RALDH1 Inhibition Shows Immunotherapeutic Efficacy in Hepatocellular Carcinoma Free

Hepatocellular carcinoma (HCC) is a devastating disease with a projected annual incidence of around one million cases worldwide by 2025 (1). Around 27,000 people die from this disease each year in the United States alone and advanced HCC has an abysmal 5-year survival rate of about 2% (2, 3). Surgical resection and liver transplantation are preferred treatments for early-stage HCC, whereas locoregional interventions such as radiofrequency ablation and transarterial chemoembolization are used in unresectable cases (3). Advanced metastatic cases present substantial management challenges with a median survival of a few months. Recent progress in systemic therapies, which currently include immune checkpoint blockers (ICB), tyrosine kinase inhibitors, and angiogenesis inhibitors, has improved patient outcomes. Nonetheless, there is substantial room for improvement, especially with immunotherapy. Single-agent ICB elicits clinical responses in a minority of patients, suggesting the existence of other biological modulators of ICB responses (4).

Macrophages and dendritic cells (DC) are the key antigen-presenting cells in solid tumors and, given the current limitations of immunotherapy, there has been increasing interest in therapeutically targeting them (5). These efforts are generally aimed at reducing the frequency of immunosuppressive macrophages, increasing the frequency of tumoricidal and pro-inflammatory macrophages, and enhancing the immunostimulatory activities of DCs (5). A number of approaches have been described previously that can achieve these effects in experimental models by targeting specific receptors and/or pathways in tumor-associated macrophage (TAM) and DCs (6, 7). In contrast, we know very little about whether and how we can target monocyte differentiation into DCs versus TAMs for cancer immunotherapy. DCs and TAMs can also originate from nonmonocyte precursors—embryonic progenitors arising from yolk sac and HSC-derived myeloid progenitors, respectively—but these progenitors are exceeding rare compared with abundant circulating monocytes (8, 9). Furthermore, DCs and TAMs within the tumor microenvironment (TME) have a finite life-span requiring a continuous influx of progenitors (10). Thus, targeting monocyte differentiation represents a viable but largely unexplored therapeutic strategy in cancer immunotherapy.

We previously discovered that some tumors produce retinoic acid (RA) that promotes differentiation of monocytes into immunosuppressive and tumor-promoting macrophages (11). Therefore, reducing RA production by tumor cells or inhibiting RA signaling in monocytes is a potential treatment approach in these tumors. Key barriers to implementation of this approach include identifying the tumors where this pathway is active and developing safe and effective inhibitors of the RA pathway. Here, we report that HCCs produce high levels of RA through overexpression of retinaldehyde dehydrogenase 1 (RALDH1), which is one of the three enzymes that catalyzes RA production (12). RA production in HCC was abrogated by RALDH1 inhibitors (Raldh1-INH) we recently developed (13). These inhibitors suppressed tumor growth in multiple mouse models of human and murine HCC. Using genetic and pharmacologic tools, we demonstrate that the HCC-suppressive effects of Raldh1-INH are driven by altered macrophage numbers and function as well as increased infiltration of tumors by activated T cells. Pharmacological and toxicological analyses revealed a favorable profile of Raldh1-INH for potential clinical use, which was also supported by observations in newly generated RALDH1 knockout (RALDH1-KO) mice. These findings provide proof of concept for the use of Raldh1-INH in HCC and establish the scientific premise for the development of isozyme-specific RALDH inhibitors as a new strategy in cancer immunotherapy.

Key reagents are listed in Supplementary Table S1.

The University of Pennsylvania Institutional Animal Care and Use Committee (IACUC) approved all mouse experiments performed at the University of Pennsylvania. Pharmacokinetics (PK) studies were performed at the NIH and Pharmacon. Animal experiments performed at the NIH were conducted in compliance with institutional (NIH) guidelines. Animal procedures performed at Pharmakon were carried out according to guidelines approved by the IACUC of Pharmaron following the guidance of the Association for Assessment and Accreditation of Laboratory Animal Care.

Wild-type (WT) C57BL/6 mice were purchased from The Jackson Laboratory (cat. no. 000664). Immuodeficient NU/J mice were purchased from The Jackson Laboratory (cat. no. 002019). dnRAR^flox mice were a generous gift from Dr. Cantorna Margherita at Pennsylvania State University. LysM^Cre mice were purchased from The Jackson Laboratory (cat. no. 004781).

Aldh1A1 knockout mice were created through the University of Pennsylvania's CRISPR/Cas9 Mouse targeting core facility. Two CRISPR RNAs (sequences below) were designed that encompassed an approximately 36 kilobase region within the mouse Aldh1A1 gene (Gencode Gene: ENSMUSG00000053279.8. Position: mm10 chr19:20,492,715–20,643,465).

Aldh1a1_5p_crRNA: CTGAGTTGGACCCTATATGG

Aldh1a1_3p_crRNA: GAGAATGTGTTGGTGCCTCG

A mix of pure Cas9 mRNA and the guide RNAs were injected into single-cell zygotes of C57BL/6 background mice. Founders were identified by a PCR-based genotyping protocol (details below) designed to detect the gene deletion. Founders were then bred to WT C57BL/6 mice to “fix” the allele. Heterozygous pups were identified by the aforementioned PCR-based genotyping and bred to each other to generate RALDH1-KO mice. Additional qRT-PCR assays were performed (described in “RNA isolation and qPCR analysis for gene expression”) to confirm absence of the Aldh1a1 transcripts.

WT_9753: CAA CCC TGA GCA AAT CCT CCA C

WT_ 9754: GAC AGA TTG AGA GCA GTG TTT ACC C

Aldh1a1-KO_F1: TGA TAT GTC CCA GGA AGA TGA A

Aldh1a1-KO_R2: GGA CCG AGC ACT TGC CTA

5 minutes at 94°C. 35 cycles of: (i) 30 seconds at 94°C, (ii) 30 seconds at 58°C, and (iii) 30 seconds at 72°C. Final extension of 7 minutes at 72°C followed by storage at 4°C. 248 base pairs band in knockouts and 665 base pairs band in WT detected by conventional gel electrophoresis.

The details of the cell lines and culture media are described below. Upon receipt, cell lines were first expanded (two passages), authenticated (described below), and then frozen into aliquots for storage. The frozen stocks were thawed and expanded (average three passages) before experiments and discarded upon completion of individual experiments. When needed, the frozen stocks were reexpanded (average of two passages), authenticated, and stored as frozen aliquots of additional stock.

The Huh1 cell line was a generous gift from Dr. Junwei Shi at the University of Pennsylvania. The cells were obtained in 2022 and authenticated on the basis of their morphology and growth characteristics in cell culture as well as the histology of the tumors formed upon transplantation into mice.

Huh7, SNU449, SNU398, HEP3B, PLC, HEPA 1–6, HEP55, and AL458A were from Dr. Celeste Simon at the University of Pennsylvania, who is also a coauthor on this article. The cells were obtained in 2021 and authenticated on the basis of their morphology and growth characteristics in cell culture as well as the histology of the tumors formed upon transplantation into mice.

Fibrosarcoma cell lines were obtained from Dr. Robert Schreiber at Washington University (St Louis, MO), and use in our laboratory has been described in a previous publication (11).

Tumor cell lines were cultured in DMEM (Thermo Fisher Scientific, cat. no. 10567014) with 10% FBS (GeminiBio, cat. no. 100–500) 1% Pen/Strep (Thermo Fisher Scientific, cat. no. 15140122), and 2 mmol/L glutamine (Thermo Fisher Scientific, 25030081). All cells were confirmed to be negative for Mycoplasma contamination as assessed by the MycoAlert Mycoplasma Detection Kit (Lonza, cat. no. LT07).

A pathologist identified human formalin-fixed, paraffin-embedded (FFPE) samples of normal liver, normal kidney, HCC, gastrointestinal stromal tumors (GIST), and colorectal cancer from patients (deidentified) who underwent surgical resection for diagnostic or therapeutic purposes. 56 archived tissue blocks were selected and deidentified before sectioning and IHC. Approval from the Institutional Review Board at the University of Pennsylvania (Protocol 851744) was obtained before initiating the IHC study. As de-identified archived tissue blocks were used, specific written consent from patients was not required.

Normal donor human monocytes and T cells were collected by the Human Immunology Core (HIC) at the University of Pennsylvania and purchased by our laboratory from HIC. Written informed consents were obtained from donors by the HIC.

Cultured Huh7, Huh1, Hepa 1–6, Hep55, and fibrosarcoma (as indicated in figure legends) tumor cells were detached using 0.25% trypsin (Gibco, cat. no. 25200056), washed once with 1x PBS, and counted before implantation. A total of 3∼6*10⁶ tumor cells were implanted subcutaneously into shaved flanks of recipient mice. Tumor dimensions were measured using a caliper starting at the day indicated in the accompanying figure legend and every two to 3 days thereafter; volume was calculated by using formula Length*Width²/2. Tumor volumes of 2,000 mm³, tumor length of 2 cm or tumor ulceration were used as endpoints for survival analyses.

Tissue samples (murine tumors of the type indicated in the corresponding figure legends) were harvested and single-cell suspensions were generated by digestion with collagenase B and Dnase I (both Sigma-Aldrich) for 45 minutes at 37°C and filtration through 70 μmol/L cell strainers. Red blood cells were lysed using RBC Lysis Buffer (BioLegend). Samples were incubated for 20 minutes on ice with anti-mouse CD16/32 Fc Block (BD Biosciences), and subsequently stained on ice with primary-fluorophore conjugated antibodies for identification of cell populations by flow cytometry. Flow cytometry was performed on an LSRII Flow Cytometer (BD Biosciences) and analyzed using FlowJo software (Treestar, version 10.8.1). A list of the antibodies used can be found in Supplementary Table S1.

A total of 1* 10⁴ ∼ 2*10⁵ of Huh7, Huh1, Hepa 1–6, and Hep55 (as indicated in figure legends) tumor cells were plated in triplicate in 48-well plates or 6-well plates. Viable cell numbers were counted every day for three to 4 days.

Cultures of tumor cells or primary monocyte-derived cells were treated with C86, C91, C99, BMS 493 (Tocris, 3509) or Win 18446 (Tocris, 4736) at a time and concentration indicated in corresponding figure legends.

The AldeRed assay (EMD Millipore) was performed according to the manufacturer's instructions to identify cells with ALDH activity. In brief, single-cell suspensions of Huh7, Huh1, SNU449, SNU398, HEP3B, PLC, Hepa 1–6, Hep55, and AL458A cultured cells or single-cell suspension generated from enzymatic digestion of tumors generated from transplantation of these cells in mice (as indicated in figure legends) were incubated with a fluorescent and nontoxic ALDH substrate (AldeRed 588-A); the fluorescent product accumulates in cells proportional to their ALDH activity. The amount of fluorescence produced is measured by flow cytometry. The ALDH inhibitor diethylaminobenzaldehyde (DEAB, provided with the AldeRed assay kit) was used as a negative control for background fluorescence assessment.

GFP-expressing Huh7 or Hep55 cells from CRISPR Knockout experiments were isolated using the FACS Jazz cell sorter at the Children's Hospital of Philadelphia Flow Cytometry Core Laboratory. Cells were identified and isolated on the basis of the GFP positivity alone on the cell sorter.

For measurement of all-trans RA (ATRA), cultured cells were detached using trypsin, centrifuged, and the cell pellets stored at −80°C. ATRA was extracted from the frozen cell pellets and quantified at the Children's Hospital of Philadelphia Metabolomics Core using LC/MS as described previously (11).

Monocytes were isolated from bone marrow (BM) of C57BL/6 mice using the Mouse BM Monocyte Isolation Kit (Miltenyi Biotec) according to the manufacturer's instructions. Purity of monocyte was assessed by flow cytometry using CD11b, Ly6C, and Ly6G.

Monocytes were isolated from mouse BM of C57BL/6 mice and counted. Subsequently, a total of 1*10⁶ monocytes were resuspended in 50 μL 1x PBS and injected directly into established tumors at day 12 after tumor-cell implantation. Tumors were harvested at specified time points and analyzed by flow cytometry. For human monocytes, a total of 1*10⁶ cells were injected intratumorally as described above. This method of intratumoral monocyte transfer is described in detail in our previous publication (14)

Mouse monocytes were isolated from BM as described above and then cultured in RPMI-1640 (Thermo Fisher Scientific, cat. no. 11875085) with 10% FBS (GeminiBio, cat. no. 100–500). GM-CSF (20 ng/mL, peprotech 315–03) and IL4 (20 ng/mL, peprotech 214–14) were added for DC differention, whereas M-CSF (20 ng/mL, peprotech 315–02) was added for macrophage differentiation. Human monocytes were cultured in RPMI-1640 (Thermo Fisher Scientific, cat. no. 11875085) with 10% FBS (GeminiBio, cat. no. 100–500). GM-CSF (50 ng/mL, peprotech 300–03) and IL4 (50 ng/mL, peprotech 200–04) were added to cultures for DC differentiation, whereas or M-CSF (50 ng/mL, peprotech 300–25) was added for macrophage differentiation. RA (200 nmol/L; Sigma-Aldrich), C86 (100 nmol/L) or tumor conditioned medium was added at specified time points for indicated differentiation assays. Cellular identity and function of differentiated monocytes was assessed by flow cytometry and quantitative PCR (qPCR).

To deplete T cells, 200 μg of mouse CD3-specific antibody (clone 17A2) was administered intraperitoneally starting 3 days before tumor implantation and repeated every 3 to 4 days until mouse sacrifice.

To deplete macrophages, 200 μL of clodronate liposome (CloLipo) or PBS-liposome (CtrlLipo; both Liposoma) was administered intraperitoneally starting 3 days before tumor implantation and repeated every 4 days until mouse sacrifice. Macrophage depletion efficacy in spleen and within tumors was confirmed by flow cytometry using the canonical macrophage marker F4/80.

Compound-86, 97 and 99 powder was dissolved in 20% HPβCD saline (2-Hydroxypropyl)-β-cyclodextrin). Drug was administered intraperitoneally or p.o. starting when tumor volume reached 50 to 150 mm³ and repeated every day. These compounds were described in our previous publication and produced at NCATs using methods outlined in that article (13).

200 μg of PD1-specific monoclonal blocking antibody (clone RMP 1–14) was administered intraperitoneally starting when tumor volume reached 50 to 150 mm³ and repeated every 2 days.

The PK studies were done at NIH or a commercial CRO, Pharmaron, using their in-house standard protocol.

• 1. For C91, IV and PO, performed at NIH using CD-1 mice; formulation: 20% HPbCD in saline

• 2. For C86, IV and PO, performed at Pharmaron using CD-1 mice; formulation: 20% HPbCD in saline

• 3. For C99, IV and PO, performed at NIH using CD-1 mice; formulation: 20% HPbCD in saline

• 4. IP PK of C86 at 10 and 30 mpk, performed at NIH using C57BL/6J mice; formulation: 60% PEG400 in DI water

• 5. Chow PK of C86, performed at Pharmaron using CD-1 mice at 10, 30, and 60 mpk, which the dose concentration in chow is 0.05, 0.15, 0.3 mg/g, respectively, based on the calculation of 5 g food consumption/mouse/d.

PK studies at the NIH were performed by the DMPK group. Male CD1 or C57BL/6J mice between 6 and 8 weeks old and weighing approximately 20 to 30 g were dosed with compound 86, compound 91, and compound 99 at 2 mg/kg (IV), 10 mg/kg (PO), and 10 or 30 mg/kg (IP). The compounds were formulated using a 20% hydroxypropyl-beta-cyclodextrin (HPbCD) solution in saline, was made on the day of dosing or directly before dosing. Each treatment group consisted of three mice, and plasma were collected at 5, 15, 30 minutes, and 1, 2, 4, 8, 12, and 24 hours after dose for intravenous administration and at 15, 30 minutes, and 1, 2, 4, 8, 12, and 24 hours after dose for PO administration. Approximately 0.025 mL of blood was collected from the dorsal metatarsal vein at each time point. The collected blood samples were then transferred into plastic microcentrifuge tubes containing heparin sodium as an anticoagulant. Samples were then centrifuged at 4,000 × g for 5 minutes at 4°C to obtain plasma. Plasma samples were then stored in polypropylene tubes, quickly frozen, and kept at −75°C until analyzed by LC-MS/MS. The following PK parameters were measured: Terminal half-life (T1/2), concentration at immediately after injection (C0), maximum concentration (C_max), time to reach max concentration (T_max), Clearance (CL), Volume of distribution (V_d), AUClast, and bioavailability (% F). Animals were also monitored during the in-life phase by once daily cage side observations; any adverse clinical signs were noted as part of the PK report.

Serum toxicological assays were performed by IDEXX Bioanalytics (Standard Tox Panel 62794). Briefly, peripheral blood was collected in regular eppendorff tubes by tail snips following a protocol approved by our institutional IACUC. Serum was prepared by letting the blood coagulate and inspected to confirm absence of hemolysis. The serum samples were stored in −80°C until shipment to IDEXX.

Hematological studies, including complete blood count, were performed at IDEXX Bioanalytics. Briefly, peripheral blood was collected in heparin coated microhematocrit capillary tubes (VWR, 15401–560) by tail snips following a protocol approved by our institutional IACUC and stored at 4 degrees overnight before shipment to IDEXX.

Total RNA from mouse tissue samples and in vitro cultured cells was isolated using GenElute Mammalian Total RNA Miniprep Kit (Sigma) following the manufacturer's protocol. Total RNA from human FFPE samples was isolated by using the Quick-RNA FFPE Miniprep Kit (Zymo Research) using the manufacturer's protocol. 1,000 ng of RNA was used for reverse transcription using High Capacity RNA to cDNA Kit (Life Technologies).The cDNA product was 10× diluted. 2.5 μL of this cDNA was used for qPCR for each sample. Three or more replicates were used for each reaction. Target gene expression was normalized to appropriate housekeeping gene indicated in the legends of figures showing RT-qPCR data. Expression fold was calculated as 2^-(C_t target gene –C_t housekeeping gene). Quantitative PCRs were run on a Viia 7 real-time PCR system (Thermo Fisher Scientific). The following target and housekeeping genes were measured using the commercially available TaqMan probes (Thermo Fisher Scientific):

Mouse Hprt (Mm03024075_m1), Mouse Irf4 (Mm00516431_m1), Mouse Zbtb46 (Mm00511327_m1), Mouse Mafb (Mm00627481_s1), Mouse Aldh1a1 (Mm00657317_m1), Mouse Aldh1a2 (Mm00501306_m1), Mouse Aldh1a3 (Mm00474049_m1), Human HPRT (4333768F), Human IRF4 (Hs01056533_m1), Human ZBTB46 (Hs01008168_m1), Human MAFB (Hs00271378_s1), Human ALDH1A1 (Hs00946916_m1), Human ALDH1A2 (Hs00180254_m1), and Human ALDH1A3 (Hs00167476_m1.

A total of 1 × 10⁷ cells were harvested and washed with PBS three times before 50 μL RIPA Lysis Buffer (Thermo Fisher Scientific, cat. no. 89901) containing proteinase inhibitor (Thermo Fisher Scientific, cat. no. 78442) was added to the cell pellets and thoroughly mixed. Cell lysates were kept on ice for 30 minutes and centrifuged at 14,000 × g for 15 minutes. The supernatant was collected, and the protein concentration measured using the BCA Protein Assay kit (Thermo Fisher Scientific, 23227). Next, 30-μg cell lysate for each sample was mixed with loading buffer (Thermo Fisher Scientific, cat. no. NP0007) to make a final volume of 20 μL and incubated at 95°C for 5 minutes. Samples were loaded on a precast 4%–15% SDS polyacrylamide gel (cat. no. 4561084, BIO-RAD), and run at 120V (constant voltage) for 40 to 60 minutes until the dye reached the bottom of the gel. Samples were transferred from the gel to polyvinylidene difluoride (PVDF) membrane in Tris-Glycine transfer buffer at 100 V for 1.5 hour at constant current (not to exceed 0.4 A). PVDF membranes were taken out from the blotting cassette and rinsed with TBST (10 mmol/L Tris-HCl, pH 8.0, 150 mmol/L NaCl, 0.05% Tween 20) for 5 minutes at room temperature (RT) 3 times. Nonspecific binding on the membrane was blocked with freshly prepared 5% nonfat dried milk (Labscientific, cat. no. M0841) for 1 hour on a shaker at RT for 1 hour. Aldh1a1-specific rabbit polyclone primary antibody (Invitrogen, cat. no. PA5–32127) and GAPDH-specific (14C10) rabbit mAb (Cell Signaling Technology, cat. no. 2118S) were diluted with 5% BSA at the ratio of 1:1,000 and incubated with the PVDF membrane at 4°C overnight. The membranes were washed three times for 5 minutes each with TBST, and incubated with horseradish peroxidase–conjugated secondary antibody (Cell Signaling Technology, cat. no. 7074) for 1 hour at RT. The membranes were washed three times for 5 minutes each with TBST, and incubated with ECL substrate (PerkinElmer, cat. no. NEL104001EA) for 1 minute before imaging using the ChemiDoc imaging system (Bio-Rad).

Non-viral delivery of Cas9-RNPs was previously described (15). In brief, crRNA and tracrRNA (both Integrated DNA Technologies) were mixed at equimolar concentrations and were annealed by heating at 95°C for 5 minutes followed directly by hybridization for 15 minutes at RT. The annealed crRNA/XT-tracrRNA duplexes were mixed with Cas9 at a 3:1 molar ratio and were complexed by incubation at RT for ≥20 minutes. Nucleofection of Cas9-RNPs along with a GFP-expressing plasmid vector (supplied as part of the kit and used to identify cells undergoing successful nucleofection) was performed using Nucleofector 2 (Lonza). For the Huh7-RALDH1 knockout cell line, GFP-positive cells were sorted by FACS Jazz and loss of RALDH1 expression was confirmed by western blot. For the Hep55-RALDH1 knockout cell line, the GFP⁺ cells were sorted by FACS Jazz and single-cell clones were established from sorted cells. Loss of ALDH1A1 was confirmed in individual clones with Sanger sequencing performed at the University of Pennsylvania core facility and western blots performed in-house.

To compare RALDH isozyme expressions in HCCs versus other human tumors in the The Cancer Genome Atlas (TCGA) dataset, we used the cBioPortal website interface for gene expression query. To examine RALDH isozyme expression between different HCC molecular subytpes, we downloaded raw sequence counts for 371 primary tumor samples in The Cancer Genome Atlas (TCGA)-LIHC from the Genomic Data Commons Data Portal, and filtered them, retaining the 183 samples that were included in iClusters 1–3 as described in a previous publication (16). On a local workstation, several Bioconductor packages in R were used for subsequent steps. The count data were annotated with biomaRt. Principal component analysis (PCA) and plots were generated with PCAtools. Normalizations and statistical analyses were done with DESeq2. Exploratory GSEA pathway analysis was done with fgsea against the hallmark pathway set from the Molecular Signatures Database (MsigDB), using the DESeq2 statistic as a ranking metric. Clustering was performed with the degPatterns function from the DEGreport package.

To examine the expression of RALDH isozymes in single-cell RNA sequencing (scRNA-seq) data of human HCC, we used a previously published and publicly available dataset (17). The authors created an atlas of this published scRNA-seq that can be queried using their webportal—http://omic.tech/scrna-hcc. ALDH1A1, ALDH1A2, and ALDH1A3 were used as gene query terms using all samples in the dataset for the scatter plot output.

To build a homology model we used MOE (Molecular Operating Environment) software with default setting. The software MOE is a suite of different software tools developed by Chemical Computing Group Inc. The homology modeling algorithm within MOE comprises of the following steps. First, Initial Partial Geometry Specification: An initial partial geometry for each target sequence is copied from regions of one or more template chains. Where residue identity is conserved between the target sequence and its template, all heavy-atom coordinates are copied; otherwise, only backbone coordinates are copied. Second, Insertions and Deletions for correction of no assigned backbone coordinates. They are modeled from fragments of high-resolution chains from the Protein Data Bank that superpose well onto anchor residues on either side of the insertion area. Third, Loop Selection and Sidechain Packing: After the indel data collection is complete, a set of independent models is created. Loops are modeled first, in random order. For each loop, a contact energy function analyzes the list of candidates collected in the segment searching stage, taking into account all atoms already modeled and any atoms specified by the user as belonging to the model environment (e.g., a ligand bound to the template, or structural waters). These energies are then used to make a Boltzmann-weighted choice from the candidates, the coordinates of which are then copied to the model. Once all of the loops have been chosen, the side chains are modeled. Sidechain data are assembled from an extensive rotamer library generated by systematic clustering of conformations from rotamer library. A deterministic procedure based on Unary Quadratic Optimization is then run to select an optimal packing. After all of the backbone segment and side-chain conformations have been chosen for an intermediate model, hydrogens are added to complete valence requirements and the model is submitted to a series of minimizations designed to first relieve any serious steric strains, and then to prepare the model to be scored. It is then written to the output database, along with a number of quality assessment measurements that can flag any serious geometric problems. The fourth stage is Final Model Selection and Refinement. The final model is based on the best-scoring intermediate model. The final model is based on the best-scoring intermediate model. In this study, we used the electrostatic solvation energy, calculated used a Generalized Born/Volume Integral methodology. After the homology modeling procedure has finished, the final model was inspected using MOE's Protein Geometry stereochemical quality evaluation tools, including Ramachandran Maps.

Huh 7 xenograft tumor slides were dewaxed with citrisolv for 20 minutes, and 100%, 95%, 85%, 75% ethanol and ddH₂O for 3 minutes for each. Peroxidase was blocked with 3% H₂O₂ for 10 minutes at RT. Slides were then washed thrice with PBS and target retrieval for 15 min under high temperature. Slides were then blocked with Avidin/Biotin blocking kit (Vector laboratories) following the manufacturer's instructions. Rabbit anti-mouse CD163 (Abcam, Ab6720) was diluted 1:200 and incubated with the slides at 4°C overnight. Anti-rabbit secondary (Abcam) was diluted 1:200 and incubated with the slides at RT for 1 hour. The signal was amplified with the VECSTAIN ABC kit (Vector laboratories) and stained with the DAB substrate kit (Vector laboratories) following the manufacturer's instructions. The stained slides were scanned using a Leica Aperio Slide Scanner and analyzed with QuPath as described previously (18).

Five-μm sections of FFPE tissue were stained using and antibody against ALDH1A1 (D9Q8E, Cell Signaling Technology 54135S, 1:400). Staining was done on a Leica Bond-IIITM instrument using the Bond Polymer Refine Detection System (Leica Microsystems DS9800). Heat-induced epitope retrieval was done for 20 minutes with ER1 solution (Leica Microsystems AR9961). All the experiment was done at RT. Slides were washed three times between each step with bond wash buffer or water. Slides were scanned on a Hamamatsu NannoZoomerS360.

Statistical significance was calculated between two groups by the Student unpaired t test. One-way ANOVA with the Tukey HSD after test was used to calculate statistical significance between multiple groups. Significance for survival was calculated by Kaplan–Meier with long-rank analysis. Analyses were performed using GraphPad Prism 8. Error bars represent SEM and a P value of <0.05 was considered statistically significant (*, P < 0.05; **P, < 0.01; ***, P < 0.001).

The data generated in this study are available within the article and its Supplementary Data Files or upon request from the corresponding author.

We previously identified an immune evasion pathway in sarcomas where tumor cell–derived RA promotes monocytes to differentiate into TAMs (11). To explore whether other types of cancer display similar RA-dependent immune evasion, we analyzed publicly available TCGA RNA-seq data from human tumors for expression of RALDH isozymes, finding high RALDH1 transcripts in HCC (Fig. 1A). This was also confirmed through qRT-PCR in archived FFPE specimens from patients in our institution (Supplementary Fig. S1A). In contrast, the other two RALDH isozymes were not highly expressed in HCC when compared with other tumors in the TCGA RNA-seq database (Supplementary Fig. S1B and S1C). Next, we examined whether high RALDH1 expression is associated with a specific subtype of HCC. A previous report described three distinct molecular subtypes of HCC based on DNA copy number, DNA methylation, mRNA expression, miRNA expression, and proteomics (16). Computational analyses of RNA-seq data downloaded from the aforementioned study (15) showed high RALDH1, but not RALDH2 or RALDH3, in all subtypes, suggesting that RALDH1 overexpression is a hallmark of HCC (Fig. 1B). This was further confirmed by performing RALDH1 IHC on primary and metastatic HCCs, as well as unrelated tumors, which revealed strong RALDH1 staining in HCCs (Fig. 1C; Supplementary Fig. S1D).

The high level of expression of RALDH1 in HCC tumors may come from tumor cells, immune infiltrates, or other stromal components. To identify the primary source, we examined a published and publicly available scRNA-seq of human HCC (17). Tumor cells and hepatocytes were the dominant producer of RALDH1 in this dataset and expressed low amounts of RALDH2 and 3 (Supplementary Fig. S1E). We next measured transcript levels of the three RALDH isozymes in five distinct human HCC cell lines, finding high RALDH1 in all (Fig. 1D). To measure RALDH enzyme activity, we used the AldeRed assay. Consistent with elevated RALDH1 transcripts, we detected AldeRed positivity in all human HCC cell lines tested (Fig. 1E). Likewise, murine HCCs also displayed high Raldh1 and AldeRed positivity (Fig. 1F and 1G). Of note, we showed that normal liver expresses all three Raldh isozymes whereas murine HCCs appear to loose/suppress Raldh2 and 3 (Supplementary Fig. S1F). Thus, HCCs are likely dependent on RALDH1 for RA production. To confirm this, we deleted RALDH1 in human HCC cells using CRISPR/Cas9 and found dramatic reductions in AldeRed activity (Fig. 1H). Finally, we confirmed high RA in HCC cells through LC-MS–based measurement of ATRA, which is the dominant biologically active isomer of RA formed through RALDH1-catalyzed oxidation of Retinaldehyde (Fig. 1I).

Taken together, results in this section demonstrate that HCCs produce high levels of RA via RALDH1.

RA can drive autocrine or paracrine signaling by binding RAR/RXR transcription factor heterodimers to regulate gene expression (12). Thus, reducing RA production by inhibiting RALDH enzymes and/or blocking RA signaling through RAR/RXR has the potential to curtail the RA-mediated tumor immune evasion we previously described (11). However, RA is an important morphogen and signaling molecule, which precludes global RA blockade as a therapeutic strategy. RAR and RXR have several isoforms that generate a diverse repertoire of RAR/RXR heterodimers. Although isoform-specific inhibitors of RARs and RXRs have been developed, they are used as tool compounds due to toxicity and lack of approved clinical indications. In contrast, RALDH isozyme–specific inhibitors as a strategy for RA blockade have not been adequately explored (19). We recently developed best-in-class RALDH1 inhibitors (Raldh1-INH) that showed good specificity and potency (13). Within this series, two inhibitors, Compounds 86 (C86) and 91 (C91), also known as NCT-505 and NCT-506, respectively, showed PK and pharmacodynamic (PD) profiles favorable to potential clinical applications (13). Thus, we examined whether C86 and/or C91 could inhibit RA production in HCC cells. Both inhibitors reduced AldeRed fluorescence in human HCCs, although C86 displayed higher potency (Fig. 2A and B; Supplementary Table S2). Thus, we primarily used C86 in subsequent experiments. C86 or C91 did not lead to compensatory increases in transcription of any other RALDH isozymes, which is consistent with the AldeRed data and demonstrates the efficacy of these inhibitors in suppressing RA production in HCC (Supplementary Fig. S2A). In contrast with human HCC, both inhibitors failed to reduce AldeRed fluorescence in murine HCC cells (Fig. 2C). LC/MS-based RA measurements confirmed the lack of RA suppression by C86 in murine HCC (Fig. 2D). Thus, C86 and C91 activity show species specificity. Homology modeling based on partial crystal structure of these Raldh1-INH suggests that differences in key drug-interacting amino acids between mouse and human may underlie this observation (Supplementary Fig. S2B). Finally, we confirmed that the reduced RA and AldeRed activity in human HCC cell lines with C86 or C91 was not due to increased cell death or reduced cell viability (Fig. 2E and F; Supplementary Fig. S2C).

Thus, results in this section demonstrate the efficacy of Raldh1-INH in abrogating RA production in human HCC cell lines.

To examine whether HCCs regulate monocyte differentiation, we cocultured primary human monocytes with either human HCC cell lines or cell culture supernatant (CM) from these cell lines. Flow cytometry analysis showed suppression of DC differentiation in the presence of HCC cells or CM (Fig. 3A). qRT-PCR analyses confirmed this, showing suppression of DC-associated genes and increased expression of macrophage-associated genes with CM (Supplementary Fig. S3A). These findings were reproduced in murine monocytes cultured with HCC cells or CM (Fig. 3B; Supplementary Fig. S3B). The effects were reversed when HCC cells were pretreated with the Raldh1-INH C86 (Fig. 3A and B; Supplementary Fig. S3A and S3B). Thus, HCC-derived RA regulates monocyte differentiation in vitro, which can be “rescued” by blocking RA production through Raldh1-INH.

Next, we tested this in vivo by transplanting immunodeficient (NU/J) mice with the human HCC cell line Huh7 and, once the tumors were established, performing intratumoral injection with primary human monocytes. Mice were treated with vehicle (control) or C86. In this setting, C86 selectively inhibited RALDH1 in the transplanted human cells (HCC and human monocytes) due to the species-specificity of this inhibitor described above. Five days after monocyte transplantation the tumors were analyzed by flow cytometry where human and murine leukocytes were distinguished using standard species-specific anti-CD45. Raldh1-INH treatment increased DC differentiation from transplanted human monocytes (Fig. 3C). This reflects the effect of reducing RA production in HCC cells and is consistent with results from the coculture experiments described above. Although a direct effect of the inhibitor on the transplanted human monocytes could also explain this observation, this is unlikely given that the host (murine) immune cells in the TME, which are insensitive to C86, also showed increased frequency of DCs and reduced frequency of macrophages (Fig. 3D). Furthermore, C86 treatment of monocytes in vitro did not alter their potential to differentiate into DCs (Fig. 3A). Finally, we confirmed RA as the key mediator of the aforementioned effects of HCCs on monocyte differentiation by using RALDH1-KO HCC cells, whereby CM from the knockout cells failed to suppress DC differentiation (Fig. 3E).

Taken together, data presented in this section show that Raldh1-INH suppresses HCC RA production and its attendant impact on monocyte differentiation.

Monocytes can differentiate into macrophages or DCs and we previously described the adaptive immune consequences of RA-mediated suppression of DC differentiation from monocytes in tumor immunity (11). However, whether and how tumors are affected by RA-induced macrophages remained unclear. We show above that HCC-derived RA can increase macrophage frequency in the TME (Fig. 3D). To examine whether RA also alters macrophage function, we adopted a macrophage-tumor cotransplantation approach. Primary human monocytes were differentiated into macrophages in the presence or absence of RA, mixed 50:50 with the human HCC cell line (Huh7), and transplanted into immunodeficient NU/J mice. HCC cells transplanted without macrophages served as an additional control. RA-treated macrophages accelerated tumor growth compared with HCC cells transplanted alone or with control macrophages (Fig. 4A; Supplementary Fig. S4A). Therefore, we next assessed the overall impact of macrophages in HCC TME by depleting TAMs through intraperitoneal liposomal clodronate (CloLipo)—a common method to deplete macrophages (20). We confirmed TAM reductions with CloLipo and found reduced tumor growth in this setting (Fig. 4B; Supplementary Fig. S4B). Thus, HCC-associated TAMs support tumor growth, a property that could be induced by high RA in the HCC TME.

To further explore this, we took the reductionist approach of coculturing RA-induced macrophages with HCC cells. First, we noticed that RA exposure increased macrophage numbers, which is consistent with our in vivo observations in the TME and suggests that RA may increase macrophage proliferation and/or survival (Fig. 4C). RA pretreated macrophages significantly increased tumor cell numbers compared with nontreated control macrophages; this effect was reversed when the macrophages were exposed to RAR signaling inhibitor BMS493 (Fig. 4D). Consistent with increased tumor cell numbers, CFSE labeling suggested increased tumor cell proliferation in the presence of RA-treated macrophages when compared with untreated control macrophages (Fig. 4E). These effects were recapitulated when HCC cells were grown with CM from RA-treated or control macrophages, suggesting that RA exposure may lead to production of a soluble “mitogenic” factor by macrophages (Supplementary Fig. S4C). Of note, tumor cells showed reduced proliferation when cultured with control (untreated) monocytes/macrophages, likely due to competition for nutrients (Fig. 4D). Pretreatment with RA abolished this suppressive effect, indicating the tumor-supportive effect of RA exposure. The experiment with CM (Supplementary Fig. S4C) clarified this further as the absence of monocytes/macrophages eliminated this competition for nutrients and “unmasked” the mitogenic effect of the soluble factor.

To summarize, results described in this section show that HCC-derived RA induces TAMs to produce factor/s that support HCC growth.

Data presented above show Raldh1-INH can alter monocyte differentiation and macrophage functions in the TME by blocking RA production in HCC cells. To examine the therapeutic implications, we transplanted huh1 and huh7 human HCC cell lines into immunodeficient mice and treated them with C86, finding significant tumor inhibition (Fig. 5A; Supplementary Fig. S5A). C86 treatment reduced RADLH activity, and hence RA production, in tumor cells but not infiltrating leukocytes (Fig. 5B). C86 displayed a dose–response, with tumor-suppressive effects at ≥10 mg/kg once daily dosing and weight loss observed only at higher doses of 40 mg/kg (Fig. 5C and D). Correspondingly, deletion of RALDH1 with CRISPR/Cas9 (RALDH1-KO) in HCC cells led to slower tumor growth in vivo, but not in vitro (Fig. 5E; Supplementary Fig. S5B). RALDH1-KO HCC tumors did not respond to C86 therapy, demonstrating that the tumor-suppressive effects of Raldh1-INH were primarily driven by on-target effects on RALDH1 (Fig. 5E). Next, we examined the role of TAMs in mediating the therapeutic effects of C86 by depleting them through CloLipo treatment. TAM depletion suppressed HCC growth and C86 treatment did not further suppress tumors in the absence of TAMs (Fig. 5F; Supplementary Fig. S5C). Thus, the therapeutic effects of reducing HCC-derived RA require the presence of TAMs.

Next, we tested whether HCC-derived RA acts directly on TAMs to promote tumor growth observed above. Toward this goal, we obtained mice that conditionally express a dominant-negative isoform of RAR from the Rosa26 locus (dnRAR^flox mouse; ref. 21). Expression of dnRAR leads to inhibition of RAR-mediated effects of RA. We crossed dnRAR^flox to Lysozyme 2-Cre mice (LysM^Cre), which express Cre recombinase in myeloid cells, including macrophages (22). Human HCC cells (Huh7) were transplanted after T-cell depletion into control and Lysm^Cre: dnRAR^flox mice, revealing significant slowing of tumor growth and reduced TAMs with myeloid-specific dnRAR expression (Fig. 5G; Supplementary Fig. S5D and S5E). This suggests that tumor-derived RA induces tumor-promoting TAMs, which is consistent with data presented in previous sections. Nonetheless, dnRAR-mediated suppression of RA signaling is partial as it only inhibits RAR-mediated pathways and not the other RAR isoforms or RXRs and very high levels of RA can still overcome RAR-inhibition by dnRAR. To further examine the impact of blocking RAR signaling, especially in the context of therapy, we treated Huh7-bearing mice with the pan-RAR blocker BMS493 alone or in combination with Raldh1-INH. Although monotherapy with BMS493 or Raldh1-INH slowed tumor growth, the combination showed the greatest effect (Fig. 5H). Thus, blocking RA production through Raldh1-INH and/or RA signaling through RAR inhibitors can suppress HCC growth.

As described above, C86 and C91 show species specificity and do not inhibit murine RALDH1 (Fig. 2C and D). Thus, the aforementioned therapeutic effects on xenotransplantation-based tumor models represent a scenario where the drug can only work on transplanted tumor cells but not host cells and there are no T cells to mount antitumor immune responses. To overcome these limitations, we revisited our original published chemical series for Raldh1-INH and based on structure identified compound-99 (C99) as a potential inhibitor of murine RALDH1 (13). In vitro, C99 suppressed RALDH1 activity in both murine and human HCC cell lines (Fig. 6A), albeit at much higher (micromolar, Fig. 6A) concentrations compared with C86 (nanomolar range, Fig. 2A). Correspondingly, we found that the IC₅₀ value of C99 to be significantly inferior to that of C86 (Supplementary Fig. S6A). Despite lower potency, C99 allowed us to overcome the aforementioned limitations of cross-species xenotransplantation models to further probe biological responses to Raldh1-INH.

C99 did not reduce proliferation or viability of the murine HCC cell line Hepa 1–6 in vitro, but it significantly suppressed tumor growth in vivo (Fig. 6B and C; Supplementary Fig. S6B and S6C). The human-specific C86 did not suppress murine Hepa 1–6 growth in vivo (Fig. 6C). C99 treatment, but not C86 treatment, reduced AldeRed activity in tumor cells and the frequency of TAMs within the TME (Fig. 6D and E). Furthermore, TAM depletion with CloLipo suppressed Hepa 1–6 tumor growth and rendered the tumors insensitive to C99 treatment (Fig. 6F). These findings mirror the effects of C86 on human HCC described in the previous sections and suggest that the tumor-suppressive activity of RALDH1 inhibitors is dependent on their ability to block RA production in tumor cells. To further confirm this, we tested C99 on the fibrosarcoma murine model of fibrosarcoma that expresses high levels of both Raldh1 and 3 and is not solely dependent on RALDH1 for RA production; C99 did not suppress fibrosarcoma growth (Fig. 6G; Supplementary Fig. S6D).

Taken together, data in this section demonstrate the efficacy of Raldh1-INH in suppressing HCC growth through inhibition of RA production.

As described above, C86 monotherapy showed therapeutic effects even in the absence of T cells in xenotransplant tumor models. C99 also suppressed tumor growth in syngeneic tumor models with intact adaptive immunity, but with two major limitations: C99 is much less potent compared with C86 or C91 and the murine Hepa 1–6 tumor cells tend to generate significant T-cell responses after subcutaneous transplantation in C57BL6/J mice, which occasionally leads to spontaneous delayed tumor rejection in our hands. These limitations make it difficult to examine the true therapeutic potential of C99 with Hepa 1–6. Nonetheless, it is important to examine the impact of full RALDH1 inhibition in the presence of T cells and test combinations with ICB. Toward this goal, we identified Hep55 as a murine HCC cell line that shows less spontaneous T-cell responses and no rejection in our hands. As shown in Fig. 1F, Hep55 cells also have high Raldh1 expression. To overcome the limited efficacy of C99, we used a genetic approach and created RALDH1 deletion in Hep55 using CRISPR/Cas9. Loss of Raldh1 gene was compatible with normal growth of Hep55 cells in vitro but led to profound tumor suppression when the cells were transplanted in vivo (Fig. 7A; Supplementary Fig. S6E and S6F). Tumor suppression was accompanied by significantly enhanced infiltration of RALDH1-KO tumors with activated T cells (Fig. 7B). Anti-PD1 treatment led to an even greater suppression of growth by RALDH1-KO Hep55 tumors (Supplementary Fig. S6G). Hep55 tumors showed infiltration with both pro- and anti-inflammatory macrophages and the frequency of pro-inflammatory macrophages increased significantly with the loss of RALDH1 activity in tumor cells (Supplementary Fig. S6H and S6I). These findings are consistent with both macrophages and T cells driving the therapeutic effects of RALDH1 inhibition. In this context, it is worth noting that growth of RALDH1-KO Hep55 tumors was suppressed to a greater extent than C86 treatment suppressed the growth of xenotransplanted human HCC. Although this could be due to a greater extent of RALDH1 inhibition with a genetic knockout, a more likely explanation is the presence of T cells in the syngeneic Hep55 model.

An important issue for the use of Raldh1-INH is on-target toxicity, especially given the expression of RALDH1 in normal liver. Although we have not observed any signs of toxicity with C86 and C91, a major caveat is the species specificity of these compounds. To examine the potential for on-target toxicity of RADH1 inhibition, we created genetic deletion of RALDH1 in mice (Fig. 7C and D). RALDH1-KO mice did not show any overt toxicity and developed normally compared with their WT siblings. Basic toxicological analyses of serum and complete blood count also did not reveal any abnormalities in RALDH1-KO and there were no significant weight differences between genotypes (Fig. 7E; Supplementary Fig. S7A and S7B). We also performed a standard panel of in vitro assays for off target effects (Eurofins Cereps Panlabs 85), which showed no concerns for off-target effects at the expected therapeutic concentration (Supplementary Table S3). Thus, the RALDH1-INH tested here are unlikely to cause serious toxicity.

Given the good efficacy and toxicity profile of RALDH1-INH, we next examined the PK and PD properties of the three RALDH1-inhibitors used in this study—C86, C91, and C99. We described above how C99 was less potent than C86 or C91. Here, we found that the half-life of C99 was also significantly inferior to the other two Raldh1-INH (Supplementary Table S4). Although a better IC₅₀ value of C86 compared with C91 (Supplementary Table S2) was our rationale for using C86 in all our experiments, we found that the half-life of C86 was lower than that of C91 when the compounds were given through oral (PO) or intravenous routes (Supplementary Table S4). Nonetheless, we used intraperitoneal delivery in all in vivo experiments above, which is associated with superior PK profile compared with PO or intravenous route (Supplementary Tables S4 and S5). C86 also showed good bio-distribution in various tissues when delivered intraperitoneally (Supplementary Table S5). Meanwhile, we formulated chow with C86 and examined PK/PD of the inhibitor over a 15-day period (Supplementary Fig. S7C). This approach showed good stability of C86 in chow and consistent drug exposure over time without affecting body weight (Supplementary Fig. S7C and S7D). Thus, Raldh1-INH show good efficacy and PK/PD profile for further preclinical development.

In summary, data presented here establish RALDH1 as a bona fide therapeutic target for HCC immunotherapy. We demonstrate its efficacy as monotherapy but given its unique mechanism of action, there are opportunities for combination with other treatment approaches. As a proof of concept, we demonstrated additive effects with ICB. Finally, we show that the basic PK/PD profile of Raldh1-INH are encouraging and support further development for eventual clinical trials. Besides illuminating a path forward for the treatment of HCC, a common, lethal, and difficult to treat cancer, our work establishes RALDH inhibition as a viable therapeutic approach in other cancers.

HCC is one of the leading causes of cancer-related death worldwide. Liver transplantation can be curative in the early stages of disease, but donor shortage leads to prolonged delays, by which time many cases progress beyond the eligibility criteria for transplantation. Therefore, even if Raldh1-INH monotherapy slows tumor progression without being fully curative, it could improve cure rates as a bridging therapy by allowing patients to successfully undergo transplantation. We also provide evidence that combination of ICB with Raldh1-INH may be more effective than either monotherapy, which warrants further investigation. Indeed, RALDH1-targeting is an entirely new approach of targeting antigen-presenting cells in the TME and could be combined with current systemic, surgical, and loco-regional therapies.

HCCs express high levels of RALDH1 but lower levels of RALDH3 and no detectable RALDH2, whereas normal liver expressed all three isozymes. Thus, HCCs are dependent on Raldh1 for RA production, which allows us to target RA production selectively in HCCs and limit on-target off tumor toxicity. Along these lines, it is reassuring to note that the RALDH1-KO mice showed no overt signs of toxicity. Nonetheless, the species specificity of RALDH1-INH poses some challenges for toxicological studies needed for human trials. Our homology modeling suggests that the inhibitor will likely work on primate RALDH1, which could be used as the animal model. Alternatively, we can introduce human RALDH1 gene into the murine locus to generate a “humanized” murine model for such studies.

The present work builds upon our previous observation that tumor-derived RA blocks DC differentiation from monocytes, instead inducing monocytes to differentiate into TAMs (11). In that article, we mostly focused on the consequences of blocked DC differentiation, showing that RA blockade can rescue DC differentiation and induce T-cell responses. A notable observation in the present study is that RA blockade leads to significant tumor suppression even in the absence of T cells. We showed that this effect is mediated through macrophages and that CM from RA-exposed macrophages promoted HCC growth, suggesting that RA may induce production and release of a soluble mitogenic factor from macrophages. Transcriptional profiling of RA–exposed macrophages showed upregulation of several genes, such as Lipocalin2, that could potentially mediate this mitogenic effect, but further studies are needed (23). Although species specificity of our Raldh1-INH precluded us from directly addressing the role of T cells, it is very likely that the presence of T cells will augment the antitumor responses of the inhibitor based on enhanced T-cell infiltration and activation observed in RALDH1-KO tumors. Further support for this notion would be provided if there is clonal dominance of intratumoral T cells in RALDH1-KO tumors compared with parental tumors, and investigation of this is something we are currently pursuing. Nonetheless, this study highlights RA's impact on the TME beyond regulation of adaptive immune responses.

HCCs typically develop in the setting of chronic liver inflammation. With the advent of preventive and therapeutic strategies in viral hepatitis, the etiology of HCC is gradually shifting to nonviral inflammatory pathologies such as alcoholic steatohepatitis, nonalcoholic steatohepatitis, and nonalcoholic fatty liver disease. Whether RALDH1-mediated RA production plays a role in the development of HCC in these inflammatory settings is an important unknown. If true, Raldh1-INH could be even considered for preventative therapy. Investigation of this could be conducted in inflammation-associated murine HCC models such as those induced by DEN. Indeed, the subcutaneous transplantation-based HCC models used in this study have some limitation and we plan to conduct additional studies with orthotopic transplants, patient-derived xenografts, and autochthonous HCC models to further explore the impact of Raldh1-INH. In closing, research presented in this article identifies RALDH1 as a new potential therapeutic target in HCC, reports development of new inhibitors of RALDH1, provides proof-of-concept for using RALDH1 inhibitors for HCC immunotherapy, and highlights a new approach to altering the tumor immune microenvironment.

S.-M. Yang reports a patent for WO2019089626 pending and issued, US11505559 issued, and US20230079399, EP3704102, EP4155300, and 63/452,978 pending. G. Rai reports a patent for 63/452978 pending. N.J. Martinez reports a patent for WO2019089626 pending and issued, US11505559 issued, and US20230079399 and 63/452978 pending. A. Yasgar reports a patent for WO2019089626 pending and issued, US11505559 issued, and US20230079399 and 63/452978 pending. A.V. Zakharov reports a patent for 63/452978 pending. A. Simeonov reports a patent for WO2019089626 and US11505559 issued, and US20230079399, 63/452978, and E-041-2021-0-US-01 pending. M. Haldar reports grants from National Institutes of Health, grants from Department of Defense, grants from Cancer Research Institute, and grants from Burrows Welcome Fund during the conduct of the study; in addition, M. Haldar has a patent for 63/452,978 pending and reports as a cofounder of Stage One Immunotherapeutics, a startup biotech developing RALDH inhibitors for cancer immunotherapy. No disclosures were reported by the other authors.

P. Yu: Conceptualization, formal analysis, investigation, methodology, writing–original draft, writing–review and editing. S. Cao: Conceptualization, formal analysis, investigation, methodology, writing–original draft, writing–review and editing. S.-M. Yang: Investigation, methodology. G. Rai: Conceptualization, formal analysis, investigation, methodology, project administration, writing–review and editing. N.J. Martinez: Investigation. A. Yasgar: Investigation. A.V. Zakharov: Investigation, methodology. A. Simeonov: Investigation. W.A. Molina Arocho: Methodology. G.P. Lobel: Investigation. H. Mohei: Investigation, methodology. A.L. Scott: Methodology. L. Zhai: Methodology. E.E. Furth: Investigation, methodology. M.C. Simon: Investigation. M. Haldar: Conceptualization, resources, supervision, funding acquisition, investigation, methodology, writing–original draft, project administration.

This work was supported by the Department of Defense grant CA200304, National Institutes of Health Grant R37CA234027, Cancer Research Institute Grant CRI3881, Burrows Welcome Fund Grant CAMS-1013413.01 (to M. Haldar), Intramural Research Program of the National Center for Advancing Translational Sciences (NCATS), and National Institutes of Health (NIH, project 1ZIATR000192–07; to S.-M. Yang, N.J. Martinez, A. Yasgar, A.V. Zakharov, A. Simeonov, and G. Rai). We would like to thank the following core facilities at the University of Pennsylvania: Cytometrics and Cell Sorting, Human Immunology Core, Pathology Clinical Service center, Molecular Pathology and Imaging Core, ant the Crispr/Cas9 mouse targeting core. We are thankful to the metabolomics core at the Children's Hospital of Philadelphia for their LC-MS support. We would like to thank Dr. B. Keith for his helpful insights and comments throughout this project. The authors also wish to thank Xin Xu and Amy Wang from NCATS for their support in obtaining in vitro and, in vivo PK studies.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.