Proteomics of immune cells from liver tumors reveals immunotherapy targets

Summary Elucidating the mechanisms by which immune cells become dysfunctional in tumors is critical to developing next-generation immunotherapies. We profiled proteomes of cancer tissue as well as monocyte/macrophages, CD4+ and CD8+ T cells, and NK cells isolated from tumors, liver, and blood of 48 patients with hepatocellular carcinoma. We found that tumor macrophages induce the sphingosine-1-phospate-degrading enzyme SGPL1, which dampened their inflammatory phenotype and anti-tumor function in vivo. We further discovered that the signaling scaffold protein AFAP1L2, typically only found in activated NK cells, is also upregulated in chronically stimulated CD8+ T cells in tumors. Ablation of AFAP1L2 in CD8+ T cells increased their viability upon repeated stimulation and enhanced their anti-tumor activity synergistically with PD-L1 blockade in mouse models. Our data reveal new targets for immunotherapy and provide a resource on immune cell proteomes in liver cancer.


In brief
Canale et al. employed mass spectrometry-based proteomics to analyze immune cells from blood, liver, and tumor tissues of 48 hepatocellular carcinoma patients. They identified significant phenotype alterations in tumor-infiltrating immune cells, including SGPL1 upregulation in macrophages and AFAP1L2 in T cells. Genetic deletion of these genes in murine macrophages and T cells, respectively, enhanced antitumor activity, indicating their potential as immunotherapy targets.

INTRODUCTION
Tumors evolve as dynamic ecosystems consisting of cancer, stromal, and immune cells. The immune infiltrate typically consists of various cell types including cytotoxic T cells that play a central role in the immune response to tumors. Stimuli in the tumor microenvironment (TME) can suppress the functionality of T cells, which then lose the ability to control tumor growth. [2][3][4] To restore T cell functionality in tumors, checkpoint inhibitors (CPIs) have been developed that block inhibitory receptors on T cells. 5,6 CPIs have yielded great successes in clinical settings, but they are not always effective. This raises the need to identify novel strategies to further enhance immune responses to tumors.
One potential avenue to therapeutically improve anti-tumor immunity is to identify and target new molecular regulators underlying T cell dysfunction in tumors. Another possibility is to target tumor-associated macrophages (TAMs), which influence T cell function, i.e., through the production of cytokines that affect T cell differentiation. In many tumor types, TAMs have a tumor-promoting phenotype and are associated with poor prognosis. 7,8 Therefore, therapeutic strategies include depletion of TAMs or functional reprogramming of tumor-promoting TAMs into inflammatory macrophages that mediate immune responses to tumors. To devise therapeutic interventions that enhance T cell functionality or reprogram macrophages, it is important to understand how immune cells are regulated in tumors.
Proteins are directly involved in most biological processes; hence, their quantification is important to understand the phenotype of a cell. As there is no simple relationship between transcripts and proteins, their abundances do not always correlate. For example, protein abundance is influenced by the translation efficiency of a given mRNA, by the stability of the protein, as well as by posttranslational modifications affecting localization and degradation. 9 Mass spectrometry (MS)-based proteomics enables the direct quantification of thousands of proteins within cells or tissues. 9,10 Owing to advances in all areas of the MSbased workflow, including sample preparation, chromatography, MS instrumentation, and data analysis, it is possible to determine deep proteomes of clinical samples for target identification. 11 Previous studies employed transcriptomics or antibody-based protein quantification to profile tumor-infiltrating immune cells, 12-15 yet mass spectrometry-based proteomic profiling of immune cells from a large cancer patient cohort is lacking. Hence, a detailed proteomic analysis of immune cells in tumors offers important new insights to identify putative immunomodulatory interventions to treat cancer.
In this study, we profiled the proteomes of cancer and immune cells from patients with hepatocellular carcinoma (HCC), for which there are currently few treatment options. 16 Although HCC is typically infiltrated by immune cells, CPIs are only partially effective. 17,18 HCC can be caused by chronic infection with hepatitis B (HBV) or C (HCV) viruses or different metabolic and inflammatory disorders related to non-alcoholic and alcohol-related steatohepatitis. [19][20][21] Our proteomic data revealed that TAMs upregulate SGPL1, which we found inhibited their anti-tumor activity. We also identified AFAP1L2 as a new target for T cell-based immunotherapies. AFAP1L2 was expressed specifically in chronically stimulated CD8 + T cells in tumors, and its genetic ablation improved anti-tumor activity.

Proteomic profiling of an HCC patient cohort
To analyze proteomes of immune and cancer cells from liver tumors, we prospectively obtained tumor tissue, adjacent liver tissue, and peripheral blood from patients with HCC undergoing surgical resection. Of 48 patients, 9 had a previous infection with HBV, 20 with HCV, and 19 had no history of viral liver infection (Table S1). From each patient, blood, dissociated liver, and tumor tissues were subjected to Ficoll gradient centrifugation to separate mononuclear immune cells from which memory CD4 + T cells, memory CD8 + T cells, CD56 + NK cells, and CD14 + monocytes/macrophages were sorted to a purity of >98% ( Figures 1A  and S1). The numbers of immune cells isolated from tumors varied considerably between patients and ranged from <100,000 to 6 million cells per gram of tumor tissue ( Figure 1B).
From 32 tumor and 33 liver cell suspensions, we recovered a cell pellet from the Ficoll gradient, which contains hepatocytes and malignant and stromal cells ( Figure 1A). After homogeniza-tion and lysis of cell pellets with 4% SDS, proteins were precipitated, digested, and analyzed by liquid chromatography coupled mass spectrometry (LC-MS). A total of 8,182 protein groups were quantified with an average of 5,209 protein groups per sample ( Figure S2A, Table S2). A differential abundance analysis between liver and tumor tissues revealed profound alterations in tumor proteomes (Figures 2A and 2B). Data are accessible through our interactive platform www.immunomics. ch/hcc; an example is shown in Figure S2B. Tumor samples had typical molecular features of HCC based on a comparison to tumors from a different HCC patient cohort 22 (Figure 2A). In our data, PYCR2 was the most strongly upregulated protein in HCC. This enzyme catalyzes the formation of hydroxyproline, a major component of collagen that is associated with HCC tumor progression 23 ( Figure 2C). Two additional proteins that mediate metabolic flux toward hydroxyproline, P4HA1 and P4HA2, were also strongly upregulated in HCC. Among the proteins that were most downregulated in HCC was ASS1, which is involved in arginine biosynthesis ( Figure 2C), consistent with previous findings. 24 In summary, our proteomic analysis of immune cell-depleted liver and tumor tissues confirmed correct classification of tumorous and non-tumorous samples and recapitulated characteristic proteomic alterations in HCC tumors, 22 highlighting metabolic alterations related to hydroxyproline.
Next, we processed a total of 497 immune cell samples for proteomic analysis according to previously established protocols. 25 A total of 9,790 protein groups were quantified with an average of 5,832 protein groups per sample ( Figure 2D). All data are accessible through www.immunomics.ch/hcc. As expected, the lineage markers CD4, CD8, and CD56 were most abundant in the respective immune cell types, and the macrophage marker MRC1 was highest in liver and tumor macrophages ( Figure 2E). Consistent with previous findings, 26,27 the exhaustion marker CD39 was upregulated in tumor-infiltrating CD4 + and CD8 + T cells and macrophages compared with cells isolated from adjacent liver tissue and blood ( Figure S2C). Thus, the proteome data were in agreement with our cell sorting strategy and captured known regulations in tumor immunology.
Tumor macrophages display THY-1, acquired from the environment To analyze phenotypic changes in macrophages that infiltrate tumors, we performed a differential abundance analysis between proteomes of liver macrophages and TAMs and identified 15 proteins that were strongly increased in TAMs (Log2 fold change > 2; p value < 0.0001) ( Figures 3A and 3B). In addition, we analyzed transcriptomes of tumor macrophages from three HCC patients by RNA-seq and estimated copy numbers of transcripts, as previously described. 28 A comparison of transcript and protein copy numbers revealed a median protein-per-mRNA ratio of 500:1, but this ratio varied considerably (Figure 3C). For example, the chemokines CXCL8 and CCL3, which are secreted by macrophages, had a ratio lower than 10:1. In contrast, the chemokine CCL15 had a protein-per-transcript ratio greater than 10,000:1, because it is not expressed by macrophages but is taken up from the environment. Macrophages express the CCR1 receptor, which captures CCL15 secreted by liver cells. 29 Among the most enriched proteins in TAMs, THY-1, AKR1B10, and GGH had a protein-per-mRNA ratio greater than 10,000:1, suggesting they were acquired from the TME. AKR1B10, a potential diagnostic marker of HCC, 30 and GGH are both soluble metabolic enzymes that were likely taken up by macrophages through macropinocytosis. In contrast, THY-1 (CD90) is a GPIanchored membrane protein that is abundant on the surface of cancer stem cells, fibroblasts, and endothelial cells 31 but has not been found to be expressed by macrophages. We confirmed the presence of THY-1 protein on tumor macrophages by flow cytometry ( Figure S3A) and the absence of THY-1 mRNA by RNA in situ hybridization ( Figure S3B).
To confirm THY-1 protein transfer, we incubated bloodderived monocytes with patient-matched HCC tumor tissue fragments for 6 days and found that monocytes indeed acquired THY-1 protein from the TME (Figures S3C and S3E). Collectively, these data demonstrate that macrophages in liver tumors acquire proteins; hence, their phenotype is not exclusively shaped by transcription but also through interactions with the environment.
Deletion of SGPL1 promotes inflammatory macrophages that inhibit tumor growth Some of the 15 proteins that are enriched in tumor macrophages may render them anti-inflammatory and contribute to a tumorpromoting phenotype. We therefore assessed whether any of these 15 proteins are upregulated in functionally polarized M2 macrophages, which represent a model for anti-inflammatory macrophages. For this, freshly isolated monocytes from four healthy donors were in vitro polarized to an M1 (LPS + INF-g; inflammatory) and an M2 phenotype (IL-4 and IL-13; anti-inflammatory), and their proteome was analyzed by LC-MS (Table S3).
A differential abundance analysis confirmed strong enrichment of typical markers, such as GBP4 and GBP5 in M1 macrophages and CD209 and ALOX15 in M2 macrophages ( Figures 3D and  S3F). Of the 15 proteins upregulated in tumor macrophages, DAB2, SGPL1, and GLUL were enriched in M2 macrophages. DAB2 and GLUL were previously found to be upregulated in TAMs and M2 macrophages, and genetic ablation of Dab2 or Glul in murine macrophages was found to improve their antimetastatic function. 32,33 However, SGPL1 (sphingosine-1-phosphate [S1P] lyase) has previously not been associated with TAMs, and its function in macrophages is unknown.
SGPL1 is an endoplasmic reticulum membrane protein that irreversibly degrades S1P, a sphingolipid involved in immune signaling and inflammation. 34 To study the function of SGPL1 in vivo, we generated bone marrow-derived macrophages (BMDMs) from Rosa-Cas9 mice and lentivirally transduced them with two different sgRNAs targeting Sgpl1 ( Figure 3E). Tracking of indels by decomposition (TIDE) analysis 35 confirmed that the Sgpl1 locus was edited in 94% and 87% of BMDMs, respectively ( Figure S3G). We first assessed in vitro the capacity of Sgpl1-edited BMDMs to upregulate the costimulatory receptor CD86 and to produce inflammatory cytokines. For this, we stimulated BMDMs with LPS and IFN-g and then used flow cytometry to analyze the abundance of intracellular IL-6 and IL-12, as well as surface CD86. Sgpl1-edited BMDMs displayed more CD86 on their surface and more frequently produced IL-6 and IL-12 than control BMDMs ( Figures 3F-3H). An analysis of cell supernatants by ELISA showed that Sgpl1-edited BMDMs secreted 7-11 times more IL-12 than control BMDMs ( Figure 3I). This confirms that knockout of Sgpl1 increases IL-12 production, which drives anti-tumor immunity. 36 We then asked whether Sgpl1-edited BMDMs promote antitumor immunity in vivo. To test this, we mixed MC38 tumor cells together with Sgpl1-edited BMDMs in a 1:1 ratio and co-injected them into wild-type recipient mice ( Figure 3J). As a benchmark, we performed the same experiments with Dab2-edited BMDMs (editing efficiency of 67%, Figure S3H), which are known to have an anti-tumor effect. After 1 week, Sgpl1-edited BMDMs significantly inhibited tumor growth compared with control BMDMs ( Figure 3J). The extent of tumor growth inhibition was comparable to Dab2-edited BMDMs. Taken together, our data indicate that macrophages upregulate SGPL1 in tumors, which in a mouse model reduced their inflammatory response, and impaired their anti-tumor activity.

Tumor NK cells upregulate AFAP1L2
A differential abundance analysis between proteomes of NK cells isolated from liver and tumor tissues identified four proteins that were increased in tumor NK cells (Log2 fold change > 2; p value < 0.01): AKR1B10, GLUL, IGHMBP2, and AFAP1L2 ( Figures 4A and 4B). A further comparison with blood NK cells revealed that only the potential HCC marker protein AKR1B10, which was also enriched in TAMs ( Figures 3A and 3B), and AFAP1L2, a cytosolic signaling scaffold protein, 37 were specifically upregulated in tumor-infiltrating NK cells. Interestingly, inspection of a previously published human immune cell proteome atlas 38 revealed that AFAP1L2 was exclusively expressed in activated NK cells, while it was not present in resting NK cells or any other immune cell type at steady state or upon canonical activation ( Figure S4A). Taken together, AFAP1L2 is upregulated in activated, tumor-infiltrating NK cells.
AFAP1L2 is induced in TILs likely due to chronic stimulation We next analyzed T cells by flow cytometry and found that only one-third of HCC patients displayed a high frequency of PD1 + CD8 + T cells in tumors ( Figure 4C). 14 out of 15 of these patients had a viral etiology of disease (HBV or HCV), while 22 out of 36 of the PD-1 neg/low patients had a non-viral etiology of disease. We then compared proteomes of CD8 + tumor-infiltrating lymphocytes (TILs) of which greater than 60% were PD-1 + (10 patients) to proteomes of CD8 + TILs of which less than 20% were PD-1 + (15 patients). This differential abundance analysis revealed 22 proteins that were strongly upregulated in CD8 + TILs with a high frequency of PD-1 + T cells (log 2 fold change > 3, p value < 0.01) ( Figures 4D and 4E). This signature that correlated with high PD-1 expression included the nuclear factor TOX, which promotes T cell exhaustion, 39-41 the surface protein CD38, which is a marker for dysfunctional T cells, 42 as well as UBASH3B (STS2), a negative regulator of TCR signaling. 43 Interestingly, several proteins associated with proliferation were upregulated, such as seven DNA replication licensing factors (MCM2-7), SMC2, a central component of the condensin complex, and the proliferation marker Ki-67 ( Figures 4D and 4E).
Consistent with the notion that PD-1 marks proliferating T cells, we recovered significantly more CD8 + T cells from tumors when the percentage of CD8 + T cells expressing PD-1 was high ( Figure 4F). These data suggest that PD-1 + CD8 + T cells in HCC tumors can be highly proliferative, consistent with previous observations in melanoma tumors. 13 We asked which of the 22 PD-1-associated signature proteins were induced by T cell activation. For this, we isolated naive CCR7 + CD45RA + CD8 + T cells from the blood of four healthy donors, activated them with plate-bound antibodies to CD3 and CD28 for 48 h, and then cultured them for an additional 48 h. Samples were collected after increasing times and analyzed by LC-MS (Table S4). We found that the abundance of 19 out of the 22 ''PD-1 signature proteins'' was strongly increased after T cell activation ( Figure 5A). However, TOX, LMCD1, and AFAP1L2 were either not detected or only at very low levels, suggesting that their presence in tumor-infiltrating T cells was not a consequence of the canonical T cell activation program.
Because tumor-specific CD8 + TILs are chronically stimulated at the tumor site, we hypothesized that TOX, LMCD1, and AFAP1L2 might be induced upon chronic stimulation. To test this, we continuously stimulated naive CD8 + T cells for 14 days with plate-bound antibodies to CD3 and CD28. For comparisons, we activated naive CD8 + T cells for only 3 days and then either cultured them for 11 days or re-stimulated them after 8 days (Figure 5B). Markers associated with T cell exhaustion such as CD39, PD-1, TIM-3, and LAG3 were consistently increased in chronically stimulated T cells when analyzed by flow cytometry (Figures 5B and S4B), indicating that upon extensive in vitro stimulation, CD8 + T cells acquired phenotypic properties similar to TILs.
We then analyzed transiently activated, re-activated, and chronically stimulated T cells by proteomics and found that the 19 ''PD-1 signature proteins'' that were upregulated by T cell activation were maintained at similar abundances following chronic stimulation ( Figure 5C, Table S5). Out of the three proteins (TOX, LMCD1, and AFAP1L2) that were not induced upon transient activation, AFAP1L2 was strongly induced upon chronic stimulation ( Figure 5C). We estimated protein copy numbers in CD8 + T cells and found that transiently activated T cells contained close to zero AFAP1L2 protein copies (Figure 5D). However, re-activated T cells contained $100,000 copies of AFAP1L2 protein, and chronically stimulated cells had $300,000 copies ( Figure 5E). Consistent with this, we observed only background levels of AFAP1L2 mRNA in transiently activated CD8 + T cells, but the abundance of AFAP1L2 mRNAs was considerably higher in re-activated T cells and further increased upon chronic stimulation ( Figure 5F, Table S6).
Since AFAP1L2 is only expressed in T cells upon repeated stimulation, it could serve as a specific target to exclusively reinvigorate chronically stimulated T cells at the tumor site. We analyzed publicly available single-cell RNA-seq datasets of CD4 + and CD8 + T cells isolated from tumors of patients with HCC, 12  Article ll In all cancer types, AFAP1L2 was present in clusters of PD-1 + CD8 + T cells that express exhaustion markers such as LAG3, TIM-3, and CD39 ( Figure S4C). Notably, these exhaustion markers were also found in CD4 + Tregs, whereas AFAP1L2 was exclusively found in exhausted CD8 + T cells. Collectively, these data indicate that AFAP1L2 is induced in repeatedly stimulated CD8 + T cells across different cancer types. Since the function of AFAP1L2 in T cells was unknown, we next characterized its functional impact.

AFAP1L2 impairs T cell viability upon chronic stimulation
Upon chronic stimulation by tumor antigens, T cells undergo apoptosis, which plays a critical role in establishing tumoral immune resistance. 46 Since AFAP1L2 is a cytosolic signaling scaffold protein involved in the regulation of cell proliferation and survival, 37 we asked whether ablation of AFAP1L2 influences proliferation and viability of CD8 + T cells. We ablated AFAP1L2 by CRISPR-Cas9 in primary human CD8 + T cells using two different sgRNAs (editing efficiency of 77% and 59%, respectively, Figure S5). To analyze proliferation, we labeled T cells with CellTrace Violet (CTV), stimulated them for 5 days with antibodies to CD3 and CD28, and analyzed cells by flow cytometry. We did not observe any differences in proliferation ( Figure 5G). Next, we chronically activated T cells for 8 and 12 days and subsequently quantified the number of live CD8 + T cells by flow cytometry. Strikingly, upon chronic stimulation, the number of live AFAP1L2-edited CD8 + T cells doubled across both sgRNAs ( Figure 5H). Thus, AFAP1L2 does not impact proliferation but promotes cell death of chronically stimulated T cells and may therefore have a negative effect on tumor control.

Knockout of Afap1l2 in murine T cells improves antitumor functions
To study whether AFAP1L2 plays a role in the T cell response to tumors, we used CD8 + OT-I T cells, which have a transgenic TCR recognizing the ovalbumin (Ova)-derived SIINFEKL peptide presented on the MHC-I allele H-2Kb ( Figure 6A). Freshly isolated OT-I T cells were activated with plate-bound antibodies to CD3 and CD28, and after 24 h, they were electroporated with a plasmid encoding Cas9, GFP, and a sgRNA targeting Afap1l2. 47 After 24 h, GFP + OT-I T cells were sorted, and the gene editing efficiency was assessed by TIDE analysis. We used two different sgRNAs to disrupt the Afap1l2 gene (editing efficiency: 58% and 42%, Figure S6A). Upon chronic stimulation, the number of live Afap1l2-edited OT-I T cells increased 2-fold compared to nontargeting controls, phenocopying our observations in primary human T cells ( Figure S6B). Next, Afap1l2-edited OT-I T cells were re-stimulated for 48 h and then transferred into C57BL/6 mice with established B16.OVA tumors, which are recognized by OT-I T cells. 5 days later, we analyzed tumor infiltrates by flow cytometry and found that Afap1l2-edited OT-I T cells were more abundant in tumors than control OT-I T cells (18% vs. 10% of total CD8 + T cells in tumors) ( Figures 6B and 6C), indicating that Afap1l2-edited T cells mounted a more robust response. An intracellular cytokine staining of tumor-infiltrating T cells showed that the percentage of Afap1l2-edited OT-I T cells producing TNFa and IFN-g was twice as high as in controls ( Figures 6D and 6E). Taken together, these data indicate that Afap1l2 knockout increases both the quantity and potency of the anti-tumor T cell response.
When we followed B16.OVA tumor sizes over time, we found that Afap1l2-edited OT-I T cells mounted a superior anti-tumor response than control T cells. The effect was persistent across two different sgRNAs targeting Afap1l2 ( Figures 6F and 6G). We additionally treated mice with PD-L1 blocking antibodies and found that PD-L1 blockade combined with ablation of Afap1l2 in T cells synergistically reduced tumor growth and significantly increased the survival of mice ( Figures 6H and 6I).
To extend our analysis to a different tumor type, we used the MC38-OVA model. MC38-OVA tumors induce a strong endogenous T cell response and are completely eradicated upon transfer of OT-I T cells. Since this setting does not allow analyzing the impact of Afap1l2 ablation in T cells on tumor growth, we used Cd3e À/À mice 48 that lack all endogenous T cells. MC38-OVA tumor cells were subcutaneously injected into Cd3e À/À mice, and after 5 days, either control or Afap1l2-edited OT-I T cells were adoptively transferred. Under these conditions, Afap1l2-edited OT-I T cells had a superior anti-tumor activity than control OT-I T cells ( Figure 6J). In summary, we found that chronically stimulated CD8 + T cells induce AFAP1L2, which when ablated improves their anti-tumor functions.

DISCUSSION
While current immunotherapies have shown clinical activity in HCC patients, the majority of patients fail to respond. 17 Therefore, it is important to identify molecular mechanisms that contribute to immunosuppression in tumors. In this study, we generated a comprehensive proteomics dataset on cancer cells, macrophages, NK cells, and T cells from an HCC patient cohort. Our data captured known aspects of tumor immunology and provided new insights into the behavior of macrophages NK cells and T cells in tumors. We used mass spectrometry-based proteomics to profile fluorescence-activated cell sorting (FACS)-sorted immune cells isolated from blood, liver tissue, and tumors. During the isolation procedure, tissues are mechanically dissociated and enzymatically digested, which can lead to the degradation of mRNAs and to a lesser extent of proteins. While RNA degradation can lead to artifactual changes in transcriptomes, protein degradation is minimal and has little impact on proteomic measurements, with the exception of enzymatically digested membrane proteins. 49 While measuring transcripts is typically used as a proxy for protein abundance, MS allows the direct quantification of proteins, including those that are not synthesized in a cell but taken up from the surrounding environment.
Our proteome data showed that liver tumors largely alter the phenotype of macrophages. Similar observations were previously made in liver, endometrial, breast, and lung cancer by employing transcriptomics. [50][51][52] We found that tumor macrophages in HCC displayed the GPI-anchored protein THY-1 on their cell surface. Tumor macrophages did not express THY-1 mRNA but acquired the protein from the HCC tumor microenvironment. Interestingly, a study in a breast cancer model showed that THY-1 is upregulated in cancer stem cells (CSCs) and mediates physical interactions with tumor macrophages. This interaction allows tumor macrophages to create a CSC niche. 53 THY-1 is also expressed on activated endothelial cells, where it mediates the adhesion of monocytes via its counter receptor CD11B (MAC1). 54 It is possible that during these interactions, macrophages extract THY-1 from the membrane of endothelial cells or CSCs, thus shaping the macrophage phenotype within the TME.
In TAMs isolated from liver tumors, SGPL1 protein was strongly enriched. Inspection of published single-cell RNA-seq data of macrophages isolated from HCC tumors 52 showed that SGPL1 mRNA was present in tumor macrophages but was not as strongly enriched as in our proteome dataset. SGPL1 degrades the signaling sphingolipid S1P in the membrane of the endoplasmic reticulum and is crucial for S1P homeostasis. S1P can be exported from cells, where it interacts with S1P receptors (S1PRs). S1PRs mediate a broad range of cellular functions, including trafficking of lymphocytes. For example, the egress of lymphocytes from lymphoid organs is dependent on S1P receptor 1 (S1PR1). Lymphocytes expressing S1PR1 migrate toward S1P, whose concentration is low in lymphoid organs but high in the blood and lymphatics. 55 In Sgpl1 À/À mice, S1P concentrations in lymphoid organs are increased, which inhibits the egress of lymphocytes from lymphoid organs, causing a decrease in the number of circulating lymphocytes. 56 Extracellular and intracellular S1P can activate the NF-kB pathway, which inhibits apoptosis in myeloid cells and enhances cytokine production. 57 For example, Sgpl1 À/À mice exhibit higher concentrations of inflammatory cytokines in their sera in response to an LPS challenge. 58 In line with an inflammatory phenotype of Sgpl1 À/À mice, we found that BMDMs in which we ablated Sgpl1 produced more of the inflammatory cytokines IL-6 and IL-12. The resultant inflammatory phenotype may be a consequence of accumulating S1P, which either acts on S1PRs or through intracellular signaling pathways. 59 Strikingly, Sgpl1-deficient BMDMs had improved anti-tumor activity, suggesting that SGPL1 and S1P metabolism are potential therapeutic targets to improve antitumor immunity. Systemic targeting of SGPL1 may likely cause side effects, given that Sgpl1 À/À mice show immunological alterations and have a reduced lifespan. 58 As such, a safer approach might be to target SGPL1 specifically in tumor macrophages, for example, by using macrophage-targeting nanoparticles. 60 Our analyses of T cell proteomes in HCC revealed that AFAP1L2 was uniquely expressed in CD8 TILs in a subset of patients with high percentages of PD1 + TILs, indicative of highly stimulated and proliferative T cells within the TME. These cells also expressed several inhibitory receptors and CD39, and therefore most likely recognize tumor antigens. 13,61,62 Mechanistically, we validated that AFAP1L2 is induced in CD8 + T cells only upon repeated triggering of the TCR. Genetic ablation of Afap1l2 in murine CD8 + T cells improved their survival and anti-tumor activity and had synergistic effects with PD-L1 blocking antibodies in the clearance of tumors. This suggests that AFAP1L2 is a potential target for T cell-based cancer immunotherapies to treat patients with HCC and other types of tumors.
Since AFAP1L2 reduces T cell survival and effector functions, it operates as a checkpoint to blunt the escalation of immune responses. As AFAP1L2 is expressed only in chronically stimulated T cells, its ablation does not interfere with T cell activation, and thus it is a promising target specific to dysfunctional T cells. Being an intracellular scaffold protein, AFAP1L2 is not accessible to blocking antibodies. Therefore, AFAP1L2 must be targeted through different strategies. For example, it could be targeted systemically with small molecule degraders, which harness the ubiquitin proteasome system to selectively target intracellular proteins. 63 Since adult Afap1l2 À/À mice do not show any significant anatomical and physiological phenotypes, 64 it is likely that systemic degradation of AFAP1L2 would have limited toxicity.
An additional therapeutic application of our findings would be to genetically ablate AFAP1L2 in CAR T cells to improve Article ll OPEN ACCESS their anti-tumor function. Such interventions are crucial for nextgeneration CAR T cells, because T cell dysfunction is a major contributor to ineffective CAR T cell therapy in solid tumors. 65,66 CAR T cell dysfunction has recently been associated with a transition toward an NK-like phenotype. 67 Consistent with this notion, we found that AFAP1L2 is only expressed in chronically stimulated T cells and NK cells. Whether AFAP1L2 has similar functions in NK cells remains to be determined. AFAP1L2 is a scaffold protein that can bind to c-Src-containing proteins and PI3K through its proline-rich domain, and it can bind to lipids within membranes through its two pleckstrin homology domains. 37 The 818-amino-acid-long protein also contains a coiled-coil domain with unknown function. In general, scaffold proteins dynamically interact with different partners and shape cell behavior. 68 Notably, the functions of AFAP1L2 are cell type specific since it curtails the survival of chronically stimulated T cells but has the opposite effect in cancer cells. Previous reports showed that silencing of AFAP1L2 in thyroid and lung cancer cells inhibits cell-cycle progression and survival. 69,70 Consistent with these observations, we found that ablation of AFAP1L2 in renal (CAKI1) and pancreatic (CAPAN-2) cancer cell lines inhibited their growth (data not shown). Thus, systemic targeting of AFAP1L2 would potentially enhance the activity of chronically stimulated T cells and directly inhibit the growth of cancer cells. How AFAP1L2 controls survival and growth in different cell types remains to be determined.
In summary, our study provides a rich dataset on cancer and immune cell proteomes across a relatively large cohort of patients with HCC. Our findings uncovered new mechanisms underlying T cell dysfunction and mechanisms by which macrophages impede tumor control. These insights are critical for advancing cancer immunotherapy.

Limitations of the study
In this study, surgically resected liver and tumor tissue was entirely used for immune cell extraction, and no histopathological images were taken to exclude the possibility of microtumors in samples designated as non-tumorous. To confirm correct classification of tumorous and non-tumorous tissue, we relied on the analysis of their proteomes, which showed that they had a clearly distinct profile. However, since we did not assess the ratio of cancer cells to tumor stroma in patient samples, some differences between tumorous and non-tumorous tissue might be masked.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

OPEN ACCESS
Data and code availability d The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE 1 partner repository with the dataset identifier PXD040957. d The raw count RNA-seq data have been deposited at GEO and are publicly available as of the date of publication. Accession numbers are GSE229400 for tumor macrophages and GSE228571 for T cells. d Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Human specimens
After obtaining written informed consent, peripheral blood, liver and tumor tissue was obtained from HCC patients undergoing liver resection at the Ospedale Sant'Orsola (Bologna), Universitä tsklinik (Mannheim) and Universitä tsklinikum Carl Gustav Carus (Dresden).

Patient samples
Fresh tumor and adjacent liver tissue samples were cut into approximately 1 mm 3 pieces in RPMI-1640 medium (Invitrogen). Samples were enzymatically digested using the MACS human Tumor Dissociation kit (Miltenyi Biotec) in a gentleMACS TM Dissociator (Miltenyi Biotec) according to manufacturer's instructions (protocol for a sample size of 0.2-1.0 g, h_tumor_01 program). Cell suspensions were filtered through 70 mm Cell-Strainers (BD) in RPMI-1640 medium and then centrifuged at 500 g for 5 min. Cell pellets were resuspended in 25 mL RPMI-1640 medium. Mononuclear immune cells were separated from tumor and stromal cells by Ficoll gradient centrifugation (15 mL of Ficoll-Histopaque solution; 18 C, 30 min, 774 g). Mononuclear cells and tumor/stroma pellets were washed twice with RPMI-1640. Tumor/stroma cell pellets were washed three times with PBS, snap-frozen in liquid nitrogen and kept at -80 C until processing. Mononuclear immune cells were resuspended in 500 mL MACS buffer (PBS with 2% FBS and 5 mM EDTA), stained and sorted as indicated in Figure S1A. Monocytes/macrophages, and memory T cells were sorted using a FACS ARIA III (BD) as indicated in Figure S1A and recovered in RPMI-1640 supplemented with 1X Glutamax, Penicilin/Streptomycin, Non-essential amino acids and Hepes (GIBCO). Then, cells were washed and resuspended in cold PBS three times before pellets were snap frozen in liquid nitrogen and kept at -80 C.
Peripheral blood was obtained from patients prior to surgery in EDTA tubes. Then, immune cells were obtained by Ficoll gradient centrifugation (5 mL of blood on top of 3 mL of a Ficoll-Histopaque solution). Mononuclear cells were washed and stained for sorting as shown in Figure S1A.

Mice
Wild type C57BL/6 mice were obtained from Charles River (Italy). CD3e -/mice were kindly provided by Dr. Bernard Malissen (CIML, Marseille) and have been described previously. 48 Rag-/-HZ OT-1 mice were obtained by crossing three different strains obtained from Jackson Laboratory: B6.129S7-Rag1tm1Mom/J, C57BL/6 Tg(TcraTcrb)1100Mjb/J and B6.SJL-Ptprca Pepcb/BoyJ. Rosa-Cas9 mice (B6(C)-Gt(ROSA)26Sorem1.1(CAG-cas9*,-EGFP)Rsky/J) were from Jackson Laboratory. Mice were maintained under specific pathogen-free conditions in the animal facility of the Institute for Research in Biomedicine (IRB). Five mice per cage were housed in ventilated cages under standardized conditions (20±2ºC, 55±8% relative humidity, 12 h light/dark cycle). Food and water were available ad libitum, and mice were examined daily. Female mice were used between 6 and 10 weeks of age. Mice were treated in accordance with the Ticino Cantonal Commission for Animal Welfare, which is in accordance with the Animal Welfare Ordinance and the Animal Experimentation Ordinance from the Swiss Animal Welfare Legislation. Tumor sizes of 1000 mm 3 was considered the limit to terminate experiments (cantonal authorization number TI 51/2019).

METHOD DETAILS
High resolution mass spectrometry For tumor/liver stroma, 2-5 mm3 of each pellet were homogenized in 4% SDS in 100 mM Tris pH 7.6 for 10 min at 300 oscillations/ minute in a TissueLyser II (Qiagen). Samples were further sonicated in a Bioruptor (Diagenode) (15 cycles, 30s on, 30s off, high mode) and incubated at 95 C for 10 min, after which the lysates were cleared by centrifugation. Proteins were precipitated overnight at À20 C in 80% acetone (VWR), pelleted by centrifugation at 13,000 rpm for 20 min at 4 C and dried on a heating block at 40 C. The protein pellets were re-suspended in 50 mM ammonium bicarbonate buffer (ABC) at pH 8 containing 8M urea (Sigma). For sorted immune cells, pellets were directly re-suspended in 50 mM ABC 8M urea and lysed by Bioruptor sonication. Disulfide bonds were reduced with 10mM DTT (Sigma) and subsequently alkylated with 50 mM iodoacetamide (Sigma). Samples were pre-digested for 2 hours with LysC (Wako Fujifilm, 1:100, w/w) and then diluted 1:4 with 50 mM ABC before trypsin (Promega, 1:100, w/w) was added and the mixtures were incubated overnight at RT. The resulting peptide mixtures were acidified and loaded on C18 StageTips 74 . Peptides were eluted with 80% acetonitrile (ACN), dried using a SpeedVac, and resuspended in 2% ACN, 0.1% trifluoroacetic acid and 0.5% acetic acid for single-shot MS measurement.