Osteopontin Expression Identifies a Subset of Recruited Macrophages Distinct from Kupffer Cells in the Fatty Liver

1Due to the increasing prevalence of obesity and associated insulin resistance, non-alcoholic fatty liver disease (NAFLD), recently renamed metabolic associated fatty liver disease (MAFLD), has become a global problem. MAFLD consists of a spectrum of disease states ranging from simple steatosis to the more end-stage of the disease termed non-alcoholic steatohepatitis (NASH), encompassing fibrosis, cirrhosis and hepatocellular carcinoma (HCC). Not all patients progress from the steatosis phase to NASH, and the reasons underlying the progression remain unclear. Additionally, as there is currently no treatment for NASH, liver transplantation is often the only option. Thus, further research is required to understand MAFLD pathogenesis and design new treatment strategies.

2It is commonly accepted that multiple hits coming from the gut and adipose tissue are key regulators of disease progression. Hepatic macrophages (macs) have been implicated in this process, being activated to become pro-inflammatory by the excess lipids and damage in the fatty liver and by signals originating from the intestine. However, it is not clear which macs are involved. Kupffer cells (KCs) are the main mac population in the healthy liver, where they reside with at least part of their body in the liver sinusoids, interacting with liver sinusoidal endothelial cells (LSECs), hepatic stellate cells (HSCs), and hepatocytes. Although there has been considerable interest in the role of KCs in MAFLD, many existing studies have relied on the use of non-specific markers to identify “KCs,” together with non-specific methods like clodronate-liposome-mediated depletion to study theirfunc- tions. It has recently become clear that these approaches cannot distinguish between resident KCs (ResKCs) and other macs, recruited to the liver in response to inflammation. As a result, these poorly defined “KCs” have been attributed many, often contradictory, roles in MAFLD, and the relative contributions of bona fide KCs and recruited macs remain largely unknown. By developing tools to identify and manipulate KCs based on their expression of the C-type lectin, CLEC4F, we have shown that ResKCs are replaced by monocyte-derived KCs (moKCs) upon depletion with diphtheria toxin and that with time, these moKCs then become resident. We have previously reported a similar replacement of the ResKC pool by moKCs in mice fed a methionine- and choline-deficient diet (MCD), a protocol that leads to a NASH- like disorder. However, whether this holds true in a more clinically relevant model of MAFLD remains to be seen.

3Here, we have employed cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq) to investigate the transcriptomes and surface epitopes of CD45+ cells in the livers of mice with MAFLD, induced by feeding a Western diet (WD) for 12, 24, or 36 weeks. In combination with flow cytometry, fate mapping, and confocal microscopy, this revealed that ResKCs are not pro-inflammatory in MAFLD. Rather, ResKCs are gradually lost as the disease progresses, being replaced by moKCs. Monocytes recruited to the liver also differentiated into a distinct subset of osteopontin-expressing CLEC4F- macs with a transcriptome similar to that of lipid-associated macs (LAMs) in adipose tissue and scar-associated macs in the fibrotic human liver and hence were termed hepatic LAMs. Notably, hepatic LAMs were differentially activated as compared with ResKCs and moKCs and had distinct abilities to metabolize lipids. Together, our data reveal considerable heterogeneity within the hepatic macrophage pool in MAFLD and suggest a need for more specific mac targeting strategies.

4To assess the roles of hepatic macs in MAFLD, mice were fed a Western diet (WD) consisting of excess fat and cholesterol and supplemented with sucrose and fructose in the drinking water, a protocol generating all stages of human MAFLD and NASH. Control mice were fed a standard diet (SD). WD-fed mice gained a significant amount of weight compared with SD-fed mice (Figure S1A) and had an increased liver weight and liver to body weight ratio (Figure S1B). From 12 weeks, WD- fed mice also showed progressively elevated serum AST and ALT, cholesterol, triglyceride and fasting insulin concentrations as well as glucose intolerance (Figures S1C–S1F). Histological analysis confirmed that WD-fed mice exhibited all features of human MAFLD, with NASH developing at later time points (Figures S1G and S1H), while some of these mice (1/82 at 12 weeks, 2/77 at 24 weeks, and 5/34 at 36 weeks) also developed HCC.

5To investigate mac heterogeneity in MAFLD, we employed CITE-Seq. To avoid bias, total live CD45+ cells from the liver were isolated from a total of 6 mice (one from each of the SD- and WD-fed groups at 12, 24, or 36 weeks), stained with a panel of 112 antibodies including isotype controls, and loaded onto the 10X Genomics Chromium platform. After sequencing, aggregation of the samples, quality control, removal of contaminating CD45- cells, and exclusion of cells resembling doublets, a total of 56,407 cells remained (6,116 cells from 12-week SD, 10,465 cells from 12-week WD, 5,357 from 24-week SD, 10,788 from 24-week WD, 11,018 cells from 36-week SD, and 12,663 from 36-week WD). 25 clusters could be identified by generating a UMAP from the transcriptome data using 20 principal components (Figure 1A). 16 discrete cell types could be identified based on the differentially expressed genes (DEGs) (Figure S2A, Table S1, www.livercellatlas.org), combined with protein expression profiles (Figures 1B, S2B, and S2C) and their expression of a set of standard cell identity genes (Figure 1C), including B cells, T cells, NK cells, patrolling monocytes, classical monocytes, KCs (Clec4f+), other macrophages (Clec4f-), pDCs, cDC1s, and cDC2s (Figure 1D). Three additional clusters (13, 19, and 24) were designated as neutrophils based on their protein expression of Ly6G (Figure 1B). However, only one of these (cluster 24) expressed mRNA for Ly6g (Figure 1C), likely reflecting the technical difficulties involved in isolating good quality mRNA from neutrophils.

6Examination of the distribution of cells originating from the mice fed the different diets at each time point suggested that many cells, including macs, B cells, and cDC1s, were altered in MAFLD (Figure 1E). To investigate this further, we assessed how the transcriptome of each population was affected. This effort found multiple cell types with a set of conserved DEGs (Figures 1F–1H and S2D; Table S1), including increased Ccl5 expression, which has previously been associated with steatosis and fibrosis and decreased expression of Ccl4 and the TIS11 family members, Zfp36, Zfp36l1, and Zfp36l2, thought to function in post-transcriptional gene degradation (Figures 1F-1H and S2D). A number of DEGs were also cell-type specific (Figures 1F-1H and S2D; Table S1). CITE-Seq analysis found 5 surface proteins to be differentially expressed by one or more cell types over time in MAFLD, including increased expression of CD8ß byTcellsand reduced TIM4on KCs (Figures 1l-1K). The increase in CD8+ T cells is in agreement with recent findings in mice and humans, highlighting the robustness of our approach. Based on our previous work, the decrease in TIM4 expression in KCs suggests that ResKCs may be replaced by newly recruited macs.

7We next used flow cytometry to explore the macs present in the liver at the different time points of MAFLD (gating strategy in Figure S3A). This found an increase in total Ly6Chi monocytes from 24 weeks on the WD compared with SD-fed mice (Figures 2A and 2B). Although only a slight increase in the number of macs defined as CD64+F4/80+ was observed at 24 weeks (Figures 2A and 2B), the numbers and proportions of CLEC4F+TIM4- KCs and CLEC4F- macs increased from 24 weeks on the WD. There was a concomitant reduction in the proportion of CLEC4F+TIM4+ ResKCs after feeding the WD for 24 weeks, and the number of these cells was reduced at 36 weeks (Figures 2A, 2C, and S3B). Notably, F4/80 and CD11b expression could not discriminate between the different macs (Figure S3C). In the few mice with HCC, CLEC4F+TIM4+ KCs were almost absent (Figures S3D and S3E). Increases in neutrophils, eosinophils, cDC1s, and cDC2s from 24 weeks on the WD (Figures S3F- S3H) were also observed, consistent with the histological evidence of inflammation (Figure S1G). While the total numbers of cDC1s and cDC2s were increased in WD-fed mice (Figure S3F), there was a reduction in the % of cDC1s and a reciprocal increase in the % of cDC2s (Figure S3G), consistent with a recent report. Moreover, as recently reported, hepatic cDC2s were heterogeneous in terms of Mgl1, CCR2, and Tbet expression (Figure S3H).

8To test the hypothesis that CLEC4F+TIM4- KCs and CLEC4F- macs were newly recruited from the bone marrow (BM) in MAFLD, we generated shielded chimeras in which CD45.2 mice were irradiated with their livers protected before being reconstituted with congenic donor BM and put on the SD or WD (Figure 2D). Chimerism of the hepatic monocyte and mac populations was then compared with that of blood monocytes, which derive exclusively from the BM. 18 weeks after reconstitution, there was no chimerism of CLEC4F+TIM4+ ResKCs in SD- or WD-fed mice, whereas CLEC4F+TIM4- KCs and CLEC4F- macs were chimeric, demonstrating their BM origin (Figure 2E). As TIM4 has been described to be expressed with time, to determine if the lack of chimerism in WD-fed ResKCs was because the CLEC4F+TIM4- moKCs had not been in the tissue long enough to acquire TIM4 expression, we also analyzed chimerism after 24 weeks. A small proportion of CLEC4F+TIM4+ ResKCs were chimeric at this time, suggesting that while some moKCs may give rise to these cells, this is likely an inefficient process (Figure 2F). Notably, local proliferation of KCs was not found to contribute to the maintenance of their numbers in MAFLD, as the number of Ki67+ CLEC4F+ KCs was identical in SD- and WD-fed mice (Figures 2G and 2H). This was not due to an intrinsic inability of the ResKCs to proliferate in the diseased liver, as administration of CSF1-Fc to SD- or WD-fed mice (24 weeks) induced proliferation of TIM4+ ResKCs irrespective of the diet (Figure 2I). To determine if loss of signals originating from LSECs or HSCs recently described to drive KC differentiation and maintenance were responsible for the reduced ResKC population, we examined these by qPCR (sorted as described in ) after feeding the diets for 36 weeks. However, none of the LSEC- or HSC-derived signals were found to be reduced (Figures 2J and 2K); thus, loss of these signals was not driving loss of ResKCs.

9To further investigate the response of the ResKC niche cells to MAFLD, we next performed scRNA-seq analysis of live CD45- cells from mice fed the SD or WD for 24 and 36 weeks (4 samples in total). After sequencing, aggregation of the samples, quality control, removal of ambient mRNA signals, alignment with the Harmony Algorithm, removal of contaminating CD45+ cells, and exclusion of cells resembling doublets, a total of 33,241 cells remained (8,747 from 24-week SD, 11,759 from 24-week WD, 7,247 cells from 36-week SD, and 5,488 from 36-week WD). 15 clusters could be identified by generating a UMAP using 20 principal components (Figure 3A, www.livercellatlas.org). 5 discrete cell types, hepatocytes, endothelial cells, HSCs, cholangiocytes, and a small population of cells resembling hepatic stem or progenitor cells (HpSCs), could be identified by analysis of DEGs (Figure 3B; Table S2). Analysis of the origins of the cells found that some clusters were predominantly composed of cells originating from the WD-fed mice, including clusters 10 and 12 (Figure 3C). Closer inspection found that these clusters were proliferating endothelial cells and hepatocytes respectively (Figure 3D). qPCR for Mki67 expression on purified cells from mice fed the diets for 36 weeks confirmed this increase in proliferation in LSECs and also identified increased proliferation in HSCs and Cholangiocytes (Figure 2E). A trend toward increased proliferation was also observed in the hepatocytes. Confocal microscopy for Ki-67 in mice fed either diet for 36 weeks further confirmed an increase in proliferation in MAFLD with Ki-67+ cells (Figure S4A).

10In addition to increased proliferation, we also investigated other changes in the transcriptomes of the niche cells. To account for the heterogeneity between clusters of the same cell type (primarily due to zonation in hepatocytes and zonation and distinct endothelial cell subsets in endothelial cells) (Figures S4B–S4D), we examined DEGs between the different clusters from mice fed the SD or WD for 24 or 36 weeks. This identified a number of conserved DEGs between the clusters (Figures 3F and S4E; Table S3). Some of these were found after feeding the WD for 24 and 36 weeks, while some were specific to the mice fed the WD for 36 weeks (Figures 3F and S4E). Any DEGs moving in different directions at 24 and 36 weeks were excluded. Two conserved downregulated genes in the niche cells (Fos & Zfp36) were also observed in the CD45+ cells (Figures 1F–1H and S2D); however, the relevance of this remains to be examined. Some of the DEGs were confirmed using qPCR (Figure 3G); this included Cd36 and Fabp4, indictive of the increased lipid load. Conversely, we were not able to validate some of the DEGs shown in the scRNA-seq analysis, such as Hspa5 (Figure 3G). Moreover, confocal microscopy for HSPA5 also did not reveal any differences in protein expression (Figure S4G); however, as this primarily stained the hepatocytes, this analysis demonstrated the increased size of hepatocytes in WD- compared with SD-fed mice (Figure S4G). These results highlight that while the expression of signals important for ResKC development and maintenance are not lost in MAFLD, the niche cells have an altered transcriptome and also undergo increased proliferation in MAFLD. Additionally, hepatocytes increase in size which could alter the distribution of signals to ResKCs. These alterations to the niche could therefore play a role in the loss of ResKCs in MAFLD.

11Having identified changes in the ResKC niche in MAFLD, we next set out to explore whether resident and recruited macs occupied the same or distinct niches in MAFLD. We therefore localized CLEC4F+TIM4+ ResKCs, CLEC4F+TIM4- moKCs, and CLEC4F- macs using confocal microscopy after feeding the SD or WD for 12, 24, and 36 weeks. This revealed that all MAFLD macs, irrespective of CLEC4F and TIM4 expression, were located in the zones between the portal and central veins, with at least a part of their body in contact with LSECs (CD31+) and hepatic stellate cells (Desmin+) (Figures 4A–4C). This location was identical to that observed for ResKCs in healthy liver. Notably, there were also some CLEC4F- macs found in close proximity to the large vessels (central & portal veins), but these were also found in the SD-fed mice (Figure 4B). Automated identification of the macs based on expression of F4/80 followed by classification into TIM4+ KCs, TIM4- KCs, or CLEC4F- macs identified similar proportions of the different subsets as retrieved by flow cytometry (Figures 4B, 4C, and S3B). However, this identified regions where CLEC4F- macs were the main mac subset present (Figures 4A and 4B). These zones were characterized by increased Desmin expression (Figures 3D and 3E), a characteristic of fibrotic stellate cells and myofibroblasts. Thus, these findings suggest that recruited macs populate the ResKC niche in MAFLD, with CLEC4F- macs being predominantly found in fibrotic zones.

12We next investigated the similarities and differences between CLEC4F- macs and KCs. To this end, we isolated cells of the monocyte and mac lineages from the CITE-seq data (18,241 cells) and re-clustered them (Figures S5A and 4F, www.livercellatlas.org). The sub-clusters revealed by this additional analysis varied between diets and time points (Figures 4G and S5B). Analysis of the DEGs and surface proteins expressed per cluster allowed them to be identified more precisely (Figures 4H, 4I and S5C–S5E). Clusters 0,1,3,5,14, and 17 were monocyte-like cells (Figures 4F, S5D, and S5E; Table S3), and of these, cluster 3 was the main population in the SD-fed mice, whereas cluster 17 was a minor population, present in all conditions but increased in WD-fed mice (Figures 4G and S5B). Cells in cluster 17 were enriched for IFN-induced genes, including Ifit1, Ifit3, and Isg15 (Figure S5D; Table S3), similar to a subset of monocytes recently reported. Cluster 11 was identified as patrolling Ly6Clo monocytes, while clusters 4, 7, 10, and 12 had features of monocytes and macs, suggesting that they were transitioning monocytes differentiating into macs (Figures 4F, 4I, S5D, and S5E; Table S3). Clusters 6, 8, 9, 15, 22, and 23 were identified as ResKCs on the basis of their expression of Clec4f and ResKC genes, including Timd4, C2, Xlr, Marco & Cd163 (Figures 4H, S5D, and S5E; Table S3). Of these, cluster 6, enriched for genes associated with inflammation, was restricted to the mouse fed the SD for 36 weeks (Figures 4G, S5B, and S5F; Table S4). However, no significant differences in the expression of these genes were observed by qPCR (Figure S5G).

13Clusters 2, 13, 16, 18, 19, 20, and 21 were also identified as macs. Of these, clusters 19 and 21 were proliferating cells containing Clec4f KCs and Clec4f macs (Figures 4H, S5D, and S5E).

14Cells in cluster 13 were identified as moKCs, expressing Clec4f and many other KC signature genes, but lacking expression of ResKC genes such as Timd4 (Figures 4H, S5D, and S5E). Based on a list of capsule mac signature genes (Table S4) generated by comparing the gene expression profile of liver capsule macs with those of ResKCs and Ly6- Chi monocytes and mapping the signature onto our data using the signature finder algorithm, cluster 20 was identified as liver capsule macs (Figure S5C). Clusters 2 and 16, which were specific to WD-fed mice, could not yet be further identified and were called Mac1 and Mac2, respectively (Figure 4F). Cluster 18 identified a group of cells that shared expression of Mmp12 (Figures S5D and S5E). However, mapping individual cells in this cluster onto the other clusters using the FindTransferAnchors function of the Seurat R package showed that this was a heterogeneous population of ResKCs, moKCs, Mac1s, and Mac2s; thus, for further analysis, they were included with the cells of the same identity.

15To determine which clusters represented the macs identified in our flow cytometry analysis, we purified the 3 subsets based on CLEC4F and TIM4 from SD- and WD-fed mice (12 and 24 weeks) and performed bulk RNA-seq (Figure S6A). PCA analysis and comparison of the 3 populations in a triwise plot revealed that the majority of DEGs between the 3 populations were either specific to the CLEC4F- macs (Spp1, Chil3, Ccr2, and Gpr183) or shared between the moKCs and ResKCs with the exception of residency genes specific to the ResKCs, including Timd4, Cd163, C6, Xlr, and Marco (Figure S6A). To define the signature of each population, we compared the cells to one another and took the genes with a Log fold change (FC) >2. This process was refined for the CLEC4F- macs by an additional comparison of their DEGs with BM monocytes. Similarly, an additional comparison was carried out between moKCs and ResKCs, to ensure selection of DEGs that were unique to the moKC population. Using the signature finder algorithm to map the resulting signatures (Table S5) onto the scRNA-seq data (Figure S6B), we confirmed the annotations of the populations in the scRNA-seq data. Notably, while the CLEC4F- macs mapped primarily onto the Mac2s, they also showed some similarities to the Mac1s (Figure S6B), indicating that the CLEC4F- cells are heterogeneous.

16To investigate the heterogeneity of the CLEC4F- macs, we used triwise plots to define the DEGs that separated the as yet unidentified Mac1s and Mac2s from the other recruited population, the moKCs (Figure S6C). Comparison of these DEGs with ResKCs confirmed the similarity between moKCs and ResKCs, while the 2 unidentified mac clusters were distinct (Figure 5A; Table S4). The Mac1s were enriched for genes such as Cx3cr1, Itgax, H2-M2, and Olfml3, while the Mac2s were enriched for Spp1, F10, Chil3, Fabp5, and Gpnmb expression (Figures 5A and S6C-S6E). Notably, the gene expression profile of Mac1s somewhat resembles moKCs (Figure 5A) and macs en route to becoming moKCs. While this manuscript was under review, Clec1b (encoding CLEC2) was proposed to be a good marker of moKCs that appeared to be expressed earlier than CLEC4F. To confirm this, we examined expression of CLEC4F and CLEC2 in Clec4f-DTR mice 3 and 6 days post-ResKC depletion during the generation of moKCs. This analysis confirmed CLEC2 to be an early marker of macs becoming moKCs, and hence, we termed the CLEC2+CLEC4F- macs “pre-moKCs” (Figure S6F). As the Mac1s seemed to resemble macs en route to becoming moKCs, we checked if these would be pre-moKCs. Fitting with the scRNA-seq data, the CLEC4F- macs could be split into 2 populations based on CLEC2 expression (Figures S6D and S6G). Thus, Mac1s are pre-moKCs. To define Mac2s, we next performed KEGG pathway analysis (Table S4). This suggested the Mac2s to be relevant in diseases, including MAFLD (Figure 5B). While Cd9 and Trem2 expression was increased across mac subsets in MAFLD (Table S6), the overall signature of the Mac2s closely resembled those of Cd9- and Trem2-ex- pressing lipid-associated macrophages (LAMs) described recently in obese adipose tissue (AT; Figure S6H), suggesting that this cluster may be their hepatic counterpart. Indeed, when we calculated the signature of the AT LAMs compared with other AT macs (Table S7), the signature finder algorithm mapped the LAM signature onto the Mac2s (Figure 5C). Thus, we called these cells hepatic LAMs. Comparison of the DEGs between hepatic and AT LAMs identified 67 DEGs with a Log FC >1.5, indicating tissue-specific influences on the differentiation of these cells (Figure S6I; Table S7). The hepatic LAMs also showed some overlap with the scar-associated macs that have recently been described in fibrotic human livers and whose signature is also enriched for expression of TREM2 and CD9 (Figure S6J). Examination of the CITE-seq data did not reveal any specific antibodies for phenotypic analysis of the hepatic LAMs, while flow cytometry analysis of potential surface markers showed reactivity that was either non-selective (e.g. CD9) (Figures 5D and 5E) or gave no signal above controls (GPNMB, CD63; data not shown). Thus, we next examined other suitable markers of the LAM population and identified Spp1 to be specifically enriched in LAMs (Figure 4I). As Spp1 encodes the che- mokine osteopontin, implicated in fibrosis and NASH, we tested whether it could be used as a marker of hepatic LAMs. Using the PrimeFlow assay to measure mRNA transcripts by flow cytometry, we detected Spp1 primarily in CLEC4F- macs, identifying the hepatic LAMs (Figures 5D and 5E). While the SPP1 antibody proved unsuitable for flow cytometry, we were able to confirm the presence of SPP1-expressing macs in MAFLD by immunofluorescence, using EPCAM to exclude SPP1-expressing cholan- giocytes (Figures 5F and 3B).

17We next sought to investigate how the different macs compared functionally. We first examined expression of a list of generic prototypical immune activation-associated genes in hepatic LAMs, pre-moKCs, moKCs, and ResKCs pooled from mice fed the WD (24 and 36 weeks). Overall, the different KCs expressed these genes similarly, while hepatic LAMs expressed lower Il18, Fpr2, Tlr4, Cd38, Tgm2, Mrc1, and Abca1 and higher Spp1, Ccl3, Itgax, Socs3, Chil3, and Plin2 (Figures 6A and S7A). We next examined expression of lipid metabolism genes, as this is proposed to be a main function of ResKCs. This revealed a similar pattern in which moKCs were similar to ResKCs, with the exception of Apoc1 (lower) and Pparg (higher), whereas LAMs showed lower or equal expression of most of these genes compared with the KC populations, with the exception of Lpl &Pparg (Figures 6B and S7B). Pre-moKCs had an intermediate phenotype between the KCs and the LAMs (Figure 6B). Consistent with these findings, CLEC4F- macs contained fewer neutral lipids than their ResKC or moKC counterparts (Figure 6C), as measured by BODIPY or LIPIDTOX staining. Despite containing more neutral lipid than CLEC4F- macs, moKCs also contained less lipid than the ResKCs (Figure 6C). To examine this further, we performed lipidomic analysis. Due to a number of factors including cell numbers required, the lack of a Spp1 antibody, the lack of knowledge of CLEC2 expression at the time of analysis, and the overall similarity between ResKCs and moKCs, we compared total CLEC4F+ KCs with CLEC4F- macs. This showed that although KCs and CLEC4F- macs shared some lipid species, others were found only in KCs and not in CLEC4F- macs (Figure 6D), further suggesting that these cells have distinct abilities to metabolize lipids. Thus, hepatic LAMs and moKCs represent distinct fates for monocytes recruited to the liver in MAFLD in terms of their transcriptome, localization, and function.

18Finally, we examined how ResKCs responded in MAFLD. The gradual loss of ResKCs during MAFLD indicates that these cells are affected by the lipemic environment, and it is generally believed that immune activation of KCs occurs in MAFLD. However, the exact effects of MAFLD on KCs are contentious. Consistent with a recent report, we did not observe immune activation of KCs at any stage of MAFLD in either the scRNA-seq or bulk RNA-seq analysis, as evidenced by expression of genes associated with immune cell activation. Although we observed a slight decrease in Tnfa, Il1b, and Il10 expression in ResKCs from WD-fed mice (Figures 7A and S7C), no differences were found in pro-inflammatory cytokine production (Figure 7B), thus demonstrating that ResKCs are not pro-inflammatory in MAFLD. Furthermore, although changes could be seen in the ResKC transcriptome when examining all the DEGs between SD- and WD-fed mice at the different time points, the scale of these changes was modest compared with the differences between KCs and CLEC4F- macs (Table S6). Using a log FC >1.5 as a cut-off, analysis of the scRNA-seq data revealed 2 DEGs in ResKCs at 12 weeks, 0 DEGs at 24 weeks, and 13 DEGs at 36 weeks (Figure 7C). Analysis of the more in-depth bulk RNA-seq data showed 1 DEG in WD ResKCs at 12 weeks and 100 DEGs at 24 weeks, compared with 781 DEGs between WD ResKCs and CLEC4F- macs at 24 weeks (Figure S7D; Table S6). KEGG pathway analysis suggested that many of the upregulated genes in WD ResKCs were associated with cell adhesion and cytokine-receptor interactions (data not shown). There was also some increased expression of genes associated with metabolically activated macs (MMe) in ResKCs from mice with MAFLD (Figures 7A and S7C); however, this did not correlate with a difference in the ResKC metabolome between SD and WD (12 weeks), although the ResKC profile was distinct from in vitro M0, M1, and M2 macrophages (Figures 7D and S7F). Moreover, LIPIDTOX and BODIPY staining showed that the neutral lipid content of ResKCs did not increase in WD-fed mice at 12, 24, or 36 weeks (Figures 7E–7G); rather, there was a slight decrease after 36 weeks (Figures 7E and 7F). Analysis of the ResKC lipid metabolism profile at the different time points on the diet identified a trend toward slightly increased expression of these genes in WD-fed mice (Figures 7H and S7E), possibly compensating for the increased lipid load.

19Focused ion beam milling combined with scanning electron microscopy (FIB-SEM) and lipidomics analysis also did not identify any overt differences in either lipid droplet volume or in the lipid species present in CLEC4F+ KCs isolated from mice fed the SD or WD (Figure S7G; data not shown). Taken together, this suggests that while ResKCs react in MAFLD, they are not pro-inflammatory, nor do they harbor an increased or altered lipid load. Thus, the biggest difference in MAFLD macs stems from the recruitment of LAMs which are distinct from ResKCs, moKCs, and pre-moKCs, highlighting the need to differentiate between these mac subsets.

20MAFLD represents a spectrum of disease states ranging from simple steatosis to NASH, which can lead to cirrhosis and HCC. However, not all patients progress from MAFLD to NASH, and while the reasons why some do and some don’t remain largely unclear, hepatic macs have been proposed to play a role in this. Recently, it has become clear that hepatic macs, especially in disease settings, do not represent a homogeneous population of ResKCs but rather can exist in multiple subsets and/or activation states. To begin to assess the roles played by the distinct mac populations in MAFLD, we set out to characterize mac heterogeneity and localize the subsets within the liver. We identified four subsets of hepatic macs in MAFLD, including ResKCs and 3 subsets of recruited macs, moKCs, pre-moKCs, and the distinct hepatic LAMs. Notably, all populations were found within the KC niche in contact with hepatic stellate cells (HSCs) and liver sinusoidal endothelial cells (LSECs). However, LAMs were predominantly found in regions characterized by increased Desmin expression, suggestive of fibrosis. During reviewing, two additional papers were published investigating macs in MAFLD. It is currently unclear exactly how the populations identified here correlate to those described by ). Comparison of the datasets would suggest that the hepatic LAMs would most likely be present in the KN-RM gate; however, with the expression pattern of F4/80, TIM4, and CD11b, some LAMs may also be in the Ly6Clo-RM gate. However, as distinct diets were used, this may also affect the populations present. With this in mind, it will also be important to examine whether LAMs are found within CLEC2- macs recruited in the MCD-model. Additionally, ) have also described a population of macs termed NASH-associated macs (NAMs). In contrast to our study, the NAMs were identified as KCs. However, only a subset of these cells expressed KC identity genes like Cd5l or Clec4f, suggesting that these may also include LAMs.

21How important is the distinction between the recruited mac subsets? While others have suggested minimal differences between TIM4- macs and ResKCs in NASH, by dividing the population into moKCs and LAMs, we identified that moKCs largely resembled ResKCs, while LAMs had more than 700 DEGs compared with ResKCs, including differences in lipid metabolism and immune activation. Moreover, as the hepatic LAMs best resemble scar-associated macrophages described in fibrotic human livers, this distinction is also clinically relevant. Moreover, hepatic LAMs expressed Sppi encoding the chemokine Osteo- pontin. Osteopontin has recently been described as a good biomarker of NASH in patient serum. Thus, as hepatic LAMs were only identified in the later stages of disease correlating with worsening disease and fibrosis, it is tempting to speculate that this increase in osteopontin observed in patients could be attributed to the hepatic LAMs. In addition, osteopontin has been implicated in driving collagen I production from HSCs and thus is thought to contribute to fibrosis. This could suggest that hepatic LAMs drive the progression to NASH. Indeed, recent studies blocking osteopontin in mice models of NASH have suggested a protective effect. Unfortunately, to date, we do not have the tools to target this population specifically to address these questions, but generating these is an important aim for the future.

22Turning our attention to the ResKCs, we found that, contrary to the current line of thinking, ResKCs are not pro-inflammatory in MAFLD. Thus, this questions the proposed role of KCs as drivers of NASH, through their role as inflammatory mediators. While not pro-inflammatory, ResKCs had an altered transcriptome in MAFLD; however, the scale of these changes was modest compared with differences between ResKCs and LAMs. Moreover, minor, if any, changes were observed in the metabolomics and lipidomics profiles of KCs. One possibility is that the modest response observed here is because we can only profile the remaining ResKCs. Perhaps if we were to catch the ResKCs just prior to their loss, increased differences would be observed. The modest response of ResKCs is in contrast to a recent report, where over 800 DEGs genes have been identified between NASH and healthy ResKCs. For example, we did not observe the altered expression of C6 or Cdl63. The reasons for these differences remain unclear but could be related to the timing or diet used. As many KC residence and identity genes were downregulated in NASH TIM4+ KCs in the study from, one possibility is that the proportion of moKCs acquiring TIM4 expression may be different between the models, leading to a more immature KC profile in NASH in the study from the Glass lab. Indeed, many of the DEGs reported between NASH and healthy KCs, including C4b, Timd4, Marco, and Hmoxi, were dysregulated between our moKCs and ResKCs. However, overall, the differences between moKCs and ResKCs in our model were also relatively modest (72 DEGs, LogFC > 1.5).

23Despite these differences, our observation that ResKCs were gradually lost from the tissue in MAFLD is consistent with the other recent studies. Regarding why ResKCs fail to self-maintain in MAFLD, one hypothesis is that they are no longer correctly adapted to the niche. While our data did not identify any major changes in the niche signals important for KC development and maintenance, we observed an increase in niche cell proliferation, suggesting an increased need to maintain cell numbers in MAFLD. Moreover, we also identified differences in the transcriptomes of these cells. However, further work is required to better understand the precise signals at play. With the knowledge that ResKCs have been suggested to be protective in NASH, our data suggesting a lack of ResKCs in HCC and the finding that osteopontin expression correlates with worse disease, perhaps if we can understand the signals driving ResKC loss and LAM generation in MAFLD, this could open the door to therapeutic options for patients to prevent and/or reverse progression to NASH and HCC.

24Together, our data highlight the heterogeneity within the macrophage pool in MAFLD and highlight the need for examination of this heterogeneity when considering therapeutic options.

25Detailed methods are provided in the online version of this paper and include the following: Key Resources Table Lead ContactMaterials AvailabilityData and Code AvailabilityAdditional Resources Experimental Model and Subject DetailsIn Vivo Animal Studies Method DetailsDiet induced MAFLDIntraperitoneal Glucose Tolerance TestInsulin ELISAColorimetric assaysHistological analysisIsolation of Liver CellsFlow Cytometry and Cell SortingConfocal microscopyGeneration of BM chimerasCSF1Fc administrationDT administrationRNA Sequencing, CITE-seq and qPCRPreprocessing DataMetabolomic profiling of MacrophagesGeneration of M0, M1 and M2 in vitro macrophagesLipidomic profiling of MacrophagesHCS LipidTOX Deep Red staining for microscopyFIB-SEMLipid droplet analysisQuantification of MicroscopyQuantification and Statistical Analysis

26Key Resources Table

27Lead Contact

28Materials Availability

29Data and Code Availability

30Additional Resources

31Experimental Model and Subject Details

32In Vivo Animal Studies

33Method Details

34Diet induced MAFLD

35Intraperitoneal Glucose Tolerance Test

36Insulin ELISA

37Colorimetric assays

38Histological analysis

39Isolation of Liver Cells

40Flow Cytometry and Cell Sorting

41Confocal microscopy

42Generation of BM chimeras

43CSF1Fc administration

44DT administration

45RNA Sequencing, CITE-seq and qPCR

46Preprocessing Data

47Metabolomic profiling of Macrophages

48Generation of M0, M1 and M2 in vitro macrophages

49Lipidomic profiling of Macrophages

50HCS LipidTOX Deep Red staining for microscopy

51FIB-SEM

52Lipid droplet analysis

53Quantification of Microscopy

54Quantification and Statistical Analysis

55The following mice were used in this study; C57BL/6J (Janvier), Ccr2GFP/+ and Clec4f-DTR. All mice were used on a C57BL/6 background and only male mice were used for diet experiments while a mix of male and female mice were used for all experiments performed with normal chow diet. Mice were put on the diets from 5 weeks of age and sacrificed at indicated time points. All other mice were used between 6 and 12 weeks of age. All mice were bred and/or maintained at the VIB (Ghent University) under specific pathogen free conditions. All animals were randomly allocated to experimental groups. All experiments were performed in accordance with the ethical committee of the Faculty of Science, UGent and VIB.

56To induce MAFLD and NASH, mice were fed a western diet (WD) high in fat, sugar and cholesterol as described previously. This consisted of 58% fat, 1% cholesterol (Research Diets; D09061703i) and drinking water was supplemented with 23.1g/L fructose (MPBio) and 18.9g/L sucrose (VWR). Control mice were fed a standard diet with 11kcal% fat with corn starch (D12328i; Research Diets).

57Following overnight fastening, mice were administered 2g/kg glucose via intraperitoneal injection. Blood glucose concentrations were measured 0, 15, 30, 60, 120 and 180 min later.

58Following overnight fasting, blood was collected from the tail, serum was collected and fasting insulin were measured on undiluted serum according to the manufacturer’s instructions using the Ultra Sensitive Mouse Insulin ELISA kit (Crystal Chem)

59After sacrifice, mice were bled by cardiac puncture and serum was collected. Serum cholesterol & cholesteryl esters were measured according to the manufacturer’s instructions using the Cholesterol & Cholesteryl Ester Quantitation Kit (Abnova). Serum triglycerides were measured according to the manufacturer’s instructions using the Triglyceride Colorimetric Assay Kit (Cayman).

60Livers were removed rapidly, a piece taken and fixed immediately in fresh 4% paraformaldehyde at RT for 24 h. Fixed livers were then dehydrated and embedded in paraffin, after which 5 μm cross sections were obtained using a Microm HM360. Sections were then deparaffinized, rehydrated and stained with hematoxylin and eosin using standard protocols. Sections were also stained for Picro- Sirius Red according to the manufacturer’s instructions using the Picro Sirius Red Stain kit (Abcam). Sections were imaged using a Zeiss AxioScan.Z1.

61Liver cells were isolated by liver perfusion and digestion as described previously. Briefly, after retrograde cannulation, livers were perfused for 1-2mins with an EGTA-containing solution, followed by a 5min (6ml/min) perfusion with 0.2mg/mL collagenase A. Livers were then removed, minced and incubated for 20mins with 0.4mg/mL collagenase A and 10U/ mL DNase at 37°C. All subsequent procedures were performed at 4°C. Samples were filtered over a 100 μm mesh filter and red blood cells were lysed. Samples were again filtered over a 40 μm mesh filter. After two centrifugation steps of 1 min at 50 g to isolate hepatocytes, remaining liver cells (leukocytes, LSECs and HSCs) were centrifuged at 400 g for 5mins before proceeding to antibody staining for flow cytometry.

620.5-5x106 cells were pre-incubated with 2.4G2 antibody (Bioceros) to block Fc receptor sand stained with appropriate antibodies at 4°C in the dark for 30-45 min. Cell viability was assessed using Fixable Viability dyes (eFluor780 oreFluor506; Thermo Fisher) and cell suspensions were analyzed with a BD FACSymphony or purified using a BD FACSAria II or III and FlowJo software (BD). Bodipy staining (Thermo Fisher) was performed after surface antibody staining for 15 min at room temperature. The full list of antibodies used can be found in the Key Resource Table. The Primeflow assay (Thermo Fisher) for Spp1 expression was performed in 96-well U-bottom plates according to the manufacturer’s instructions. Ki67 and Tbet staining was performed by fixing and permeabilizing extracellu- larly stained cells according to the manufacturer’s instructions using the FoxP3 Transcription factor staining buffer Kit (eBioscience). Lipidtox staining was performed after fixing and permeabilizing extracellularly stained cells according to the manufacturer’s instructions using the Fixation/Permeabilization Solution Kit (BD Cytofix/Cytoperm). To measure intracellular cytokines, 0.5-5x106 cells were incubated for 3.5 h at 37°C in DMEM with 10%FCS, 1% glutamax, 0.1% ß-mercaptoethanol, 0.5% gentamicin with 1X mon- ensin and 1X brefeldin A (Biolegend). After incubation, cells were stained extracellularly as above. Intracellular cytokine staining was performed by fixing and permeabilizing extracellularly stained cells according to the manufacturer’s instructions using the Fixation/ Permeabilization Solution Kit (BD Cytofix/Cytoperm).

63Confocal staining was performed as described previously. Immediately after sacrificing mice with CO2, inferior vena cava were cannulated and livers were perfused (4 mL/min) with Antigenfix (Diapath) for 5 min at room temperature. After excision, 2-3 mm slices of livers were fixed further by immersion in Antigenfix for 1 h at 4°C, washed in PBS, infused overnight in 34% sucrose and frozen in Tissue-Tek OCT compound (Sakura Finetek). 20 μm-thick slices cut on a cryostat (Microm HM 560, Thermo Scientific) were rehydrated in PBS for 5 min, permeabilized with 0.5% saponin and non-specific binding sites were blocked for 30 min with 2% bovine serum albumin, 1% fetal calf serum and 1% donkey or goat serum for 30 min. Tissue sections were labeled overnight at 4°C with primary antibodies followed by incubation for 1 h at room temperature with secondary antibodies. When two rat antibodies were used on the same section, the directly conjugated rat antibody was incubated for 1 h after staining with the unconjugated and anti-rat secondary and after an additional blocking step with 1% rat serum for 30 min. Slides were mounted in ProLong Diamond, imaged with a Zeiss LSM780 confocal microscope (Carl Zeiss, Oberkochen, Germany) with spectral detector and using spectral unmixing and analyzed using ImageJ and QuPath software.

64Bone marrow chimeras were generated as described previously. Briefly, 5 week old male C57BL/6J wild-type mice (CD45.2) on normal chow were anaesthetized by intraperitoneal administration of Ketamine (150 mg/kg) and Xylazine (10 mg/kg). Mice were lethally irradiated with 8 Gy, with the livers being protected with a 3-cm-thick lead cover. Once recovered from the anesthesia, mice were reconstituted by intravenous administration of 5-10 × 106 BM cellsfrom CD45.1/CD45.2 wild-type mice. Mice were placed on either the SD or WD immediately after reconstitution. Chimerism was assessed 18 or 24 weeks after irradiation by flow cytometry.

65Mice received the SD or WD for 24 weeks, before being injected subcutaneously with 1 mg/kg CSF1Fc or PBS for 4 days. Mice were sacrificed on day 6 and the % and number of KCs was assessed by flow cytometry.

66Clec4f-DTR mice or WT littermate controls were injected intraperitoneally with 500ng DT. Mice were sacrificed on 3 or 6 days after DT administration.

6740000-160000 live CD45+ cells or CD45- cells from livers of a C57BL/6 mouse fed the SD or WD for 12,24 and 36 weeks were purified, centrifuged at 400 g for 5 min and stained with 2.4G2 antibody to block Fc receptors and CITE-seq antibodies for 20mins at 4°C, before being washed in excess PBS with 2% FCS and 2mM EDTA. Cells were then resuspended in PBS with 0.04% BSA at ~1000 cells/mL. Cell suspensions (target recovery of 8000-10000 cells) were loaded on a GemCode Single-Cell Instrument (10x Genomics, Pleasanton, CA, USA) to generate single-cell Gel Bead-in-Emulsions (GEMs). ScRNA-Seq libraries were prepared using GemCode Single-Cell 3’Gel Bead and Library Kit (10x Genomics, V3 technology) according to the manufacturer’s instructions. Briefly, GEM-RT was performed in a 96-Deep Well Reaction Module: 55°C for 45min, 85°C for 5 min; end at 4°C. After RT, GEMs were broken down and the cDNA was cleaned up with DynaBeads MyOne Silane Beads (Thermo Fisher Scientific, 37002D) and SPRIselect Reagent Kit (SPRI; Beckman Coulter; B23318). cDNA was amplified with 96-Deep Well Reaction Module: 98°C for 3 min; cycled 12 times: 98°C for 15 s, 67°C for 20 s, and 72°C for 1 min; 72°C for 1 min; end at 4°C. Amplified cDNA product was cleaned up with SPRIselect Reagent Kit prior to enzymatic fragmentation. Indexed sequencing libraries were generated using the reagents in the GemCode Single-Cell 3’ Library Kit with the following intermediates: (1) end repair; (2) A-tailing; (3) adaptor ligation; (4) post-ligation SPRIselect cleanup and (5) sample index PCR. Pre-fragmentation and post-sample index PCR samples were analyzed using the Agilent 2100 Bioanalyzer.

68RNA was extracted from 25000 sorted macrophages (CLEC4F+ TIM4+, CLEC4F+ TIM4- or CLEC4F- TIM4-) or from 5000-25000 Hepatocytes, LSECs, HSCs or cholangiocytes from livers of C57BL/6 mice fed the SD or WD for 12, 24 and 36 weeks using a RNeasy Plus micro kit (QIAGEN). Sensifast cDNA synthesis kit (Bioline) was used to transcribe total RNA to cDNA. Real-time RT-PCR using SensiFast SYBR No-Rox kit (Bioline) was performed to determine gene expression, therefore a PCR amplification and detection instrument LightCycler 480 (Roche) was used. Gene expression was normalized to ß-actin gene expression.

69Libraries were constructed using the Illumina TruSeq RNA Preparation Kit. RNA sequencing was performed at the VIB Nucleomics Core using Illumina NextSeq500 with these parameters: High Output v2.5, 75 bp, Single Reads (76-8-8-0), 1.4 pM + 1.73% PhiX. All samples passed quality control based on the results of FastQC (v0.11.9). Reads were mapped to the mouse reference genome (mm10) via HiSat2 (v2.2.0) with max-intronlen set to 1000000. The aligned reads were counted via FeatureCounts (v2.0.0). The R package limma (v3.42.2) was used to normalize the data and to perform differential expression analysis. Genes that didn’t have a count per million (cpm) value > 1 in at least 5 samples were removed. As such, we continued with 13421 genes. Benjamini-Hochberg was used to adjust the p values for multiple testing. The DE genes were defined based on a log2-ratio >1 or < -1 and adj.P value < 0.05. For the heatmap we first transformed the normalized expression table as ‘log2(2^expTable + 1)’ and subsequently scaled the values per gene by calculating the mean expression per gene and then subtracting that mean value of each expression value. The R package triwise (v0.99.5) was used to create the triwise plots.

70Sequencing libraries were loaded on an Illumina HiSeq with sequencing settings recommended by 10X Genomics (26/8/0/98 –2.1pM loading concentration, ADT and cDNA libraries were pooled in a 25:75 ratio). Sequencing was performed at the VIB Nucleomics Core (VIB, Leuven). The demultiplexing of the raw data was performed using CellRanger software (10x– version 3.1.0; cellranger mkfastq which wraps Illumina’s bcl2fastq). The reads obtained from the demultiplexing were used as the input for ‘cellranger count’ (CellRanger software), which aligns the reads to the mouse reference genome (mm10) using STAR and collapses to unique molecular identifier (UMI) counts. The result is a large digital expression matrix with cell barcodes as rows and gene identities as columns. The aggregation of the samples was performed using ‘cellranger aggr’ (CellRanger software).

71Preprocessing of the data was done by thescater R package (v1.12.2) according to workflow proposed by the Marioni lab. Outlier cells were identified based on 3 metrics (library size, number of expressed genes and mitochondrial proportion) and cells were first tagged as outliers when they were x median absolute deviation (MADs) higher or lower than the median value of each metric. For library size 4 MADs was used, for the number of expressed genes 4 MADs for used and for the mitochondrial proportion 5 MADs was used. For the patient data were respectively 3 MADs, 3 MADs and 5 MADs used. As second filtering, a principal component analysis plot was generated using the runPCA function of the scater R package (default parameters). Outlier cells in this principal component analysis plot were identified using the R package mvoutlier. Forthe CD45-cells, due to inclusion of hepatocytes an additional cleanup was necessary to remove ambient mRNA signals. Samples were corrected for ambient RNA using the R package FastCAR (v0.1.0). We defined an empty droplet as a cell barcode with at least 10 UMIs and less than a certain cut off. To define this cut off we first made a plot showing on the y axis how many UMIs are present in a barcode and on the x-as the number of barcodes (same plot as in the CellRanger web_summary output). By making this plot, a clear ‘drop’ in UMI counts becomes visible and we put ourcut off just before this ‘drop’. This cut off is always lower than the minimal UMI count from the ‘filtered’ CellRanger output. The corrected counts were merged together using the R package Harmony (v1.0). For all samples, after these initial analysis steps, additional low quality (low UMI counts, > 8% of mitochondrial genes), contaminating (potential doublets) and actively proliferating cells were also removed from the analysis. Low-abundance genes were removed by removing all genes that weren’t expressed (count > 0) in at least 3 cells. The counts were normalized and log2 transformed via the NormalizeData function of the Seurat R package (v3.1.1) using default parameters. Detecting highly variable genes, finding clusters and creating UMAP plots was done using the Seurat pipeline. For the clustering, the first 20 principal components were used and 0.6 was used as the resolution parameter. The monocytemacrophage UMAP was generated starting from the counts and using specific cells from the full UMAP (based on Mafb, Ly6c2, Fcgr1 and Adgre1 expression, 21156 cells). Then the same Seurat pipeline was followed as described above. The first 20 PCs were used and 0.8 was the resolution parameter. Differential gene expression was assessed using the findMarker function of the Seurat pipeline.

72For the CITE-Seq data, we first checked in how many cells each antibody was expressed. The lowest expressed antibody was expressed in only 1 cell and was hence excluded from further analysis, while the second lowest antibody was expressed in 41% of the cells. The expression data were processed using the Seurat workflow, with CLR normalization and scaling of the data performed using the default parameters. Marker Enrichment Modeling was performed on the CITE-Seq data using the MEM R package (v2.0.0).The MEM heatmap was generated using pheatmap (v1.0.12). For each antibody we checked per cluster if the MEM values were similar to those of the isotype control antibodies. When MEM values of an antibody are ± 0.5 compared with the MEM values of the isotype across all clusters, the antibody was considered to not give a real signal above background and was hence removed from further analyses. The plots showing the expression of specific antibodies (Figure 1B) were generated by collating all cells with a count > 0 and calculating the 98% quantile cut-off of counts, with all the cells above this value being colored in red.

73ResKCs (10x104 - 25x104) were sorted in DMEM supplemented with 10% FCS (Bodinco) using a BD FACSARIA II and ARIA III. Cells were then incubated for 30 min at 37°C, before being pelleted at -9°C and washed in 1ml ice cold NaCl (0.9%). Cells were then pelleted at -9°C and snap frozen in liquid nitrogen. Samples were than extracted in 100uL of 30:50:20 acetonitrile:methanol:milli-Q H2O pre-cooled at -20°C, dried using speed vac and resuspended in 20uL of 30:50:20 acetonitrile:methanol:milli-Q H2O for metabolites that are identified in negative ionization mode (polar), or in H2O for metabolites that are identified in positive ionization mode (nonpolar). Metabolite quantification was carried out using Agilent 1290 Infinity II UHPLC in line with a Bruker impact IIQTOF-MS operated in full scan (MS1) mode. LC separation was performed on a Waters CSH-C18 column (100 × 2.1 mm, 1.7 μm particles) using a binary solvent gradient of 100% buffer A (0.1% formic acid in water) to 97% buffer B (50:50 methanol:acetonitrile). Flow rate was 400 μL/min, autosampler temperature was 4°C, and injection volume was 3 μL. Data processing was performed using TASQ™ Software (Target Analysis for Screening and Quantification) (Bruker Daltonics Inc.).

74Bone marrow cells were isolated from tibias and femurs of WT mice and grown in complete medium (RPMI-1640 medium containing 10 mM glucose, 2 mM L-glutamine, 100 U ml-1 penicillin-streptomycin and 10% FCS) with 20 ng ml-1 murine macrophage colonystimulating factor 1 (CSF-1; Peprotech) for 7 days, and supplemented with CSF-1 on days 3 and 5. On day 7 macrophages were harvested and then maintained in 20 ng ml-1 CSF-1 for subsequent experiments in which they were either maintained in medium alone (M0), or stimulated with 50 ng ml-1 IFN-μ (R&D systems) and 20 ng ml-1 LPS (M1), or 20 ng ml-1 IL-4 (M2) (Peprotech) for 18 h.

75The protocol for lipid extraction was adapted from Matyash et al.. Frozen cell pellets (60000-150000 cells) were resuspended in ice cold PBS and transferred to glass tubes before the addition of methanol and methyl tert-butyl ether. The tubes were then shaken for 1 h at 4°C. Water was added to separate the phases before centrifugation at 1,000 x g for 10 min. The upper organic phase was collected and dried in a Genevac EZ2 speed vac. Samples were resuspended in 2:1:1 isopropanol:acetonitrile:- milli-Q H2O priorto analysis. LC-MS was carried out using an Agilent Zorbax Eclipse Plus C18 column using an Agilent 1290 Infinity II UHPLC inline with an Agilent 6495 Triple Quad QQQ-MS. Lipids were identified by fragmentation and retention time, and were quantified using Agilent Mass Hunter software.

7625000 ResKCs were sorted into DMEM supplemented with 10% FCS (Bodinco) using an ARIA II and ARIA III (BD Biosciences). Cells were incubated on coverslips pre-coated with poly-lysine for 30 min at 37°C. Cells were washed and then fixed using Antigenfix (Di- apath). Cells were washed and stained with DAPI (Invitrogen) and LipidTOX Deep Red staining (Invitrogen) for 1 h at room temperature. Slides were mounted in ProLong Diamond, imaged with a Zeiss LSM780 confocal microscope (Carl Zeiss, Oberkochen, Germany) and analyzed using ImageJ software.

77100000 CLEC4F+ KCs were sorted into DMEM supplemented with 10% FCS (Bodinco) using a BD FACSARIA II and ARIA III. Cells were resuspended in DMEM +10% FCS at room temperature and were incubated for 2 h at 37°C to adhere onto coverslips. Following that, samples were incubated in freshly prepared fixative (2% paraformaldehyde (PFA, Applichem), 2.5% glutaraldehyde (GA, EMS) in 0.1M sodium cacodylate (Sigma-Aldrich) buffer, pH7.4) at RT for 30 min. Fixative was removed by washing in 0.1M cacodylate buffer and samples were incubated in 1% osmium (OsO4, EMS), 1.5% potassium ferrocyanide (Sigma-Aldrich) in 0.1M cacodylate buffer for 40 min at RT. After washing in ddH2O, samples were incubated overnight at 4°C in 1:3 UAR in H2O (Uranyl Acetate Replacement, EMS). The next day, UAR was removed by washing in ddH2O. After final washing steps the samples were dehydrated using solutions of increasing EtOH concentration (30%, 50%, 70%, 90%, 2x 100%) for 3 min each. Subsequent infiltration with resin (Spurr’s, EMS) was done by first incubating in 50% resin in ethanol for 2 h, followed by at least 3 changes of fresh 100% resin (including 1 overnight incubation). Next, samples were embedded in fresh resin and cured in the oven at 65°C for 72 h.

78For FIB-SEM imaging, embedded cells were mounted on aluminum SEM stubs (diameter 12 mm) using conductive epoxy (Circuit Works) and samples were coated with ~20nm of Platinum (Quorum Q150T ES). FIB-SEM imaging was performed using a Zeiss Crossbeam 540 system with Atlas5 software. The Focused Ion Beam (FIB) was set to remove 5nm sections by propelling gallium ions at the surface. Imaging was performed at 1.5kV and 1nA using an ESB (back-scattered electron) detector.

79The FIB-SEM data was segmented using deep convolutional neural networks, specifically U-Net. In total, 2292 patches of size 256x256 were selected across 8 samples. The pixels in these patches that correspond to lipid droplets were labeled by an expert. The obtained labeled dataset was split in 66/34% train/test data. For training the network, we employed weighted cross entropy minimization (weight selected to balance classes) batch normalization, 16 initial feature maps (i.e., 4 times smaller as in) and standard data augmentations (small scaling, rotations, flips, random deformations and noise perturbations). A threshold is then applied to the final probability maps to obtain a segmentation of the droplets. This segmentation was further cleaned by removing noise components (objects smaller than 3x3x3 = 27 voxels) and applying a morphological opening with a spherical structure element (radius of 2 voxels). This final segmentation allowed us to extract the total volume of the droplets. The neural network was implemented in PyTorch and post-processing was performed using MATLAB. The code is publicly available on www.github.com/jorisroels/lipid-droplets.

80To detect the Macrophages, a custom groovy script (used with the eleventh milestone release on the path to QuPath v0.2.0) has been written using the F4/80 marker, and then cells have been classified based on their CLEC4F, TIM4 and F4/ 80 expression. The script splits the image into tiles, extracts the F4/80 channel image, applies a median filter to remove the noise, applies a threshold to keep only the positive signal object and uses the Analyze Particles from ImageJ to filter the smallest particle by area. The ROIs (Region of Interest) are sent to Qupath as annotation, the adjacent annotations due to the tiling are merged back and holes within the annotations are filled and converted to cell detections. Script is available upon request.

81Feature extraction was performed using QuPath and detected cells classified using QuPath based on their CLEC4F, TIM4 and F4/ 80 expression. Identification of Desminhi areas was performed using QuPath: based on pixel intensity from the Desmin channel, a pixel classifier was trained, the high/low Desmin Areas detected and the number of each cell type was then counted in Desminhi and Desminl¤ areas. Large vessels were identified manually based on expression of Desmin and CD31 and these regions were excluded from the analysis.

82In all experiments, data are presented as mean ± SEM and/or individual data points are presented unless stated otherwise. Statistical tests were selected based on appropriate assumptions with respect to data distribution and variance characteristics. Student’s t test (two-tailed) was used for the statistical analysis of differences between two groups. One-way ANOVA with Bonferroni’s multiple comparisons test was used for the statistical analysis of differences between more than two groups. Details of the precise test used for each analysis can be found in the figure legends. Statistical significance was defined as p < 0.05. Sample sizes were chosen according to standard guidelines. Number of animals is indicated as “n”. The investigators were not blinded to the mouse group allocation.