenrichment−1)×infusionrate :MATH] Other key hepatic fluxes (glucose production from pyruvate, citrate synthase flux and the contributions of glucose and FAs to total mitochondrial oxidation) were measured using enrichment in glucose, alanine (measured by GC–MS) and glutamate (measured by liquid chromatography–MS/MS), as described previously^[389]89. Because glucose production from the tumour cannot be distinguished from absolute glucose production, the tumour’s contribution to net hepatic glucose production was assumed to scale with the relative mass of the tumour compared with the non-tumour liver, and absolute fluxes were calculated accordingly. Plasma lipids and lipoprotein profile analysis Mice were fasted for 12–14 h before blood samples were collected by retro-orbital puncture. Plasma was separated by centrifugation and stored at −80 °C until analysis. Plasma total cholesterol and HDL cholesterol (HDL-C) were detected by enzymatic cholesterol detection assay (Wako; 999–02601) and TG levels were also determined by standard enzymatic assay (Wako; 10752–440). Plasma cholesterol fractions (VLDL, intermediate-density lipoprotein/LDL and HDL) were detected by fast-performance liquid chromatography gel filtration on Superose 6 HR 10/30 size-exclusion column (Pharmacia) as described previously^[390]90. Hydroxyproline measurement Approximately 0.2 g of liver tissues was collected and homogenized in 5% trichloroacetic acid (Sigma; T6399) using a cell homogenizer at 8,000g for 2 min. Cells were centrifuged at 2,500g for 20 min and the supernatant was washed twice with distilled water. Then, 6 N HCl was added at 110 °C and allowed to react for 16 h. Toluene (Sigma; 244511) was then added, and the mixture was agitated for 20 min before collecting the top organic layer and adding p-dimethylamino benzaldehyde (Sigma; D2004). Hydroxyproline was detected using a spectrometer at 560 nm. FA and glucose uptake FA and glucose uptake were evaluated in Huh7 cells by BODIPY C16 (Invitrogen; D3821) and 2-NBDG (Invitrogen; N13195) staining. For BODIPY C16 uptake, Huh7 cells were maintained in PBS with 1% FA-free BSA and 1 μM BODIPY C16 for 30 min at 37 °C. For glucose uptake, 100 μM 2-NBDG was added to the culture medium and maintained at 37 °C for 2 h. Samples were analysed on an LSR II flow cytometer and quantified through FlowJo. Measurement of liver TG and CEs Snap-frozen liver tissue was homogenized in NaCl and lipids were extracted using a solvent chloroform:methanol (2:1). TGs (Abcam; [391]Ab206386) and CEs (Wako; 999–02601) were quantified using a commercially available assay according to the manufacturer’s instructions. IL-4 polarization of BMDMs and arginase activity BMDMs were cultured for 5 days to allow for complete differentiation and treated with respective drugs or siRNA before stimulation for 24 h with 15 ng ml^−1 of IL-4 (Peprotech; 214–14). Arginase activity (Abcam; [392]Ab180877) was measured according to the manufacturer’s instructions. Antigen-presenting assay and CD8^+ T cell co-culture Differentiated BMDMs were treated with FABP5 siRNA (Dharmacon; L-043807-01-0010) or SBFI-103 and polarized with IL-4, as described above. For the antigen-presenting assay, BMDMs were collected and incubated for 1 h with 100 ng ml^−1 SIINFEKL peptide (Sigma; S7951) at 37 °C, followed by three washes with PBS and culturing in RPMI 1640 complete medium. OT-I^+CD8^+ T cells were isolated from splenocytes through a naive CD8^+ T cell kit (Stem Cell; 19858) according to the manufacturer’s protocol and stained with CTV (Invitrogen; [393]C34557) for cell division tracking. Co-culture was performed at a 1:5 BMDM T cell ratio by combining 10,000 macrophages with 50,000 T cells. Analysis of T cell activation (CD25 and CD69) and cell division (CTV) by flow cytometry were performed at 24 h and 48 h after co-culture. CD8^+ T cells cultured with 100 ng ml^−1 SIINFEKL peptide were used as a positive control. Antitumor cytotoxicity in CD8^+ T cells BMDM co-cultured CD8^+ T cells were assayed for antitumor cytotoxicity through cytokine secretion and tumour co-culture experiments. At 48 h after BMDM co-culture, GolgiPlug containing brefeldin A (BD; 555029) was added to block intracellular protein transport. Intracellular IFNγ production was analysed by flow cytometry. The B16-OVA tumour cell line (ATCC; cat. no. CRL-6322) was collected and stained with 5 μM NucView caspase 3 substrate (Biotium; 10405) for 30 min at RT. Then, 4,000 tumour cells were added to each 96-well and activated T cells were added at a ratio of 2:1 (8,000 cells) or 4:1 (16,000 cells) for co-culture. Flow cytometry analysis of caspase 3 activity and total apoptotic cells were measured at 4 and 24 h after co-culture. Droplet-based scRNA-seq library construction Live cell enriched PCs and NPCs were encapsulated into droplets and processed following the manufacturer’s specifications using 10x Genomics GemCode technology. Equal numbers of cells per sample were loaded on a 10x Genomics Chromium controller instrument to generate single-cell gel beads in emulsion (GEMs) at the Yale Center for Genome Analysis. Lysis and barcoded reverse transcription of polyadenylated mRNA from single cells were performed inside each GEM, followed by complementary DNA generation using Single-Cell 3′ Reagent kits v.2 (10x Genomics). Libraries were sequenced on an Illumina HiSeq 4000 as 2 × 100 paired-end reads. Pre-processing of scRNA-seq data scRNA data from this project were processed using CellRanger software (v.2.1.1) as previously described^[394]91. First, sample demultiplexing, aligning the read to the mouse genome (University of California Santa Cruz mm10 reference genome) with Software Tools for Academics and Researchers and unique molecular identifier (UMI) processing was performed. The raw gene expression matrix was filtered with the following criteria. Cells with over 40% mitochondrial gene expression in UMI counts were removed; cells with under 500 detected genes were removed; and cells with more than 30,000 UMI were removed. After filtering, 29,066 cells from CD-HFD-induced HCC progression and 15,999 cells from SBFI-103 treatment in HCC were identified for further analysis. Dimension reduction, unsupervised clustering and cell cluster annotation The processed gene expression matrix with all retained cells for each sample was imported to the Seurat R package (v.3.1.0) for downstream analyses^[395]92. Data were normalized using the ‘NormalizeData’ function. UMI counts for each gene were divided by the total UMI counts in each cell and multiplied by the scale factor of 1,000, followed by natural-log transformation and adding a pseudo count of 1 for each gene. Based on the normalized expression matrix, the 2,000 most-variable genes were identified using the ‘FindVariableFeatures’ function with the ‘vst’ method. Highly variable genes were used for the PCA to identify the top 50 principal components using the ‘RunPCA’ function of Seurat, which was then applied to dimension reduction using the ‘RunUMAP’ function in Seurat. Uniform Manifold Approximation and Projection (UMAP) visualization indicated that cells from different samples were well mixed into the shared space^[396]93. We built a shared nearest neighbour graph for clustering using principal components 1 to 50 and k = 25 nearest neighbours. Then, the Louvain clustering algorithm was used to group the cells into different clusters. Cell clusters were annotated based on the top differentially expressed marker genes and mapped to established cell signatures. Identification of DEGs and pathway enrichment analysis DEGs were identified in different clusters by the ‘FindAllMarkers’ function in Seurat using the ‘MAST’ test and setting min.pct = 0.25. The MAST algorithm fits a hurdle model to the expression of each gene, consisting of logistic regression for the zero values (the undetectable expression level) and linear regression for the non-zero values (the expression level)^[397]94. Volcano plots and heatmaps for the DEGs were generated by customized R code using ggplot2 (v.3.3.3, R package). GSEA was used to analyse pathways with DEGs across samples. Re-clustering of hepatocytes, macrophages and T cells Cell types of interest (hepatocytes/cancer cells, macrophages and T cells) were extracted for re-clustering analysis. The selected subpopulations were applied with the same pipeline as described above. Subclusters within each cell population were identified and annotated according to the DEGs from the ‘FindAllMarkers’ function in Seurat. Building a single-cell trajectory For cell trajectories constructed with Monocle3 (ref. [398]95) and Slingshot^[399]30, the dimensionality of the scRNA-seq data was reduced by PCA with 50 components and UMAP was applied to visualize the result of data dimensionality reduction. For Monocle3, we chose the cell derived from the control diet, with the highest hepatocyte marker ‘Alb’ expression as the ‘root’ of the trajectory. Monocle3 ordered each cell along a learned trajectory according to its transcriptional progress. For Slingshot, lineages were identified by creating tracing paths through a minimum spanning tree between cell clusters. The default parameters in Monocle3 and Slingshot were used for the analysis. Regression analysis using ‘fit_models’ (Monocle3) or ‘tradeseq’ (Slingshot) was used to identify genes that change as a function of pseudotime. GO analysis was carried out to evaluate the pathway of DEGs across pseudotime. NicheNet immune crosstalk analysis To define ligand–target interactions between immune cells after SBFI-103 treatment, NicheNet analysis was performed to infer ligand–target links based on the established gene regulatory network^[400]66. DEGs in tumour-associated lymphocytes were used to determine ligand activity scores for MPs and CD8^+ T cells, followed by z-score normalization for comparison. For circular visualization of links between ligands and target genes, the circlize R package was used^[401]96. Bulk RNA sequencing Huh7 cells treated with 5 μM SBFI-103 or FABP5 siRNA were lysed in 500 μl RLT plus buffer (QIAGEN). RNA was isolated using a RNeasy Plus micro kit (QIAGEN) and RNA libraries were constructed using TrueSeq Small RNA Library preparation (Illumina) and sequenced on the Illumina NovaSeq platform. Pre-processing of the RNA sequencing data was performed by Trimmomatic. The adaptors were cut and reads were trimmed when the quality dropped below 20. Reads with a length <35 were discarded. All samples passed quality control based on the results of FastQC. Reads were mapped to the mouse reference genome via Tophat2 and counted via HTSeqCount. For visualization of PCA, the top 25% of genes with the most variable expression were used. DESeq2 with default settings was used to identify differentially expressed up- and downregulated genes. Overlapping genes from SBFI-103 and FABP5 siRNA-treated conditions were used for downstream GO pathway analysis using the online Pantherdb (v.17.0) platform. FABP5 expression and survival in human HCC GEPIA ([402]http://gepia.cancer-pku.cn/) was used to evaluate FABP5 expression and correlation with survival across 369 HCC and 160 healthy liver samples from TCGA and GTEx. Gene expression levels were displayed with box plots, with statistical significance evaluated through the Wilcoxon test. Overall survival was assessed by a FABP5 expression cutoff at the mean (above mean indicates high expression, below mean indicates low expression) and a hazard ratio was calculated based on the Cox proportional hazards model. Statistical analysis Statistical analysis was performed with Prism 7.0 (GraphPad) by a two-tailed paired Student’s t-test, two-tailed unpaired Student’s t-test or one-way analysis of variance (ANOVA) with Newman–Keuls’s test. A two-way ANOVA was performed to compare tumour growth curves. n represents the number of mice used in the experiment, with the number of individual experiments listed in the legend. Graphs show individual samples and centre values indicate the mean. P values < 0.05 were considered significant. Data distribution was assumed to be normal, but this was not formally tested. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Extended Data Extended Data Fig. 1 |. Long-term CD-HFD feeding induces metabolic abnormalities, hepatic inflammation, fibrosis, and HCC. Extended Data Fig. 1 | [403]Open in a new tab (A) Developmental timeline of obesity-induced HCC in the CD-HFD model outlining key experiments to examine metabolic, transcriptomic, and pathogenic phenotypes. (B) Body weight in C57BL/6 mice fed with CD-HFD or control diet. N = 20 for ND-fed mice and N = 40 for CD-HFD-fed mice. (C) Body composition analysis by MRI of fat mass from 3-, 6-, and 12-months CD-HFD fed C57BL/6 mice. N = 6 for each ND-fed and CD-HFD-fed mice at 3 months, CD-HFD-fed mice at 6 months. N = 8 for ND-fed mice at 6 months, ND-fed and CD-HFD-fed mice at 12 months. (D) Representative H&E staining of 3- and 6-month CD-HFD fed mice demonstrating steatosis and inflammation. (E) Representative Oil Red O staining illustrates lipid accumulation in 3- and 6-month CD-HFD-fed mice. (F) Quantification of liver triglyceride content in 3- and 6-month CD-HFD fed mice. N = 6 for each experimental condition. (G) Quantification of liver cholesterol esters (CE) content in 3- and 6-month CD-HFD fed mice. N = 6 for each experimental condition. (H) Quantification of circulating total cholesterol in 3-, 6-, 12-, and 15-months CD-HFD fed mice. N = 8 for each experimental condition. (I) Cholesterol content of FPLC-fractionated lipoproteins from pooled plasma of 6 months of CD-HFD or control diet. Plasma samples were pooled from 4 mice for each condition. (J) Fasting glucose measured from 3-, 6-, 12-, and 15-months CD-HFD fed mice. N = 6 for experimental conditions at 3 months, N = 10 for experimental conditions at 6 months, N = 8 for experimental conditions at 12 and 15 months. (K) Glucose tolerance test performed with 6-month CD-HFD-fed mice with Area Under Curve (AUC) shown (top right). N = 8 for each experimental condition. (L) Insulin tolerance test (left) performed with 6-month CD-HFD-fed mice with Area Under Curve (AUC) shown (top right). N = 8 for each experimental condition. (M) Flow cytometry quantification of hepatic CD45^+ population in C57BL/6 mice after 6 months of CD-HFD feeding. N = 4 for ND-fed mice and N = 5 for CD-HFD-fed mice. (N) Flow cytometry quantification of hepatic CD11b^low F480^high Kupffer Cells (KCs), CD11b^high F480^low Monocyte-derived Macrophages (MoMPs) and CD44^+, CD62L^+ activated CD8^+ T cell after 6 months of CD-HFD feeding. N = 4 for ND-fed mice and N = 5 for CD-HFD-fed mice. (O) Picosirius red staining analysis (left) and quantification (right) in the liver after 6 months of CD-HFD feeding. N = 4 for each experimental condition. (P) Representative gross images of High and Low AFP liver from 15 months CD-HFD and ND-fed mice. Total of 16 ND livers, 18 CD-HFD Low AFP livers and 14 CD-HFD High AFP livers. (Q) Representative H&E image from steatotic and non-steatotic HCC sections of 15 months CD-HFD fed mice. Scale bars: 200 μm in (d-e); 20 μm at highest magnification in (q). Statistical Analysis: (b) one-way ANOVA followed by Dunnett’s post hoc test; (c, f-h, j-o) non-parametric two-sided t-tests. P-value of < 0.05 considered statistically significant. For (c, f-h, j-o) each dot represents an individual animal and bar height indicates mean and SEM. Data (c, f-q) representation of 2 or more independent experiments. Extended Data Fig. 2 |. Single-Cell RNA sequencing identifies distinctive inflammatory signatures in Mononuclear Phagocytes and T cells. Extended Data Fig. 2 | [404]Open in a new tab (A) UMAP representation of 29 discrete clusters from 15 months CD-HFD and ND-fed mice. (B) UMAP representation of G1, G2M and S cell cycle phase cells from 15 months CD-HFD and ND-fed mice. (C) Feature plot visualization of representative cell marker genes across all sequenced cells. (D) Heatmap showing top 3 differentially expressed markers across all identified cell types. (E) Composition of immune cell subtypes in total PTPRC^+ immune cells for experimental conditions after 15 months CD-HFD and ND feeding. (F) UMAP representation of 6 distinctive cell clusters corresponding to Monocyte-derived Macrophages (MoMP) or Kupffer Cells (KC) after subsetting on Mononuclear Phagocytes (MPs). (G) Composition of MoMP and KC sub-clusters for experimental conditions after 15 months of CD-HFD and ND feeding. (H) Expression of pro-inflammatory and anti-inflammatory gene signatures in MPs across experimental conditions after 15 months of CD-HFD and ND feeding. (I) Flow cytometry analysis of CD11b^+ Ly6C^+ inflammatory MPs across experimental conditions after 15 months of CD-HFD and ND feeding. N = 3 for ND, Low AFP and HCC conditions. N = 4 for High AFP condition. (J) UMAP representation of 6 distinctive cell clusters corresponding to CD8^+ T cells (CD8), CD4^+ T cells (CD4), Natural Killer cells (NK) or Natural Killer T cells (NKT) after on CD3E^+ lymphocytes. (K) Violet plot analysis of NR4A2, TOX and PDCD1 expression across experimental conditions after 15 months CD-HFD and ND feeding. N = 4 for all experimental conditions. (L) Flow cytometry quantification of CD8^+ T cells across experimental conditions after 15 months of CD-HFD and ND feeding. N = 4 for all experimental conditions. (M) Flow cytometry analysis of TOX and PD1 in CD8^+ T cells across experimental conditions after 15 months of CD-HFD and ND feeding. N = 4 for all experimental conditions. Statistical Analysis: (i, l-m) non-parametric two-sided t-tests. P-value of < 0.05 considered statistically significant. For (i, l-m) each dot represents an individual animal and the bar height indicates the mean and SEM. Data (i, l-m) representation of 2 or more independent experiments. Extended Data Fig. 3 |. Single-cell RNA sequencing analysis of hepatocyte gene signatures by slingshot pseudotime analysis and comparison analysis in DEN + CCL4 model. Extended Data Fig. 3 | [405]Open in a new tab (A) UMAP representation of 4 experimental conditions in hepatocytes after 15 months CD-HFD or ND-fed mice. (B)Heatmap showing 5 periportal and central vein gene marker expression across hepatocyte subtypes (left). Feature plot representation of select periportal and central vein gene markers in hepatocytes (right). (C) Expression of upregulated genes ENO1, FBP1, GAPDH as a function of pseudotime value. (D) Expression of downregulated genes mt-CO1, mt-CYTB, mt-NO3 as a function of pseudotime value. (E) UMAP representation of 4 distinctive experimental conditions after re-clustering by slingshot from 15 months CD-HFD or ND-fed mice. (F) UMAP representation of 5 distinctive clusters after re-clustering by slingshot from 15 months CD-HFD or ND-fed mice. Identified trajectory shown by connecting central nodules between each cluster. (G) Pseudotime value visualization by slingshot across hepatocytes and cancer cells from 15-month CD-HFD and ND-fed mice. (H) GO ontology analysis of upregulated differentially expressed genes identified as a function of slingshot pseudotime progression. (I) GO ontology analysis of downregulated differentially expressed genes identified as a function of slingshot pseudotime progression. (J) UMAP representation of combined analysis including 15 months CD-HFD or ND-fed mice and WT mice treated with DEN + CCL4 sequenced at 3 days, 10 days and 30 days post-tumour initiation. (K) Expression of AFP in hepatocytes and cancer cells in diet-induced and carcinogenic HCC models by Feature plot. (L) Expression of FABP5 in hepatocytes and cancer cells in diet-induced and carcinogenic HCC models by Feature plot. (M) Dotplot expression of FABP5 and AFP from single-cell analysis from diet-induced and carcinogenic HCC models. (N) Dotplot expression of mt-Cytb, mt-Nd4 and mt-Co1 from single-cell analysis from diet-induced and carcinogenic HCC models. Extended Data Fig. 4 |. Single-cell RNA sequencing identifies FABP5 upregulation during HCC progression. Extended Data Fig. 4 | [406]Open in a new tab (a) Feature plot representation of AFP and FABP5 expression in hepatocytes and transformed cancer cells. (b) Feature plot representation of CSF1R and FABP5 expression in Mononuclear Phagocytes. (c) Immunofluorescence (IF) imaging of FABP5 and AFP staining in the CD-HFD model. (d) IF imaging of FABP5 and CD68 staining in the CD-HFD model. (e) IF imaging of FABP5 and Phalloidin staining in the CD-HFD model. (f) Whole tumour IF imaging of FABP5 (green) in HCC and adjacent healthy liver. (g) Whole tumour H&E staining in HCC and adjacent healthy liver. (h) Whole tumour CD68 immunohistochemistry staining in HCC and adjacent healthy liver. Scale bars: 100 μm in (c); 50 μm in (d); 37 μm in (e) and 1 mm in (f), (g) and (h). Data (c-e, f-h) representation of 2 or more independent experiments. Extended Data Fig. 5 |. FABP5 inhibition and silencing leads to lipid peroxidation, ER stress and ferroptosis. Extended Data Fig. 5 | [407]Open in a new tab (A) RT-PCR analysis of FABP5 mRNA expression in Huh7 cell lines treated with SBFI-103 for 48 hours. N = 3 for both experimental conditions. (B) Principle Complement Analysis (PCA) of SBFI-103 treated Huh7 cells for RNA sequencing analysis. N = 3 for both experimental conditions. (C) Volcano plot of differentially expressed genes in SBFI-103 treated Huh7 cell lines. Select upregulated and downregulated genes are indicated. (D) Top up- and downregulated GSEA pathways from differentially expressed genes in SBFI-103 treated Huh7 cells. (E) RT-PCR analysis of FABP5 mRNA expression in Huh7 cell lines treated with FABP5 siRNA for 48 hours. N = 3 for both experimental conditions. (F) PCA of FABP5 siRNA-treated Huh7 cells for RNA sequencing analysis. N = 3 for both experimental conditions. (G) Volcano plot of differentially expressed genes in FABP5 siRNA-treated Huh7 cell lines. Select upregulated and downregulated genes are indicated. (H) Top up- and downregulated GSEA pathways from differentially expressed genes in FABP5 siRNA-treated Huh7 cells. (I) Flow cytometry analysis of BODIPY C16 MFI in 5uM SBFI-103 and FABP siRNA-treated Huh7 cells. N = 3 for each experimental condition. (J) Flow cytometry analysis of 2-NBDG MFI in 5uM SBFI-103 and FABP5 siRNA-treated Huh7 cells. N = 3 for each experimental condition. (K) MDA concentration in FABP5 siRNA-treated Huh7 cells as measured through colorimetric assay. N = 3 for both experimental conditions. (L) Flow cytometry analysis of BODIPY 581/591 C11 MFI in FABP5 siRNA-treated Huh7 cells. N = 3 for both experimental conditions. (M) Flow cytometry analysis of CellRox MFI in FABP5 siRNA-treated Huh7 cells. N = 3 for both experimental conditions. (N) Western blot of PERK, ATF4, IRE1A and BIP in 5uM FABP5 and control siRNA-treated Huh7 cells. N = 3 for both experimental conditions. Statistical Analysis: (a, e, i-m) non-parametric two-sided t-tests. P-value of < 0.05 considered statistically significant. For (a, e, i-m) each dot represents an individual animal and bar height indicates mean and SEM. Data (i-m) representation of 2 or more independent experiments. Extended Data Fig. 6 |. FABP5 inhibition and silencing promotes mitochondrial respiration and oxidation. Extended Data Fig. 6 | [408]Open in a new tab (A) Violin plot of FABP5, FABP1, and FABP2 expression in SBFI-103 and vehicle-treated cancer cells. (B) Downregulated GSEA pathways from differentially expressed genes in SBFI-103 treated cancer cells. (C) EM quantification of mitochondrial size from vehicle-treated liver, vehicle-treated HCC, and SBFI-103-treated HCC. N = 3 for each experimental condition. (D) Vcs flux and Vpc/Vcs ratio analysed by PINTA and ex vivo NMR in the vehicle-treated liver, vehicle-treated HCC and SBFI-103-treated HCC. N = 3 for each experimental condition. (E) Western blot analysis of electron transport complexes in the vehicle-treated liver, vehicle-treated HCC and SBFI-103-treated HCC. N = 3 for each experimental condition. (F) Schematic outlining FABP5 inhibition and silencing treatment conditions on Huh7 cell lines before metabolic analysis. (G) ATP production in Huh7 cells treated with SBFI-103 for 5 μM 24 hours. N = 3 for each experimental condition. (H) ATP production in Huh7 cells treated with FABP5 siRNA for 24 hours. N = 3 for each experimental condition. (a) Oxygen consumption rate (OCR) was measured after 24 hours of treatment with FABP5 siRNA in Huh7 cells in and without the presence of 20 μm Etoximir (Eto) by Seahorse extracellular flux analyser. Basal and spare respiratory capacity were quantified after adding 1 μm Oligomycin, 2 μm FCCP and 2.5 μm Rotenone + Antimycin. N = 6 for each experimental condition. Statistical Analysis: (c-d, g-i) non-parametric two-sided t-tests. P-value of < 0.05 considered statistically significant. For (d, g-i) each dot represents an individual animal and bar height indicates mean and SEM. Data (e-i) representation of 2 or more independent experiments. Extended Data Fig. 7 |. Genetic ablation of FABP5 does not affect body weight, circulating glucose and cholesterol, and hepatic inflammation after 15 months of CD-HFD feeding. Extended Data Fig. 7 | [409]Open in a new tab (a) RT-PCR analysis of FABP5 mRNA expression in non-parenchymal cells (NPCs) and hepatocytes after tamoxifen administration in Fabp5^HKO mice. N = 3 for both experimental conditions. (b) Body weight before and after 15 months of CD-HFD feeding in Fabp5^HKO mice. N = 8 for WT mice before diet, N = 5 for Fabp5^HKO mice before diet, N = 12 for WT mice after diet and N = 13 for Fabp5^HKO mice after diet. (c) Fasting glucose in Fabp5^HKO mice after 15 months of CD-HFD feeding. N = 10 for WT mice and N = 13 for Fabp5^HKO mice. (d) Total cholesterol in Fabp5^HKO mice after 15 months of CD-HFD feeding. N = 10 for WT mice and N = 13 for Fabp5^HKO mice. (e) ALT activity in Fabp5^HKO mice after 15 months of CD-HFD feeding. N = 9 for WT mice and N = 13 for Fabp5^HKO mice. (f) AST activity in Fabp5^HKO mice after 15 months of CD-HFD feeding. N = 9 for WT mice and N = 13 for Fabp5^HKO mice. (g) UMAP representation of 7 distinctive immune cell types from Fabp5^HKO or WT HCC. (h) UMAP representation showing immune cells belonging to Fabp5^HKO or WT HCC (left). Quantification of percentage composition from each cell type is shown (right). (i) UMAP representation of 4 distinctive mononuclear phagocyte clusters from Fabp5^HKO or WT HCC (left). Top 5 signature genes from each cluster are shown via heatmap (right). (j) Expression of infiltrating (CHIL3, CCR2) and resident markers (CLEC4F, CD5L) in Fabp5^HKO and WT mononuclear phagocytes. (k) UMAP representation showing mononuclear phagocytes belonging to Fabp5^HKO or WT HCC (left). Quantification of percentage composition from each mononuclear phagocyte cluster is shown (right). (l) Flow cytometry quantification of CD11b^+ Ly6G^− Ly6C^+ Infil MPs in Fabp5^HKO HCCs. N = 3 for WT mice and N = 2 for Fabp5^HKO mice. (m) Flow cytometry analysis of CD86 expression in CD11b^+F4/80^+ TAMs in Fabp5^HKO HCCs. N = 3 for WT mice and N = 2 for Fabp5^HKO mice. (n) Flow cytometry analysis of CD80 expression in CD11b^+F4/80^+ TAMs in Fabp5^HKO HCCs. N = 3 for WT mice and N = 2 for Fabp5^HKO mice. (o) Flow cytometry analysis of CD206 expression in CD11b^+F4/80^+ TAMs in Fabp5^HKO HCCs. N = 3 for WT mice and N = 2 for Fabp5^HKO mice. Statistical Analysis: (a-d) non-parametric two-sided t-tests. P-value of < 0.05 considered statistically significant. For (a-d) each dot represents an individual animal and the bar height indicates the mean and SEM. Extended Data Fig. 8 |. FABP5 inhibition promotes macrophage costimulatory receptor expression and T cell proliferation in HCC and subcutaneous tumours. Extended Data Fig. 8 | [410]Open in a new tab (A) Expression of CD9 and TREM2 by Feature plot in SBFI-103 and vehicle-treated HCC. (B) Expression of CD9 and TREM2 by Vioin plot in SBFI-103 and vehicle-treated HCC. (C) Zoomed out RNAscope representation of CX3CR1 (green) and CLEC4F (red) mRNA expression in SBFI-103 or vehicle-treated HCC. (D) Flow cytometry quantification of CD11b^+ F4/80^+ TAMs in SBFI-103 treated HCC. N = 4 for vehicle-treated and N = 5 for SBFI-103-treated HCC. (E) Flow cytometry quantification of CD11b^+ Ly6G^− Ly6C^+ Infiltrating MPs in SBFI-103 treated HCC. N = 4 for vehicle-treated and N = 5 for SBFI-103-treated HCC. (F) Flow cytometry analysis of Ly6C expression in CD11b^+F4/80^+ TAMs upon SBFI-103 treatment. N = 4 for vehicle-treated and N = 5 for SBFI-103-treated HCC. (G) Flow cytometry analysis of CD206 expression in CD11b^+F4/80^+ TAMs upon SBFI-103 treatment. N = 4 for vehicle-treated and N = 5 for SBFI-103-treated HCC. (H) Flow cytometry quantification of CD44^+, CD62L^− intratumoral CD8^+ T cells upon SBFI-103 treatment. N = 4 for vehicle-treated and N = 5 for SBFI-103-treated HCC. (I) CD8 immunohistochemistry analysis (left) and quantification (right) in HCC after SBFI-103 treatment. N = 4 for each experimental condition. (J) Flow cytometry quantification of PD1^+ intratumoral CD8^+ (left) and CD4^+ (right) T cells upon SBFI-103 treatment. N = 4 for vehicle-treated and N = 5 for SBFI-103-treated HCC. (K) Circos plot of upregulated secreted factors and downstream ligands from immune cells in SBFI-103 treated conditions identified by NicheNet analysis. Secreted factors are labelled in purple, blue or yellow while downstream signalling ligands are labelled in red. Scale bars: 60 μm in (c). Statistical Analysis: (d-i) non-parametric two-sided t-tests. P-value of < 0.05 considered statistically significant. For (d-i) each dot represents an individual animal and bar height indicates mean and SEM. Data (c-i) representation of 2 or more independent experiments. Extended Data Fig. 9 |. FABP5 inhibition promotes macrophage costimulatory receptor expression and T cell proliferation in the subcutaneous MC38 tumour model. Extended Data Fig. 9 | [411]Open in a new tab (a) Schematic outlining induction of subcutaneous tumour by injection of MC38 syngeneic tumour cells in the flank and daily treatment with SBFI-103. (b) Quantification of MC38 tumour volume after one week of SBFI-103 treatment. N = 6 for both experimental conditions. (c) Quantification of MC38 tumour weight after one week of SBFI-103 treatment. N = 6 for both experimental conditions. (d) Flow cytometry quantification of CD11b^+ Ly6C^+ immune cells in SBFI-103 treated MC38 tumours. N = 6 for both experimental conditions. (e) Flow cytometry quantification of CD86 expression in TAMs from SBFI-103 treated MC38 tumours. N = 6 for both experimental conditions. (f) Flow cytometry quantification of CD8^+ T cells in SBFI-103 treated MC38 tumours. N = 6 for both experimental conditions. (g) Flow cytometry quantification of Ki67 MFI in CD8^+ T cells from SBFI-103 treated MC38 tumours. N = 6 for both experimental conditions. Statistical Analysis: (b) one-way ANOVA followed by Dunnett’s post hoc test; (c-g) non-parametric two-sided t-tests. P-value of < 0.05 considered statistically significant. For (c-g) each dot represents an individual animal and bar height indicates mean and SEM. Data (b-g) representation of 2 or more independent experiments. Extended Data Fig. 10 |. FABP5 silencing in macrophages enhances CD8^+ T cell co-stimulation to promote CD8 proliferation and cytotoxicity. Extended Data Fig. 10 | [412]Open in a new tab (A) RT-PCR analysis of ARG1, MRC1, RETNLA, and YM1 mRNA expression in FABP5 siRNA-treated IL-4 polarized BMDMs. N = 3 for both experimental conditions. (B) Arginase activity in FABP5 siRNA-treated IL-4 polarized BMDMs. N = 3 for both experimental conditions. (C) Flow cytometry gating strategy for the identification of monocultured BMDMs before subsequent analysis. (D) Flow cytometry gating strategy for identifying co-cultured OT-I^+ CD8^+ T cells before subsequent analysis. (E) Flow cytometry analysis of CD86 MFI in FABP5 siRNA-treated BMDMs in nonpolarized and IL-4-treated. N = 3 for each experimental condition. (F) Flow cytometry analysis of CD80 MFI in FABP5 siRNA-treated BMDMs in nonpolarized and IL-4-treated conditions. N = 3 for each experimental condition. (G) Flow cytometry analysis of CD69^+, CD25^+ OT-I^+ CD8^+ T cells after 48 hours of co-culture with FABP5 siRNA-treated BMDMs in nonpolarized and IL-4-treated conditions. N = 3 for each experimental condition. (H) Flow cytometry analysis of CellTrace Violet (CTV) in OT-I^+ CD8^+ T cells after 48 hours of co-culture with FABP5 siRNA-treated BMDMs in nonpolarized and IL-4-treated conditions. N = 3 for each experimental condition. (I) Flow cytometry analysis of interferon-gamma (IFNγ) OT-I^+ CD8^+ T cells after 48 hours of co-culture with FABP5 siRNA-treated BMDMs in nonpolarized and IL-4-treated conditions. N = 3 for each experimental condition. (J) Flow cytometry analysis of Caspase 3^high, Live/Dead Aqua^high B16-OVA Cancer Cells after 24 hours of co-culture with OT-I^+ CD8^+ T cells previously activated in the presence of FABP5 or control siRNA-treated BMDMs. N = 3 for each experimental condition. Statistical Analysis: (a-b, e-j) non-parametric two-sided t-tests. P-value of < 0.05 considered statistically significant. For (a-b, e-j) each dot represents an individual animal and bar height indicates mean and SEM. Data (a-b, e-j) representation of 2 or more independent experiments. Supplementary Material Supplementary Information (Figures) [413]NIHMS2101773-supplement-Supplementary_Information__Figures_.pdf^ (2.7MB, pdf) Supplementary Tables [414]NIHMS2101773-supplement-Supplementary_Tables.xlsx^ (2.2MB, xlsx) Supplementary information The online version contains supplementary material available at [415]https://doi.org/10.1038/s42255-024-01019-6. Acknowledgements