Tryptophan 2,3-Dioxygenase Expression Identified in Human Hepatocellular Carcinoma Cells and in Intratumoral Pericytes of Most Cancers

Tryptophan catabolism is one of the immunosuppressive mechanisms acting in the tumor microenvironment (1–3). Two distinct enzymes can convert tryptophan into kynurenine: indoleamine 2,3-dioxygenase 1 (IDO1) and tryptophan 2,3-dioxygenase (TDO protein encoded by gene TDO2; refs. 4, 5). Although both enzymes are located in the cytosol, their activity induces tryptophan depletion and kynurenine accumulation in the extracellular space, because of the action of amino-acid transporters (6–10). The concerted effects of tryptophan depletion and kynurenine accumulation make the tumor microenvironment strongly immunosuppressive, reducing the proliferation of effector T lymphocytes and favoring the differentiation of regulatory T cells (11–14).

In steady state, the expression of IDO1 is restricted to endothelial cells in the placenta and lung, to mature dendritic cells in secondary lymphoid organs and to scattered epithelial cells in the female genital tract (15). However, in inflammatory conditions, IDO1 expression is strongly induced by interferon gamma (IFNγ; refs. 16, 17). It is therefore highly expressed in inflammatory sites, where it appears to contribute to the negative feedback of the immune response that prevents immunopathology (15). Placental IDO1 expression prevents immune rejection of the fetus by the maternal immune system (18). IDO1 is also expressed in many cancers and contributes to their resistance to immune rejection (1, 15). Inflamed tumors can express IDO1 as an adaptive resistance mechanism, in which IDO1 is induced by the IFNγ produced by tumor-infiltrating T cells (19). However, a number of human tumors express IDO1 constitutively, as an intrinsic resistance mechanism that prevents T-cell infiltration and may account for a number of cold, poorly immune cell–infiltrated, tumors (20). Such constitutive IDO1 expression, which depends on oncogenic signaling through the MAPK and/or PI3K pathways, is exerted via COX2 inducing an autocrine loop of prostaglandin secretion, and can be repressed by clinically available COX2 inhibitors, thereby promoting tumor rejection (20). IDO1 activity can also be blocked by small-molecule inhibitors, several of which are currently in clinical development (21). The most advanced IDO1 inhibitor, epacadostat, recently completed a phase III clinical trial testing the combination of epacadostat with anti–PD-1, pembrolizumab, in advanced melanoma (22). This trial showed no clinical benefit of the combination compared with anti–PD-1 alone. The potential reasons for this failure are numerous, but include a suboptimal dosing regimen and a lack of inclusion of tumors that constitutively express IDO1 in the trial. Another potential reason is the involvement of additional actors mediating tryptophan catabolism in the tumor microenvironment (23, 24).

The other enzyme able to degrade tryptophan into kynurenine is TDO, which is expressed at a high level in the liver and degrades excess dietary tryptophan to maintain stable levels of systemic tryptophan (25, 26). Expression of TDO is regulated in part by glucocorticoids (27), whereas TDO is posttranslationally stabilized by its substrate (28, 29). TDO expression is also detected in the decidualized endometrium, where its function is unknown (30–32), and in the brain where it might contribute to the synthesis of neuroactive compounds (33). TDO2 mRNA is also found in a number of human tumor samples, including hepatocarcinomas, glioblastomas, melanomas, and bladder carcinomas (12, 34). Expression of TDO in mouse tumors can prevent their immune rejection, an effect reverted by systemic treatment of mice with a small-molecule inhibitor of TDO (34, 35). These results prompted the preclinical and early clinical development of TDO inhibitors (26, 34–36). TDO is constitutively expressed in a number of human tumor cell lines, including glioblastoma, colorectal, head and neck, lung and gall bladder carcinoma cell lines (34). In these lines, TDO protein expression and activity was confirmed by enzymatic assays (34). In human tumor samples, TDO2 expression is detected by RT-qPCR (12, 34), but the lack of a reliable TDO-specific antibody prevented confirmation of protein expression as well as identification of the cell type(s) expressing TDO. These uncertainties have hampered clinical development of TDO inhibitors. In this report, using novel TDO-specific monoclonal antibodies (mAb), we confirm TDO protein expression in many human tumor samples, and identify the TDO-expressing cells in these samples.

DBA/2 mice (Harlan Laboratories), C57BL/6J Ola Hsd mice (Envigo), and C57BL/6 TDO-KO mice (provided by Dr Hiroshi Funakoshi and Dr Toshikazu Nakamura from Osaka University; ref. 25) were bred at the animal facility of the Ludwig Institute for Cancer Research, Brussels, Belgium. C57BL/6 and C57BL/6 TDO-KO mice were mated to obtain heterozygous offspring, which were intercrossed to obtain Tdo2^+/−, Tdo2^+/+ (WT), and Tdo2^−/− (TDO-KO). Homozygous TDO-KO and WT mice littermates were selected by PCR on genomic DNA using primers: mTdo2-forward: GTA-TCT-ATG-GAG-GAC-AAT-GAA-G, mTdo2-reverse: GAT-GAA-TAG-GTG-CTC-GTC-ATG; neomycin-resistance-gene-forward: GTT-CTT-TTT-GTC-AAG-ACC-GA, neomycin-resistance-gene-reverse: TTT-CCA-CCA-TGA- TAT-TCG-GC. Mice were used at 8 to 12 weeks of age. Animal studies were conducted in accordance with national and institutional guidelines for animal care and with the approval of the Comité d'Ethique pour l'Expérimentation Animale from the Secteur des Sciences de la Santé, UCLouvain (2011/UCL/MD/015 and 2015/UCL/MD/015).

We created a pcDNA3 plasmid construct (Thermo Fisher Scientific) encoding a fusion protein comprising human TDO at the C-terminal end of the type II transmembrane human CD134 ligand (hCD134L) protein (Supplementary Fig. S1A). We immunized DBA/2 mice with this plasmid three times at 2-week intervals by electrotransfer of 20 μg DNA into tibialis muscles (10 μg into each muscle) using the pulse generator Clinivet (IGEA). Electrotransfer comprised eight pulses of 200 V/cm amplitude lasting 20 ms followed by a pause of 480 ms. Two weeks after the last immunization (i.e., on day 42), we boosted mice by injecting in the footpad 10⁵ living P1.HTR cells stably transfected with pCD1 hCD134L hTDO expression vector, which were created as described (ref. 37; Supplementary Fig. S1B). The boost was repeated on days 56, 70, and 115 using 10⁵, 10⁶, and 90 × 10⁶ cells, respectively. On day 119, the mouse with the best TDO antibody response was sacrificed, and lymphocytes from the spleen and lymph nodes (inguinal and mesenteric) were hybridized with cells from fusion partner SP2neo using standard procedures (polyethylene glycol or electrofusion). Hybridomas were selected in HAT medium, screened for production of TDO antibodies by ELISA, and confirmed by FACS.

P1.HTR cells were derived in our lab from the P815 mastocytoma cell line (38). HEK-293-EBNA (293E) cells were purchased from InvivoGen. 293E cells transfected with human TDO cl119 were previously created (34), and confirmed to still express enzymatically active TDO (Supplementary Fig. S2A). HEK-293T cells were a generous gift from Prof. Kris Thielemans in 2006. A172 (glioblastoma) cells were purchased from Antisense Pharma. Huh-7 (liver cancer) cells were a gift from A. Patel (Oxford). TDO-expressing LB159-CRCA cells were derived in 1991 at the Ludwig Institute for Cancer Research Brussels Branch from a colorectal carcinoma patient sample from the Cliniques universitaires Saint-Luc, Brussels (34). They were cultured with homemade ACL-4 supplement. LB159-CRCA TDO^−/− cells were generated using CRISPR/Cas9 as described below. SK-CO-11 (colon cancer) and MZ-CHA-3 (gall bladder cancer) were received from collaborators Alex Knuth and Thomas Wölfel in Mainz in 1991. LB1317-SCCHN (head and neck cancer) cells were derived at the Ludwig Institute Brussels Branch from a patient sample received from Marc Hamoir and Vincent Grégoire (Cliniques universitaires Saint-Luc, Brussels) in 1996. U-87 MG (glioblastoma) cells were purchased from Cell Lines Services GmBH in 2011. All cell lines were tested for Mycoplasma in October 2019. Human cell lines were authenticated in November 2019 by short tandem repeat profiling (Promega Powerplex hs 16). To prepare formalin-fixed, paraffin-embedded (FFPE) cell pellets, cells were fixed for 20 minutes with 4% formaldehyde, washed with PBS, and resuspended in 3% low-melting point agarose-PBS. The agarose blocks were embedded in paraffin using the Vacuum Infiltration Processor (Tissue-Tek). Briefly, they were dehydrated using ethanol-water baths with increasing ethanol concentrations, incubated in xylene, and finally embedded in paraffin.

LB159-CRCA TDO^−/− cells were generated using CRISPR/Cas9 as follows. Two different pairs of guide RNA (GR1-forward: CACCGTTTAAAAAACTCCCCGTAGA, GR1-reverse: AAACTCTACGGGGAGTTTTTTAAAC; GR2-forward: CACCGGCAAAGGAGGTCTTATCTAT, GR2-reverse: AAACATAGATAAGACCTCCTTTGCC) were cloned into lentiCRISPR v2 plasmid (Addgene). Target guide sequence cloning was performed as follows: Digestion and dephosphorylation of lentiviral v2 CRISPR plasmid were performed with BsmBI (NEB, #R0580S) and SAP (NEB, #M0371S) for 30 minutes at 37°C, followed by inactivation for 15 minutes at 65°C. Gel purification of the digested plasmid was performed using QIAquick Gel Extraction (Qiagen 28706). Phosphorylation and annealing of each pair of oligonucleotides (GR1 and GR2) were performed using T4 Ligase (NEB, #M0202S) and the T4 Ligation Buffer containing ATP (NEB, #B0202S) with a cycle of 37°C for 30 minutes, 95°C for 5 minutes, and cooling to 25°C at 5°C/minute. Ligation was performed using the Quick Ligation protocol (NEB, #M2200S) according to the manufacturer's instructions. XL-1 blue competent cells (Agilent 200249) were transformed, and the plasmids were purified with NucleoSpin Plasmid (Macherey Nagel). Sequencing of selected plasmid clones was performed prior to generating the lentiviruses. Lentiviruses were produced in HEK-293T cells seeded at 2 × 10⁶ cells/plate (standard petri dish). Twenty-four hours after plating, the cells were transfected with 5 μg/mL lentiviral plasmid and a second-generation lentivirus packaging system. Plasmids were diluted in FBS-free DMEM with Turbofect reagent (Thermo Fisher, #R0531) and incubated at room temperature for 15 to 20 minutes. The plasmid/turbofect mix was then added to the cells in a drop-wise fashion, and the plate was gently rocked to achieve even distribution of the plasmids. Six hours after transfection, the medium was exchanged with fresh medium. The supernatant of the cells was collected 24 and 48 hours after the transfection, filtered using a 0.45-μm filter and stored at −80°C. LB159-CRCA cells were infected with both lentiviruses. LB159-CRCA cells were seeded at 1 × 10⁶ cells/well in 6-well plates with 4 mL of medium. Twenty-four hours after seeding, the medium was removed, and 1 mL of lentivirus-GR1 or -GR2 supernatant collected 48 hours after transfection was added. Cells were incubated for 4 hours prior to adding 1 mL of fresh culture medium. Cells were cultured for 72 hours prior to selection with puromycin (6 μg/mL). The knockout of TDO was validated in clone 1 by DNA sequencing and by enzymatic assays (Supplementary Fig. S2A).

Normal tissues and tumors were obtained from surgical resections or autopsies from 171 patients. The tissues used in this publication were collected, fixed in formaldehyde, embedded in paraffin and provided by (i) the BioLibrary of the Institut Roi Albert II of the Cliniques universitaires Saint-Luc, Brussels, Belgium, projects CDCUCLR27 and CDCUCLR16-2016; (ii) the Service d'Anatomie Pathologique of the CHU-Brugmann, Brussels, Belgium; (iii) BHUL—Université de Liège—CHU de Liège, Belgium. The study was approved by the Commission d'Ethique Biomédicale Hospitalo-Facultaire from the UCLouvain (B403201316588) and conducted under the guidelines of the Helsinki declaration. Patients gave informed consent for the use of their tissues in this study.

Five μm-thick paraffin sections were deparaffinized in 3 baths of Histo-Clear (3 minutes each), washed in butanol for 3 minutes and progressively rehydrated in 100%, 90%, 70%, 50% ethanol and in demineralized water (3 minutes each). Antigen retrieval was performed with Tris/EDTA buffer at pH 9 in a 2100 Antigen Retriever (Aptum) using a predefined heating cycle. All the following steps were performed at room temperature. Endogenous peroxidases were blocked with Peroxidase Block (Dako) for 15 minutes and protein blocking was done for 1 hour with TBS-Tween containing 2% milk, 5% bovine serum albumin (biotin-free BSA), and 1% human IgG. The primary antibodies (Table 1) were diluted in immunohistochemistry (IHC) diluent (Enzo) and incubated for 1 hour. After washing, the secondary antibodies (Table 1) were incubated for 1 hour. For IHC, the staining was revealed with HIGHDEF DAB substrate (Enzo) or with AEC+ Substrate Chromogen (Dako) between 5 and 20 minutes, and counterstaining was performed with hematoxylin. Slides were incubated for 5 minutes in hematoxylin, washed with demineralized water, washed in tap water for 5 minutes and transferred again to demineralized water. Slides were mounted with HIGHDEF IHC mount (Enzo). For multiplex immunofluorescence (IF), the staining was revealed with the Tyramide Signal Amplification system (Table 1). Tyramide reagent was used at 1:50 in borate buffer. The antibodies were then eluted with citrate buffer at pH 6 using microwave treatment at 900 W until boiling followed by 15 minutes at 90 W. The following staining was started with protein blocking and performed as described above. After the last staining, microwave treatment was performed, the nuclei were counterstained for 5 minutes with Hoechst 33342 (Invitrogen) used at 20 μg/mL in TBS-Tween 0.1% with 10% BSA, and the slides were mounted with HIGHDEF IHC fluoromount (Enzo). The slides were digitalized using a Pannoramic 250 Flash III tissue scanner (3DHISTECH) at ×20 magnification. Multiplex-immunostained entire tumor sections were evaluated quantitatively using the image analysis tool Oncotopix version 2017.2 (Visiopharm).

Five μm-thick paraffin sections were deparaffinized and progressively rehydrated as described for IHC and IF. Endogenous peroxidases were blocked with RNAscope H₂O₂ (ACD) for 10 minutes at room temperature (RT). Antigen retrieval was performed with RNAscope Target Retrieval Reagents (ACD) for 15 minutes in a steam cooker. The slides were immediately transferred to water at RT, washed with ethanol and dried for several minutes up to overnight at RT. The slides were incubated with RNAscope Protease Plus (ACD) for 30 minutes at 40°C in a Dako Hybridizer. After washing with RNAscope Wash Buffer Reagents (ACD), the slides were incubated for 2 hours at 40°C with the primary probes: RNAscope Probe Hs-TDO2 (ACD, #416471), RNAscope Negative Control Probe dapB (ACD, #310043), RNAscope Hs-POLR2A (positive control, ACD, #310451). After extensive washing, the slides underwent incubation with 6 successive amplification probes from the revelation kit RNAscope 2.5 HD Detection Reagent-BROWN (ACD, #322310): AMP1 (30 minutes, 40°C), AMP2 (15 minutes, 40°C), AMP3 (30 minutes, 40°C), AMP4 (15 minutes, 40°C), AMP5 (30 minutes, RT), AMP6 (15 minutes, RT). The staining was revealed with DAB from the same kit for 10 minutes at RT. Counterstaining was done with hematoxylin, and the slides were mounted with HIGHDEF IHC mount (Enzo).

RNA of frozen cells and tissues was extracted with NucleoSpin RNA (Macherey Nagel) according to the manufacturer's instructions. Whole tissues were beforehand crushed in the lysis buffer of the kit using a TissueLyser LT (Qiagen). RNA of FFPE tissue sections was extracted with NucleoSpin total RNA FFPE XS (Macherey Nagel) according to the manufacturer's instructions. RNA was quantified with a NanoDrop spectrophotometer and a defined amount of RNA was retrotranscribed by the RevertAid RT Kit (Thermo Fisher Scientific). TaqMan qPCR was performed with Takyon ROX Probe 2X MasterMix dTTP blue (Eurogentec) in a StepOnePlus thermal cycler (Applied Biosystems) using the following program for human IDO1, human TDO2, and murine Tdo2: 3 minutes at 95°C, then 40 cycles of 10 seconds at 95°C and 1 minute at 60°C; for human EF1 and murine β-actin: 3 minutes at 95°C, then 40 cycles of 3 seconds at 95°C and 30 seconds at 60°C. The following primers were used (F = forward, R = reverse, P = probe):

hIDO1:

F 5′-GGTCATGGAGATGTCCGTAA-3′

R 5′-ACCAATAGAGAGACCAGGAAGAA-3′

P 5′-CTGTTCCTTACTGCCAACTCTCCAAGAAACTG-3′

hTDO2:

F 5′-CATGGCTGGAAAGAACTC-3′

R 5′-CTGAAGTGCTCTGTATGAC-3′

P 5′-TTTAGAGCCACATGGATTTAACTTCTGGG-3′

hEF1 (EEF1A1):

F 5′-GCTTCACTGCTCAGGTGAT-3′

R 5′-GCCGTGTGGCAATCCAAT-3′

P 5′-AAATAAGCGCCGGCTATGCCCCTG-3′

mTdo2:

F 5′-GTATCTATGGAGGACAATGAAG-3′

R 5′-GATGAATAGGTGCTCGTCATG-3′

P 5′-CCTCCTTTGCTGGCTCTGTTTACACC-3′

mβ-actin (Actb):

F 5′-CTCTGGCTCCTAGCACCATGAAG-3′

R 5′-GCTGGAAGGTGGACAGTGAG-3′

P 5′-ATCGGTGGCTCCATCCTGGC-3′

The probes were coupled to 5′ FAM and 3′ TAMRA. Standard curves were added for each gene. Mβ-Actin or hEF1 were used for normalization.

Tryptophan and kynurenine were quantified in the cell culture supernatants by HPLC based on the retention time and the UV absorption (280 nm for tryptophan, 360 nm for kynurenine; ref. 34).

This protocol was used in Supplementary Fig. S2A. The cells were lysed in a homemade lysis buffer (0.1% SDS, 1% sodium deoxycholate, 0.5% Nonidet P40) with complete Protease Inhibitor Cocktail (Sigma-Aldrich). A defined amount of protein was heated at 70°C for 10 minutes with NuPAGE LDS Sample Buffer (Thermo Fisher Scientific) and NuPAGE Sample Reducing Agent (Thermo Fisher Scientific). Proteins (20 μg) were loaded on Bolt 4%–12% Bis-Tris Plus Gels (Thermo Fisher Scientific) and separated by gel electrophoresis using Bolt MOPS SDS Running Buffer (Thermo Fisher Scientific). Dry transfer was performed by iBlot (Thermo Fisher Scientific) using iBlot Gel Transfer Stacks, nitrocellulose, regular size (Thermo Fisher Scientific), and a transfer program of 7 minutes (1 minute at 20 V, 4 minutes at 23 V, 2 minutes at 25 V). The membranes were incubated overnight at 4°C with the primary antibodies: mouse anti-TDO clone III at 1 μg/mL, clone V at 1 μg/mL, or mouse anti-β-actin clone AC-15 at 1:10,000 (Sigma-Aldrich, #A5441). After washing in PBS Tween 20, the membranes were incubated with HRP-linked goat anti-mouse IgG (R&D Systems, #HAF007) at 1:5,000 for 1 hour at RT and washed again. HRP was revealed using SuperSignal West Pico (Thermo Fisher Scientific) and detected with Fusion Solo S (Vilber Lourmat). TDO and β-actin were stained on the same membrane, and stripping between both stainings was performed by 15-minute incubation in Restore PLUS Western Blot Stripping Buffer (Thermo Fisher Scientific).

This protocol was used in Supplementary Fig. S2B and S2C. The cells and whole livers were lysed in Pierce Ripa buffer (Thermo Fisher Scientific) with Halt Protease and Phosphatase Inhibitor Cocktail (Thermo Fisher Scientific). The livers were crushed using a TissueLyser LT (Qiagen). A defined amount of proteins was heated at 95°C for 5 minutes with Laemmli loading buffer. Proteins (20 μg) were loaded on NuPAGE Bis-Tris 4–12% gels (Thermo Fisher Scientific) and separated by gel electrophoresis using NuPAGE MOPS SDS Running Buffer (Thermo Fisher Scientific). Dry transfer was performed like in protocol #1. After blocking, the membranes were incubated for 2 hours at RT with the primary antibodies: mouse anti-TDO clone III at 1 μg/mL, clone V at 1 μg/mL or mouse anti-vinculin clone hVIN-1 at 1:10,000 (Sigma-Aldrich, #V9131). After washing, the membranes were incubated with HRP-linked goat anti-mouse IgG (BioLegend, #405306) at 1:2,500 in blocking buffer for 1 hour at RT and washed again. HRP was revealed like in protocol #1.

Statistical analyses were performed using Prism 6 (GraphPad Software). P values of ≤ 0.05 were considered statistically significant. Correlation analysis was performed using two-tailed Spearman rank correlation. Two paired groups were compared using the two-tailed Wilcoxon matched-pairs signed rank test.

Given the lack of fully validated antibodies recognizing TDO, we first produced new TDO-specific mAbs by immunizing mice with DNA and boosting with cells engineered to express full-length human TDO at the cell surface (37). We selected two clones (clone V, IgG2aκ, and clone III, IgG2bκ) and validated them based on the following criteria: They recognized a single band of the expected size (47 kDa) only in human cell lines that express TDO2 mRNA (Supplementary Fig. S2A and S2B). This band disappeared after CRISPR-Cas9 genetic inactivation of TDO2 (Supplementary Fig. S2A and S2B). In TDO2-negative human cells, the same band appeared upon transfection of TDO2 (Supplementary Fig. S2A). This band was absent in cells expressing IDO1 but not TDO2, indicating specificity (Supplementary Fig. S2B). Although both mAbs recognized human TDO, only mAb V recognized mouse TDO, and stained a single band in liver lysates from WT but not from TDO^−/− mice (Supplementary Fig. S2C). Of note, human and murine TDO are 89% identical (protein BLAST database). IHC of fixed pellets of human cells showed a cytoplasmic staining in cell lines expressing TDO but not in TDO-negative cell lines (Supplementary Fig. S2D).

We used the aforementioned antibodies to analyze the expression of TDO in human tumors and determine the proportion of TDO expression and the type of cells that express TDO. We stained a large number of human tumors of various histologies with both mAbs and found TDO-positive cells in most of them (Fig. 1A and Table 2). The results obtained with both TDO mAbs were identical. Importantly, in all cases we used adjacent sections to detect the presence of TDO mRNA by RT-qPCR and/or in situ hybridization (ISH) to validate the specificity of the stainings (Fig. 1B). Samples were considered positive only when they were stained by IHC with both antibodies and contained TDO2 mRNA.

With the exception of astrocytomas, we found TDO-positive samples in all the tumor types tested. All hepatocarcinomas expressed TDO and showed cytoplasmic staining in most tumor cells (Fig. 1A). This is in line with the physiologic expression of TDO in normal hepatocytes (39). Other tumors, including glioblastomas, bladder, pancreatic and colon carcinomas, also showed a high proportion of TDO-positive samples, but here the staining was limited to some cells in focal tumor areas, usually corresponding to vascular structures. Such focal staining pattern was observed in all the TDO-positive tumors other than hepatocarcinomas (Fig. 1A; Table 2). Interestingly, although only half of the tested melanomas and breast carcinomas were TDO positive, more than 80% of metastatic lymph nodes contained TDO-expressing cells, which were absent from nonmetastatic lymph nodes (Table 2).

The vascular structures expressing TDO were often located at the tumor periphery or surrounded necrotic and hemorrhagic areas (Fig. 1A). In tumor-infiltrated lymph nodes, TDO-positive vessels were found mainly at the borders of the metastases and at the interface between tumor and lymphatic tissue. We then wondered whether the level of TDO expression correlated with clinical features, such as the age of the patients, their treatment, the tumor grade, and the presence of lymph node or distant metastases. Considering all the tumor types tested (except hepatocarcinoma), we found that the proportion of TDO-expressing cells was inversely correlated with tumor differentiation, with higher proportions in higher histologic grades (Fig. 1C). An illustrative example is glioma: whereas grade II and III astrocytomas were all TDO-negative, 97% of glioblastomas multiforme (grade IV gliomas) contained TDO-positive cells (Table 2).

To identify the cell type expressing TDO in these vascular structures, we performed a series of costainings (Fig. 2A). We observed that all TDO-positive cells expressed the beta-type platelet-derived growth factor receptor (PDGFRβ), a marker of mesenchymal cells [fibroblasts, pericytes, vascular smooth muscle cells (vSMCs); ref. 40], whereas not all PDGFRβ-positive cells expressed TDO (Fig. 2A). Some, but not all, TDO-positive cells expressed α-smooth muscle actin (αSMA) and NG2 (neural/glial antigen 2; Fig. 2A). TDO-positive cells also surrounded CD34-positive endothelial cells, but TDO never colocalized with CD34 (Fig. 2A). We therefore concluded that TDO-expressing cells corresponded to vascular smooth muscle cells or pericytes. These two cell types, which have very similar phenotypes, can be identified according to their localization toward the basement membrane in normal vessels, but can no longer be distinguished in disorganized tumor vessels (41, 42). In addition to TDO-positive pericytes, which were identified in all tumor types including hepatocellular carcinomas (HCC), some tumors also contained TDO-expressing tumor cells, including, as mentioned above, all HCCs, but also 10 of 39 glioblastomas, in which TDO was partially coexpressed with glial fibrillary acidic protein (GFAP) and β3-tubulin (Fig. 2A), and 1 of 10 kidney carcinoma (Fig. 1B). These TDO-positive tumor cells surrounded TDO-positive vessels bordering necrotic tumor areas. In none of the tested tumors was TDO expressed by macrophages or dendritic cells (HLA-DR-positive; Fig. 2A).

Although most tumors contained TDO-positive cells, these cells were few in numbers. In the 34 analyzed glioblastoma multiforme tumors (the tumor type with the highest amount of TDO-expressing pericytes), only 5.9% of the PDGFRβ-stained area was also positive for TDO (workflow described in Supplementary Fig. S3). We used PDGFRβ rather than αSMA or NG2 as a pericyte marker, because not all TDO-positive cells expressed αSMA or NG2 (Fig. 2A). However, PDGFRβ is not exclusively expressed by pericytes, but also by cancer-associated fibroblasts (43). Therefore, 5.9% is probably a slight underestimation of the proportion of TDO-positive pericytes.

TDO-positive tumor vessels often displayed abnormal features that were similar to the vascular anomalies previously observed in tumors (42). These features included the accumulation of little vessels surrounded by several layers of pericytes (Fig. 2B, first line), and enlarged vessels with a single, irregular layer of pericytes that did not completely cover the endothelial cells (Fig. 2B, second line). We also found normal TDO-positive capillaries (with a diameter of about 10 μm and a single pericyte layer that completely surrounded the endothelial cells), but they were much less abundant (Fig. 2B, third line). We therefore wondered whether TDO-positive vessels were well perfused with blood or whether these tumor areas were hypoxic. We found no costaining of TDO with the hypoxia marker carbonic anhydrase 9 (CA9; Fig. 2C). This result suggested that TDO-positive tumor regions were well oxygenated, but did not exclude that blood and oxygen were supplied by the neighboring TDO-negative vessels.

We wondered if TDO colocalized with IDO1, the other tryptophan-metabolizing enzyme. IHC stainings of glioblastoma multiforme showed that IDO1 was expressed by very few cells, which were either tumor cells or cells with a dendritic cell–like shape (Supplementary Fig. S4A). Costainings with TDO were performed on five tumors and showed that both proteins never colocalized (Supplementary Fig. S4B).

In mouse tumor models, we previously described the immunosuppressive role of TDO when expressed homogeneously by tumor cells (34). This situation was nicely reflected in human hepatocarcinomas, which expressed TDO in all tumor cells (Fig. 1A). It was unclear, however, whether the low proportion of TDO-expressing vascular cells we observed in other tumors would suffice to catabolize tryptophan in a meaningful immunosuppressive way. This was difficult to address in mouse models, in which we never observed a similar expression of TDO in tumor vessels (36). To try to address this question, we costained 29 human glioblastomas for T lymphocytes (CD3 and CD8) and TDO and used an image analysis software to quantify lymphocyte infiltration in TDO-positive versus TDO-negative tumor areas (workflow described in Supplementary Fig. S5). Glioblastomas have the advantage that they rarely expressed IDO1 (ref. 15; Supplementary Fig. S4). As expected, T-cell infiltration was low in these glioblastomas (Fig. 3A; ref. 44). Tumor regions within 60 μm of a TDO-expressing cell (whether a pericyte or a tumor cell) contained slightly more—not less—CD3-expressing lymphocytes, whether or not they also expressed CD8 (Fig. 3A; Supplementary Fig. S5). Moreover, T lymphocytes present in TDO-positive areas tended to be more proliferative, as indicated by Ki67 staining (Fig. 3B; Supplementary Fig. S5). These results suggested that the presence of TDO-positive pericytes does not prevent but might instead favor lymphocyte infiltration in glioblastoma, although this enrichment might result from the fact that the TDO-positive areas were close to vessels, whereas the TDO-negative areas we used for comparison were not, for technical reasons, selected to be close to vessels. Although our approach did not assess lymphocyte function or differentiation, these data do not support an important immunosuppressive role of TDO-expressing pericytes in the tumor microenvironment.

We then took advantage of our validated TDO-specific mAbs to evaluate the expression profile of TDO in normal tissues. We stained nontumoral sections from the organs harboring the tumors listed in Table 2, including 5 pancreata, 1 uterus, 3 colons, 2 stomachs, 2 lungs, 6 kidneys, 10 livers, and 10 bladders. These tissues were negative for TDO, with the exception of liver and lung. As expected, all liver samples showed TDO expression in hepatocytes (Fig. 4A).

Unexpectedly, both lung samples contained TDO-positive pericytes (Fig. 4A and B). To determine whether TDO expression was linked to the presence of a tumor in the lung, we stained 11 additional nontumoral lung samples collected at a distance from the lesion in patients with or without benign or malignant tumor, as indicated in Table 3. Five of them contained TDO-expressing pericytes (Fig. 4A and B). Interestingly, all 7 samples containing TDO-positive cells also showed signs of inflammation, some of them also containing granulation tissue, which is tissue formed during wound healing and characterized by neovascularization, extracellular matrix formation, and the presence of immune cells (Table 3). In contrast, none of the six samples without inflammation contained TDO-expressing vessels (Table 3). These results indicated that TDO expression in lung pericytes was associated with inflammatory processes.

Because the murine decidua is known to express TDO (30–32), we also tested human deciduas of the first trimester and ectopic pregnancies by IHC. We found TDO expression only in a few cells, which corresponded to pericytes (PDGFRβ-positive) located inside chorionic villi, and to interstitial syncytiotrophoblasts (multinuclear cells expressing cytokeratin; Fig. 4A and B). Trying to identify the mechanism responsible for TDO expression in pericytes, we treated murine mesenchymal stem cells (MSC), which are precursors of pericytes, with dexamethasone, or with different combinations of pregnancy hormones or cytokines able to induce inflammation or differentiation to fibroblasts/vSMCs. However, we could not detect any induction of TDO2 mRNA (Supplementary Table S1). As TDO is expressed in the murine decidua, the maternal part of the placenta (31, 32), we also treated decidualized and undecidualized uterine cells. Except for a small increase of TDO expression by LPS and IFNγ in decidualized cells, TDO was not induced (Supplementary Table S1).

In conclusion, we observed massive TDO expression in human hepatocarcinoma tumor cells, but a limited expression in other tumors, restricted to pericytes in tumor vessels located in angiogenic areas. Similar TDO-expressing pericytes were observed in nontumoral inflammatory lesions of the lung and in placental tissue.

So far, the evidence supporting TDO expression in human tumors is 3-fold. First, TDO2 mRNA is detectable in human tumors by RT-qPCR performed on bulk tumor RNA (34). Second, a number of human tumor lines express TDO at the mRNA and protein level and effectively degrade tryptophan into kynurenine, which is inhibited by TDO inhibitors (12, 34). Third, IHC studies of several human tumor types (colon carcinoma, breast cancer, non–small cell lung carcinoma, ovarian carcinoma, renal cell carcinoma, and brain metastases of malignant melanoma) report widespread and homogeneous expression of TDO in tumor cells (12). Based on the expression of TDO in human tumor lines and on the staining patterns reported by Opitz and colleagues (12), it is considered that TDO is expressed by tumor cells themselves in human tumors. However, no validated TDO-specific mAb was available at the time, so that the reported IHC stainings were performed with nonvalidated polyclonal antibodies, which carry the inherent risk of cross-reactivity, particularly when used for IHC, which does not control for the molecular size of the signal (15). In this study, we described and fully validated two highly specific mAbs recognizing TDO. They stained a single band of the expected molecular weight when used in western blots and stained, by both western blot and IHC, cell lines in a TDO expression–dependent manner. Using these mAbs, we confirmed TDO expression in a large number of human tumors. In HCC, TDO was highly expressed in tumor cells themselves. In most other tumors, however, we did not observe TDO expression in tumor cells, but only in scattered cells often located next to necrotic or hemorrhagic areas. The TDO specificity of this staining was further confirmed by ISH, which detected a similar expression pattern of TDO2 mRNA in adjacent sections. TDO-expressing cells belonged to vascular structures with abnormal morphology, typical of vascular anomalies previously reported in tumor vessels (42). Further stainings identified these TDO-expressing cells as pericytes/vSMCs, expressing PDGFRβ. The lack of TDO expression in tumor cells other than HCC came as a surprise, particularly given the expression of TDO in several human tumor lines of glioblastoma and carcinoma of the colon, head and neck, lung and gall bladder (12, 34). This disconnection between tumor lines and tumor samples is intriguing, even more so when considering human HCC, in which the opposite disconnection was observed: although TDO was expressed in tumor cells in HCC samples, HCC lines tested proved negative for TDO expression (34).

The function of hepatic TDO is to regulate systemic tryptophan concentrations, as illustrated by the 9-fold increased level of tryptophan in the serum of TDO-KO mice (25, 36). By analogy to the tryptophan-degrading enzyme IDO1, the proposed function of TDO in tumors is the inhibition of immune-mediated tumor rejection (12, 34). The impact of IDO1 on immune evasion is well defined via several in vivo experiments using different tumor models (1, 20, 45, 46). In humans, IDO1 is expressed by a large number of tumors, and high IDO1 expression is positively correlated with tumor progression and decreased survival (2, 21, 47). Several IDO1 inhibitors are now in clinical development (21). Similar to IDO1, TDO negatively affects T lymphocyte activity and proliferation as seen in several in vitro and in vivo experiments (12, 34, 48). TDO-positive cells block in vitro IFNγ production and proliferation of T lymphocytes (48) and TDO favors escape of murine tumors to immune rejection (12, 34). The outcome of LPS-induced endotoxemia is worsened in TDO^−/− mice due to the increased secretion of proinflammatory cytokines (49). The dominant expression of TDO in human HCC confirms the potential of TDO as a therapeutic target in this cancer type. Although checkpoint inhibitors were recently approved in HCC (50, 51), their clinical efficacy remains limited and could be improved upon combination with TDO inhibitors, several of which have been developed (26, 34–36).

In tumor types other than hepatocarcinomas, including gliobastomas, melanomas, and carcinomas of the pancreas, colon, stomach, lung, endometrium, breast and kidney, the role of TDO expression in pericytes associated with abnormal tumor vessels is unclear at this stage. A local immunosuppressive function is possible and in line with previous reports indicating that tumor-derived vascular pericytes can cause T helper cell anergy (52). Pericytes derived from human glioblastomas have stem cell–like and immunosuppressive properties associated with TDO expression, albeit at a very low level (53). Because we could not find expression of TDO in murine tumor–associated pericytes, we could not experimentally test the immunosuppressive role of TDO-expressing pericytes in vivo (36). In human glioblastoma, our quantification of T cells in tumor areas containing TDO-positive cells did not show a reduced infiltration or proliferation of CD3- or CD8-positive cells. This result suggests that TDO-expressing pericytes did not negatively affect the recruitment and infiltration of T cells into the tumor, although they could still affect the functional status or anergic state of tumor-infiltrating lymphocytes. However, although present in most tumor samples, TDO-expressing pericytes comprised only about 6% of pericytes within a given sample. It was unclear whether such a low proportion of TDO-expressing cells was sufficient to induce the level of tryptophan degradation that is required to induce immunosuppression in the whole tumor mass.

We observed that TDO-positive vessels were often located around necrotic and hemorrhagic tumor areas, which were characterized by neovascularization. We also found TDO-expressing vessels in nontumoral lung granulation tissue, which also involved neovascularization, and in chorionic villi at the time of embryonic vessel formation. TDO could therefore play a role in neovascularization, by way of the vasoactive function of one of the tryptophan-derived metabolites. Kynurenine produced by endothelial cells expressing IDO1 during inflammation can induce vessel relaxation, and the severity of septic shock is directly correlated to the expression of IDO1 (54, 55). IDO1-deficient mice have a reduced density of pulmonary blood vessels (56) and show reduced neovascularization in models of oxygen-induced retinopathy and lung metastasis (57). In addition, a newly described tryptophan-derived tricyclic hydroperoxide, formed by IDO1 in the presence of H₂O₂, causes vessel relaxation of IFNγ-pretreated mouse abdominal aortas (58). This or a similar compound could also be produced by TDO, given the similar enzymatic mechanism of both enzymes. A link between TDO expression and neovascularization would also be in line with our observation that glioblastomas, which display much more neoangiogenesis as compared with astrocytomas, also contained much more TDO-expressing vessels.

It is unclear what the mechanism that triggers TDO expression in tumor pericytes is. Rat hepatocytes increase TDO2 expression upon treatment with glucocorticoids (27), whereas murine decidual cells do so upon exposure to estrogen and progesterone (32). However, we were unable to induce TDO2 expression by adding dexamethasone, different combinations of pregnancy hormones or cytokines or inflammatory stimuli to murine MSC or to uterine decidualized cells. As we observed TDO-positive pericytes in angiogenic tumor areas, granuloma tissues and in placenta, the induction of TDO might have resulted from a complex interplay of different physiologic and pathologic phenomena related to hypoxia and inflammation. In line with this, recent reports describe the induction of TDO in fibroblasts, which, like pericytes, are derived from MSCs: TDO was upregulated in uterine stromal fibroblasts by natural killer cells (59) and in normal lung fibroblasts by galectin-1 secreted from lung tumor cell lines (60).

In sum, our results confirm TDO expression at the protein level in a large range of human cancers. The high TDO expression by HCC cells warrants the development of TDO inhibitors in this indication. In cancers without massive TDO expression in tumor cells, TDO inhibitors may synergize with immunotherapy in a different manner, by blocking hepatic TDO and increasing systemic tryptophan levels, as we report in the accompanying manuscript (36).

B.J. Van den Eynde is a scientific advisory board member for and has ownership interest (including patents) in iTeos Therapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: D. Hoffmann, B.J. Van den Eynde

Development of methodology: D. Hoffmann, J.-C. Renauld, N. van Baren, B.J. Van den Eynde

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D. Hoffmann, T. Dvorakova, V. Stroobant, M. Solvay, S. Klaessens, M.-C. Letellier, N. van Baren

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D. Hoffmann, V. Stroobant, C. Bouzin, A. Daumerie, N. van Baren, J. Lelotte, E. Marbaix, B.J. Van den Eynde

Writing, review, and/or revision of the manuscript: D. Hoffmann, N. van Baren, E. Marbaix, B.J. Van den Eynde

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D. Hoffmann, T. Dvorakova, V. Stroobant, C. Bouzin, A. Daumerie, M. Solvay, S. Klaessens, M.-C. Letellier, J.-C. Renauld, N. van Baren, J. Lelotte, E. Marbaix, B.J. Van den Eynde

Study supervision: B.J. Van den Eynde

Other (contribution of material for the study): E. Marbaix

The authors thank Dr. Hiroshi Funakoshi, Dr. Toshikazu Nakamura, and Prof. Michael Platten for providing TDO-KO mice; Guy Warnier and his team for the production of TDO-KO mice; the BioLibrary of the Institut Roi Albert II of the Cliniques universitaires Saint-Luc (Brussels, Belgium), the Service d'Anatomie Pathologique of the CHU-Brugmann (Brussels, Belgium), BHUL—Université de Liège—CHU de Liège (Belgium), and Prof. Christine Galant for providing tissue samples; Vanesa Bol for the ISH protocol; Dr. Pascal Brouillard, senior manager of the Genomics Platform of UCLouvain, for genotyping of the cell lines; and Luc Pilotte and Auriane Sibille for editorial assistance. D. Hoffmann was supported by FNRS-FRIA (grant number: 1.E082.14), T. Dvorakova by FNRS-Télévie (grant number: 7.4597.18), V. Stroobant and B.J. Van den Eynde by the Ludwig Institute for Cancer Research, C. Bouzin, A. Daumerie, and J.-C. Renauld by UCLouvain, M. Solvay by FNRS-Aspirant (grant number: 1.A385.16), S. Klaessens by FNRS-FRIA (grant number: 1.E100.14), M.-C. Letellier by iTeos Therapeutics, N. van Baren by de Duve Institute, J. Lelotte by Cliniques universitaires Saint-Luc, and E. Marbaix by Cliniques universitaires Saint-Luc.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.