N-glycosylation Regulates Intrinsic IFN-γ Resistance in Colorectal Cancer: Implications for Immunotherapy

Mouse Models of Colon Carcinogenesis

Colitis-associated carcinogenesis was induced in 6- to 8-week-old mice in 2 independent facilities using azoxymethane (AOM) and dextran-sulfate sodium (DSS) with slight protocol variations. Intraperitoneal injection of AOM was performed on day 1 (Ifngr2^ΔIEC: 10 mg/kg body weight Ifngr2^ΔIEC−2: 12.5 mg/kg body weight) and DSS-containing drinking water (Ifnγr2^ΔIEC: 2% Ifngr2^ΔIEC−2: 2.5% DSS solution) was given for 5 (Ifngr2^ΔIEC−2) to 7 (Ifngr2^ΔIEC) days. DSS supplementation was repeated twice, separated by 14 days of normal drinking water. Animals were killed around day 80. Spontaneous colon carcinogenesis was evaluated in Ifngr2^ΔIEC−2Apc; CPC mice or control mice (Ifngr2^fl/wtApc; CPC). Mice were killed at 5 months old. Colon length, tumor number, and diameter were determined macroscopically, and the tumor load was calculated as the cumulative tumor diameter in millimeters per mouse.

Patients

CRISPR-Cas9 Gene Editing

Single-guide RNAs were designed using Benchling (Biology Software; 2017; retrieved from https://benchling.com). Cloning was performed using the pSpCas9(BB)-2a-Puro (PX459) V2.0 vector (Addgene #62988) and pL-CRISPR.EFS.tRFP vector (Addgene #57819), and confirmed by sequencing. HT-29 cells were transfected with the PX459 plasmid by Lipofectamine 2000 and, 24 hours later, were selected with puromycin for another 72 hours. Afterward, single-cell expansion was performed in a 96-well plate to grow clonal cells. HT-29 cells were then transduced by viral particles generated from the transient transfection of HEK293TN cells with 3 different plasmids encoding VSV-G (pMD2.6, Addgene plasmid 12259), packaging genes (psPAX2, Addgene plasmid 12260) and the pL-CRISPR.EFS.tRFP plasmid. After 72 hours of transduction, the cells were sorted by fluorescence-activated cell sorting for RFP-positivity and single cells were seeded in 96-well plates for further clonal expansion. The clones were screened for knockout of the IFNγRα via Western blotting.

Inhibition of the IFN-γ Response in Intestinal Epithelial Cells Promotes Tumorigenesis

To determine whether the specific loss of sensitivity to IFN-γ by tumor cells influences intestinal tumorigenesis, we generated a mouse model with conditional deletion of the IFN-γ receptor beta chain (IFNγRβ encoded by Ifngr2) in intestinal epithelial cells by crossing Ifngr2^fl/fl mice with Villin-Cre mice (Supplementary Figure 1A). The resulting Ifngr2^ΔIEC mice showed a reduction in Ifngr2 mRNA expression in colon tissue, and an absence of murine IFNγRβ in colon epithelial cells while the protein was still present in stromal cells (arrows), confirming the specificity of the knockout (Figure 1A and B). Colon carcinogenesis was induced in Ifngr2^ΔIEC and Ifngr2^fl/fl control mice by treatment with AOM and DSS. Ifngr2^ΔIEC mice developed more and larger tumors than Ifngr2^fl/fl control mice (Figure 1C–F). This increase in tumorigenesis was not due to an increase in inflammation because Ifngr2^ΔIEC mice treated solely with DSS showed less colonic inflammation than control Ifngr2^fl/fl mice (Supplementary Figure 1B–E). These results could be replicated independently in another mouse facility using an independent strain (Ifngr2^ΔIEC−2, Supplementary Figure 1F and G). The conditional deletion of Ifngr2 in a sporadic CRC mouse model (Apc; CPC mice, which spontaneously develop colon tumors) similarly increased tumor number and load compared with controls (Figure 1G and H). Hence, genetic mouse models demonstrated that tumor cell intrinsic resistance to IFN-γ increases both colitis-associated and spontaneous colon tumorigenesis.

Loss of IFNγRα Expression Correlates With Decreased Disease-specific Survival in Patients With CRC

The prognostic value of IFN-γ pathway gene expression was assessed in patients with CRC. The mRNA expression of IFNGR1, but not IFNGR2, STAT1, JAK1, or JAK2, was associated with disease-free survival in The Cancer Genome Atlas cohort (Figure 1I and Supplementary Figure 2A). These results were confirmed using an independent cohort of patients with CRC, where low IFNGR1 mRNA expression correlated with a reduced disease-specific survival (Figure 1J, Supplementary Table 2). We also observed a modest reduction (less than 2-fold) in IFNGR1 mRNA expression in CRC tissues compared with matching patient normal tissues (Figure 1K, Supplementary Table 3). We could confirm that mutations in the IFN-γ response pathway genes are rare in CRC (Supplementary Figure 2B) and not related to disease-specific survival (Supplementary Figure 2C).

Gene expression analysis in whole tissue mRNA samples is not systematically predictive of protein levels, nor does it allow the specific evaluation of tumor cell–associated expression. Therefore, we analyzed the protein expression of the IFN-γ receptor α chain (IFNγRα, encoded by IFNGR1) by immunohistochemistry on a tumor tissue microarray from a cohort of 416 patients with colon cancer (Figure 1L, Supplementary Table 1). Overall, the quality criteria for analysis were met in 310 of the 416 tumors. a loss of IFNγRα expression in tumor cells was observed in approximately half of the cases (158 of 310), and was associated with increased tumor size (χ² test, P = .04), lymph node invasion (χ² test, P < .0001), extramural venous invasion (χ² test, P < .0001), distant metastasis (χ² test, P = .019), and Union for International Cancer Control stage (χ² test, P < .0001). Furthermore, the absence of IFNγRα expression in tumor cells (n = 158, 50.9%) correlated with a shorter disease-specific survival (Figure 1L).

IFNγRα Expression Level and Pattern Correlate With IFN-γ Resistance in CRC Cells

To investigate whether IFNγRα loss correlates with IFN-γ resistance in CRC tumor cells, we examined the induction of ISG expression and cell death by IFN-γ in human CRC cell lines together with IFNγRα expression. Among the 11 cell lines tested, 6 were resistant to IFN-γ–induced cell death, Stat1 phosphorylation and ISG expression (GBP1, IDO1, CASP1) (Figure 2A and B, and Supplementary Figure 3A and 4B; resistant cell lines highlighted in red). In the other 5 cell lines (in blue), the induction of cell death by IFN-γ correlated with Stat-1 phosphorylation and ISG expression at the protein and mRNA levels (Figure 2A and B, and Supplementary Figures 3A and 4B).

Four of 6 IFN-γ–resistant cell lines (RKO, HCT116, SW480, SW620) displayed a strong reduction in IFNγRα mRNA and protein expression compared with the IFN-γ–sensitive cell lines as shown by Western blotting, intracellular and surface immunofluorescence staining, or flow cytometry (Figure 2C–G, Supplementary Figures 3C, 4A and B, and 5C). The 2 remaining resistant lines (DLD-1 and Caco2) expressed IFNγRα, but the protein was detected with a reduced apparent molecular weight (Figure 2C and Supplementary Figure 3C). In these cells, intracellular staining revealed a perinuclear accumulation of IFNγRα (Figure 2D, arrows), associated with the Golgi apparatus (GM130) but not with the endoplasmic reticulum (calnexin) (Supplementary Figure 5D and E). The quantification of Golgi-associated IFNγRα localization showed an increase in DLD-1 (P < .0001) and Caco2 (P = .1124) cells compared with HT-29 cells (Figure 2E). Although this finding was indicative of some intracellular retention, the cell surface expression of IFNγRα in both cell lines was similar to that in IFN-γ–sensitive cells (Figure 2F and G, Supplementary Figure 5B). Similar differences in the pattern of IFNγRα expression were observed in CRC tissue extracts (Supplementary Figure 4C), including both differences in expression level and apparent molecular size shifts toward lower molecular weight.

In contrast, mRNA and protein expression of the other mediators of the IFN-γ response (IFNGR2, STAT1, JAK1, and JAK2) was observed in all cell lines except Caco-2. In the latter, no STAT1 expression was detected at the protein level, and JAK2 mRNA expression was reduced (Supplementary Figures 3C and 4A). Therefore, Caco-2 cells were excluded from further functional tests with IFNγRα because of the presence of additional defects in the JAK-STAT pathway. According to the data retrieved from the Cancer Cell Line Encyclopedia, few mutations were found for IFNGR1, IFNGR2, STAT1, JAK1, and JAK2 in the 11 cell lines investigated, and they did not correlate with IFN-γ resistance (Supplementary Table 4). Furthermore, independent sequencing of IFNGR1 mRNA from IFN-γ–resistant cells showed no alteration, indicating that the IFNγRα size shift observed in DLD-1 and Caco-2 cells was not due to a truncating mutation or alternative splicing (Supplementary Table 4). Of note, among the 11 CRC cell lines investigated here, none harbored a frameshift JAK mutation and only 1 resistant line (RKO) exhibited overexpression of programmed death ligand 1 (PD-L1) (Supplementary Figure 3B), which was not mutually exclusive with down-regulated IFNγRα expression. Overall, these results suggested that defects in IFNγRα expression or post-translational maturation correlate with IFN-γ resistance in CRC cell lines, whereas other signaling components of the pathway are either not or only marginally involved.

IFNγRα Is Aberrantly Glycosylated in IFN-γ–resistant CRC Cells

To determine whether epigenetic gene silencing may be responsible for the downregulation of IFNGR1 expression in RKO, HCT116, SW480, and SW620 cells, cells were treated with the DNA methylation inhibitor decitabine. Decitabine treatment led to a modest but statistically significant increase in IFNGR1 mRNA expression in all 4 cell lines (Figure 3A), which was, however, not accompanied by an increase at the protein level (Figure 3B).

The expression of IFNγRα was then reconstituted by transfection in RKO, HCT116, SW480, and SW620 cells to explore whether this could restore the IFN-γ response. Although ectopic IFNγRα expression achieved by transfection in RKO, HCT116, SW480, and SW620 cells yielded protein levels comparable to those of the IFN-γ–sensitive control cells (Figure 3C), and increased IFNγRα expression at the cell surface (Supplementary Figure 6A), it failed to restore the response to IFN-γ (Figure 3D and Supplementary Figure 6B). In addition, IFNγRα showed a reduced apparent molecular weight (Figure 3C and D and Supplementary Figure 6B and C) together with an intracellular accumulation at the Golgi apparatus, similar to what was observed in DLD-1 and Caco2 cells (Figure 3E, arrows, and Supplementary Figure 6D and E). The altered migratory pattern and Golgi retention of IFNγRα observed in all IFN-γ–resistant lines, either endogenously or after ectopic expression, suggested differences in N-glycosylation.²⁵ This was supported by the observation of a similar apparent molecular weight shift after ectopic expression of the endothelial surface protein Endoglin in IFN-γ–resistant, but not in IFN-γ–sensitive CRC cells (Supplementary Figure 6F).

To assess whether IFNγRα was differently glycosylated in IFN-γ–sensitive and –resistant cell lines, we used 2 endoglycosidases with different specificity, protein-N-glycosidase F (PNGase-F) and endoglycosidase-H (Endo-H). PNGase-F cleaves all N-glycans, whereas Endo-H specifically removes high-mannose N-glycan chains (but not mature complex N-glycans). PNGase-F treatment resulted in a shift of IFNγRα band size to approximately 70 kDa in all cell lines tested, indicating that the receptor undergoes N-glycosylation to some extent in both IFN-γ–sensitive and IFN-γ–resistant cell lines (Figure 3F). In IFN-γ–sensitive cells (HT29), IFNγRα was resistant to Endo-H digestion, indicating a mature complex N-glycosylation (Figure 3F and G). In all IFN-γ–resistant CRC cell lines, IFNγRα was highly sensitive to Endo-H digestion, regardless of whether ectopically (RKO, HCT116, SW480, SW620) or endogenously (DLD-1) expressed, indicating the presence of high-mannose N-glycans characteristic of an immature, low-complexity glycosylation.

Endogenously expressed IFNγRα was then immunoprecipitated from HT-29 (IFN-γ–sensitive) and DLD-1 (IFN-γ–resistant) cells, and its modification with complex N-glycans was detected by Western blotting using the specific lectins Phaseolus vulgaris erythroagglutinin (PHA-E) and Phaseolus vulgaris phytohemagglutinin-L (PHA-L). Binding of both PHAE and PHA-L to IFNγRα was reduced in DLD-1 cells compared with HT-29 cells, indicating a decreased complexity of IFNγRα N-glycosylation in IFN-γ–resistant cells (Figure 3H).

IFNγRα Stability and Signaling Are Regulated by N-glycosylation

In the next step, we investigated whether changes in N-glycosylation were indeed able to affect IFNγRα function. Treatment of IFN-γ–sensitive HT-29 cells with the N-glycosylation inhibitors tunicamycin and 2-deoxyglucose induced a shift in the IFNγRα apparent molecular weight and inhibited IFN-γ signaling (Figure 4A, Supplementary Figure 7A–C). We then investigated whether a non-glycosylated form of IFNγRα obtained by mutation of its 5 putative glycosylation sites^25,26 (ΔG-IFNγRα) could signal in response to IFN-γ (Figure 4B). As recipient cells, we chose N-glycosylation-competent HT-29 cells with a CRISPR-Cas9–mediated knockout of IFNGR1 (IFNGR1-KO HT-29, clone sg 2.21) that exhibited a complete inhibition of IFNγRα protein expression, IFN-γ signaling, and ISG expression compared with controls (NTC) (Figure 4C and Supplementary Figure 8A–C). Ectopic expression of wild-type IFNγRα in IFNGR1-KO HT-29 cells resulted in electrophoretic migration at the expected molecular weight (~90 kDa), as well as induction of pStat-1 phosphorylation, ISG expression, membrane localization, and cell death upon treatment with IFN-γ (Figure 4D and Supplementary Figure 8D–H). In contrast, ectopically expressed ΔG-IFNγRα was detected at approximately 70 kDa by Western blot, below the molecular weight of wild-type IFNγRα and similar to the PNGase-F-treated fully deglycosylated receptor (Figure 4D). Contrary to wild-type IFNγRα, ΔG-IFNγRα had a strongly reduced IFN-γ response, both in pools of transduced cells and in single clones (Figure 4D and Supplementary Figure 8D–G). Furthermore, a dose-response analysis suggested that ligand binding of ΔG-IFNγRα is impaired compared with that of wild-type IFNγRα (Supplementary Figure 8F and G).

To determine whether immature N-glycosylation of IFNγRα might increase protein degradation, we treated RKO, HCT116, SW480, and SW620 cells with the proteasome inhibitor MG132. MG132 induced a strong dose-dependent increase in IFNγRα protein levels in all 4 IFN-γ–resistant cell lines (Figure 4E), which was, in comparison, not observed in HT-29 cells (Supplementary Figure 8I). The protein expression level of ΔG-IFNγRα was also enhanced by treatment with MG132 (Figure 4F), confirming that N-glycosylation defects can trigger proteasome-dependent degradation of IFNγRα.

Reconstitution of MGAT3 Expression Rescues IFNγRα N-glycosylation and Signaling

To determine which enzymes of the N-glycan synthesis pathway might be responsible for the downregulation of IFNγRα complex glycosylation in IFN-γ–resistant CRC cells, a quantitative reverse transcription polymerase chain reaction array was performed (Figure 5A). Multiple genes displayed differential expression patterns between IFN-γ–sensitive and –resistant cells. MGAT3 (N-acetylglucosaminyltransferase III, or GnT-III) was the only gene consistently down-regulated in all IFN-γ–resistant cells compared with IFN-γ–sensitive control cells (HT-29 and SW948; Figure 5A, arrow). MGAT3/GnT-III catalyzes the addition of β1,4-linked GlcNAc on the central mannose of the trimannosyl core of N-linked oligosaccharides, generating so-called “bisected” N-glycans. The downregulation of MGAT3 expression in IFN-γ–resistant cell lines was confirmed at the mRNA and protein levels (Figure 5C and Supplementary Figure 9A and B). In addition, the level of bisected N-glycans detected by PHA-E lectin binding correlated with MGAT3 expression, and was reduced in all IFN-γ–resistant cell lines (Figure 5B–D).

Therefore, we examined whether the modulation of MGAT3 expression influences IFNγRα complex N-glycosylation. Chinese hamster ovary (CHO) cells without MGAT3 activity (Pro-5) or with an MGAT3 gain-of-function mutant (Lec10B cells) were used to ectopically express the IFNγRα protein.²⁷ Western blot analysis revealed an increased apparent molecular weight and expression level of IFNγRα in Lec10B CHO cells compared with control Pro5 CHO cells (Supplementary Figure 9C), suggesting that MGAT3 activity increases IFNγRα bisected glycosylation and protein levels. All-trans retinoic acid (ATRA) has been shown to increase MGAT3 expression levels and to enhance the addition of bisecting N-glycans to proteins in vitro.²⁸ In the IFN-γ–resistant/MGAT3-low RKO cells, ATRA induced a dose-dependent increase in MGAT3 and IFNγRα protein expression (Supplementary Figure 9D; quantification of IFNγRα: middle panel, and MGAT3: lower panel), accompanied by a higher IFNγRα molecular weight (Supplementary Figure 9D; middle panel; green: IFNγRα upper band, gray: IFNγRα lower band), confirming that MGAT3 promotes bisected glycosylation and stabilization of the IFNγRα protein.

The addition of bisecting N-glycans by MGAT3 has been shown to reduce the affinity of cell surface proteins to the galectin lattice, particularly to galectin-3.²⁹ Consistent with this, the level of bisected N-glycans observed on immunoprecipitated endogenous IFNγRα inversely correlated with galectin-3 binding (Figure 3H and Supplementary Figure 9E), indicating that a reduced MGAT3-dependent addition of bisecting N-glycans to IFNγRα increases its association with the galectin lattice.

To assess whether the restoration of MGAT3 expression would improve IFNγRα signaling in IFN-γ–resistant cells, we stably expressed MGAT3 in IFN-γ–resistant RKO cells, resulting in a higher level of bisected N-glycans (Figure 6A, PHA-E). The increase in MGAT3 expression correlated in a dose-dependent manner with an increase in IFNγRα protein levels and a shift toward higher molecular weights as shown in 2 independent clones (cl. 1 and 2) (Figure 6B). Most importantly, responsiveness to IFN-γ was restored in the MGAT3-positive cells (cl. 2) compared with the RKO cells transfected with the empty vector (Figure 6C–E).

To confirm the relevance of our results at the clinical level, we investigated protein expression of MGAT3 and IFNγRα in tumor tissues by immunohistochemistry. a strong positive correlation between MGAT3 and IFNγRα protein levels was observed in human CRC samples (Pearson r = .8215, P = .001), which supported our in vitro data (Figure 6F and G and Supplementary Figure 9F).

Finally, we investigated whether the modulation of bisected N-glycosylation might influence colon tumor growth and checkpoint inhibitor therapy. Using a syngeneic colon tumor mouse model, we observed that ATRA treatment reduced tumor growth in 3 of 5 animals (Supplementary Figure 10A). The mean tumor diameter was decreased by 28.5% in the ATRA-treated group after 30 days but did not reach statistical significance (Supplementary Figure 10B and C, P = .0823). Anti-PD-1 treatment resulted in complete shrinkage of tumors within 2 weeks, and addition of ATRA reduced the half-life of anti-PD-1 treatment from 3.141 to 1.617 day (Supplementary Figure 10D). These results suggested that the addition of ATRA might increase the efficacy of checkpoint inhibitor treatment. Further investigations (eg, using lower anti-PD-1 concentrations) are needed to establish whether ATRA can synergistically improve the efficacy of checkpoint inhibitors.