Mutual regulation of CAF autophagy and activation in pancreatic cancer correlates with desmoplasia, an immunosuppressive TME, and poor patient survival.
Consistent with previous studies (13), IHC and IF revealed activated autophagy in CAFs in pancreatic cancer (Figure S1A, S1B and Figure 1A). By classifying our cohorts into autophagy-high and autophagy-low groups based on their IHC intensity and area, we found that autophagy in the cancer cells of patients was correlated with poor survival. It was also uncovered that autophagy in the patient’s CAFs was correlated with poor survival (Figure 1B and C). To determine whether CAFs autophagy is relevant to AIR, we next evaluated collagen deposition and tumor-infiltrating immune cell populations in human PDAC tissues. Impressively, autophagy in CAFs, but not in cancer cells, were significantly correlated with increased collagen deposition and with fewer infiltrating CD8+ T cells (Figure 1D-G). Thus, the results indicated that autophagy in CAFs, but not in tumor cells correlated well with the induction of AIR in pancreatic cancer.
To determine the relationship between CAFs autophagy and activation, the correlation between LC3B and α-SMA was first identified in multiple assays, including, but not limited to transcription in the TCGA database and tumor tissue microarrays (Figure S1C-S1F). Furthermore, the transmission electron microscopy (TEM), immunofluorescence staining and immunoblotting results demonstrated that there was a significant reduction in the number of autophagosomes and inhibition of autophagic flux in CAFs treated with all-trans-retinoicacid (ATRA) to decrease CAFs activation (Figure 1H-L). In addition, the inhibition of genetic (ATG5 knockdown) and chemical autophagy (Chloroquine) can promote CAFs to enter a quiescent state by the immunofluorescence staining and immunoblotting analyses (Figure 1M-P). The above results suggest that CAFs autophagy and activation in pancreatic cancer regulate each other, and correlate with AIR induction and poor patient survival.
Genetic inhibition of CAF autophagy induces PD-L1-upregualation, immune escape, and desmoplastic disruption in both immune-competent mice and pancreatic cancer cells
To investigate the effect of CAF autophagy on tumor cells in vivo, we orthotopically transplanted KPC with mouse CAFs(mCAFs)-WT or mCAFs-ATG5 KD cells into immunocompetent and immunodeficient mice (Figure 2A). The results in immunodeficient mice demonstrated that suppressing autophagy in mCAFs attenuated the effect of mCAFs on tumor growth and prolonged overall survival (OS) (Figure 2B-D), as previously reported (13). Intuitively, based on the results in immune-deficient mice and previous studies in our group (14, 15), it is expected that mCAFs-ATG5 KD tumors can significantly decrease the tumor size and improve OS in immunocompetent C57BL/6 mice compared to immunodeficient nude mice; however, no significant differences were observed between the mCAFs-ATG5 KD groups and mCAFs-WT group regarding tumor growth and survival in immunocompetent mice (Figure 2E-G). Moreover, we observed that the activity (GZMB+) of infiltrating CD8+ T cells was significantly decreased, whereas no difference was observed in the number of CD8+ T cells in the mCAFs-ATG5 groups by flow cytometry and immunohistochemistry (Figure 2H, I, Figure S2A and S2B). Unexpectedly, we observed an obvious upregulation in the level of PD-L1 protein expression in the tumor cells in the mCAFs-ATG5 KD group, which was associated with a significant decrease in the tumor-infiltrated GZMB+CD8+ T cell population (Figure 2H-M, Figure S2A and Figure S2B). To confirm this conclusion, we performed immunofluorescence, immunoblotting analysis, and flow cytometry in pancreatic cancer cell lines cocultured with CAFs-WT or CAFs-ATG5 KD to detect the level of PD-L1 expression. Similarly, these results were confirmed in vitro (Figure 2Q-P and Figure S3A-S3E). Functionally, we demonstrated that KPC cells cocultured with CAFs-ATG5 KD rendered the cells more resistant to activated CD8+ T cells in T cell-mediated tumor cell-killing assays (Figure 2S and 2T), which was similar to the in vivo results. Surprisingly, we found that MHC-1 expression was unchanged in tumor cells with CAFs-ATG5 KD by flow cytometry and immunoblotting analysis both in vivo and in vitro (Figure 2H, 2I, 2N, S3F and S3G). Similarly, no differences were found between the ctrl and chloroquine diphosphate (CQ)-treated groups in immunocompetent mice, compared to immunodeficient mice. In addition, tumours with a large number of CAFs are more resistant to CQ therapy than those with little or no CAFs (Figure S4). Collectively, these findings suggest that the inhibited CAF autophagy induces PD-L1 expression in tumour cells, which promotes pancreatic cancer cell immune escape both in vitro and in vivo.
Next, to examine the effect of genetic inhibition of CAF autophagy on disruption of high desmoplastic TME, IHC was performed to assess collagen deposition, CAF activation, and the microvascular area of the tumor tissues of immunocompetent C57BL/6 mice. The results demonstrated that the collagen area and α-SMA+CAFs was sharply reduced, which indirectly promote tumour microvascular formation in KPC with the CAF ATG5 KD group (Figure S2A and S2B). Overall, these data indicate that the genetic inhibition of CAF autophagy induced PD-L1 upregulation and disrupted desmoplasia in the PDAC mouse model. These changes function to convert TME type I (PDL1−/TIL−) into type II (PDL1+/TIL+), which may represent a prerequisite for enhancing the efficacy of ICB treatment for PDAC.
Inhibition of CAF autophagy improved the in vivo anti-tumor effect of immunotherapy.
Given that the inhibition of CAF autophagy can convert TME type I (PDL1−/TIL−) into type II (PDL1+/TIL+) in pancreatic cancer, one critical question is whether targeting CAF autophagy will have an effect on the therapeutic response to ICB treatment. Thus, we constructed a genetic mouse model ATG5f/f α-SMAcreERT2 mice to knockout ATG5 in the CAFs of PDAC in vivo to investigate the effect of inhibiting CAF autophagy on enhancing the efficacy of ICB treatment of PDAC (Figure 3A). As expected, the genetic deletion of CAF autophagy can enhance the therapeutic efficacy of PD-1-targeted drugs (Figure 3B-F). These tumours in ATG5f/f α-SMAcreERT2 combination with αPD-1 displayed increased infiltration and cytotoxic effects with bulk CD8 T cells and decreased number of Tregs (Figure 3G, 3H and Figure S5). In addition, the upregulation of PD-L1 expression in tumor cells was confirmed in ATG5f/f α-SMAcreERT2 mouse model by flow cytometry and CyTOF (Figure 3I and J; Figure S6). In particular, we found that the immune escape induced by inhibiting CAFs autophagy in immune-competent mice was substantially abrogated in Cd274 knockout KPC mice. This finding confirms that the immune escape induced by inhibiting CAF autophagy is primarily dependent on the PD-1/PD-L1 signalling pathway (Figure S7A and S7B). At the endpoint, the efficiency of PD-L1 depletion using tumor cells from the mice was confirmed by flow cytometry (Figure S7C). In addition, CD8+ T cells or NK cells were depleted prior to an inoculation with KPC tumor cells with CAFs, and combined antibody treatment to determine which immune cell types were important for the effects of combination therapy. The efficacy of αPD-1 treatment was substantially abrogated by the administration of αCD8α in mice bearing KPC tumors, whereas αNK1.1 partially inhibited the antitumor response in the combined treatment, indicating that NK cells play a more limited role compared to CD8+ T cells (Figure 3K). Finally, the survival analysis also demonstrated that survival time was significantly prolonged in an ATG5f/f α-SMAcreERT2 combination with αPD-1 compared with that in the other groups (Figure 3L).
Similar to αPD-1 treatment for PDAC, we observed that αPD-L1 and αCTLA4 treatment in ATG5f/f α-SMAcreERT2 mice also significantly decreased the tumor weight and prolonged the survival time compared with the other groups (Figure S8 and S9). To be closer to the clinical treatment, overt ATG5 f/f α-SMAcreERT2 mice bearing with tumors were treated with GEM or/and αPD-1. As expected, the ATG5 f/f α-SMAcreERT2 group were sensitized to GEM or GEM+αPD-1 therapy, and exhibited drastically increased survival (Figure S10A-S10G). We further analyzed the response rate through three independently repeated treatment experiments in an orthotopic mouse model using an IVIS imaging system. The results showed that the response rate of the ATG5 f/f α-SMAcreERT2 with the GEM+αPD-1 treatment group was significantly higher compared with that of the control and WT with GEM+αPD-1 groups (Figure S10H).
Taken together, these results suggest that combining the inhibition of CAF autophagy and ICB may provide an effective treatment strategy to enhance the therapeutic efficacy of pancreatic cancer.
Deletion of CAF autophagy decreased the secretion of IL-6, which further increased PD-L1 expression via the ubiquitin proteasome system in pancreatic tumor cells.
To investigate the detailed mechanism of inhibited CAF autophagy induced PD-L1-upregualtion immune escape, we determined the difference in cytokine expression between CAFs WT and CAFs ATG5 KD using a cytokine antibody array. The levels of IL-6, IL-11, M-CSF, RANTES, and sTNFRII expression were significantly decreased in CAFs after ATG5 KD (Figure 4A and 4B). Furthermore, reduced levels of IL-6 derived from ATG5 KD CAFs were confirmed by ELISA (Figure 4C). In addition, the abundance of IL-6 expression in CAFs from pancreatic cancer tissue was significantly higher than other four cytokines in patient PDAC tissues and GEMM-KPC tumour tissues (Figure S11).
Next, to explore whether reduced IL-6 from ATG5 KD CAFs induced tumour PD-L1 expression, PD-L1 expression was examined in tumor cells with or without Tocilizumab treatment (IL-6R blockade) in CAFs-conditioned medium (CM) by flow cytometry, immunofluorescence, and immunoblotting analysis. The results demonstrated that the presence of increasingly elevated of PD-L1 expression in tumour cells under CAFs-CM treated with Tocilizumab (Figure 4D-4H). Similar to previous reports, IL-6 upregulates PD-L1 expression by enhancing its association with N-glycosyltransferase STT3A in hepatocellular carcinoma (16). Thus, pancreatic cancer cells were treated with IL-6 to detect the level of PD-L1 expression by flow cytometry and immunoblotting analysis. Interestingly, decreased in PD-L1 expression treated with IL-6 in SW1990 and Panc02 (PD-L1-high cell lines), while increased in PD-L1 expression treated with IL-6 in PANC-1 and KPC (PD-L1-low cell lines), were detected by immunoblotting analysis (Figure S12A and S12B). By contrast, membrane PD-L1 expression remained unchanged under IL-6-treatment (Figure S12C). In addition, HLA-ABC/MHC-1 was not significantly affected by IL-6-treatment (Figure S12A). Indeed, CAFs-CM plus αIL-6R significantly attenuated the T cell-mediated cytotoxic effect in vitro compared with CAFs-CM (Figure 4I and 4J).
To determine how the IL-6 pathway blockade regulates the level of PD-L1 expression, the level of PDL1 mRNA remained unchanged in the tumor cells under CAFs treated with ATG5 KD-CM or Tocilizumab or αIL-6R (Figure 4K and 4L). Moreover, treatment with Tocilizumab or mouse αIL-6R induced PD-L1 protein expression, whereas the addition of the proteasome inhibitor, MG132, not the lysosome inhibitor, CQ, blocked PD-L1 protein degradation (Figure 4M), indicating that the IL-6 pathway blockade regulates PD-L1 expression via the ubiquitin proteasome system at the post-translational level. Indeed, PD-L1 in tumor cells under CAF-CM combined with Tocilizumab or αIL-6R treatment exhibited a longer half-life than the control groups, as well as lower levels of ubiquitination (Figure 4N-4P). In addition, we performed GeneOntology (GO) and enrichment plot of RNAseq in BxPC-3 with or without Tocilizumab treatment cocultured with CAFs for 24 h. The GO analysis showed that proteasomal protein catabolic process and ubiquitin protein ligase binding were significantly altered under Tocilizumab treatment (FigureS13A and Figure S13B).
Based on the above observations, our results indicate that the genetic inhibition of CAF autophagy decreased the level of IL-6 secretion, which further increased the level of PD-L1 expression by the ubiquitin proteasome system in pancreatic tumor cells.
Transcriptional activation of USP14 by STAT3 interacted with and negatively regulated PD-L1 in pancreatic cancer.
To further illustrate the underlying mechanism by which a IL-6 pathway blockade mediates PD-L1 regulation, we first performed an intersection analysis of the proteasome-mediated degradation pathway using RNAseq in BxPC-3 cells treated with or without Tocilizumab (50 ng/mL) and cocultured with CAFs and CD274 flag-IP-LCMS. 35 genes/proteins were shown in the intersection analysis (Figure 5A-C). We also identified deubiquitinating enzyme 14 (USP14) to represent a critical regulator of IL-6 pathway blockade-mediated PD-L1 regulation by an immunoblotting analysis (Figure 5D and Figure S13C). Bioinformatic analyses across multiple cancers demonstrated that USP14 was significantly upregulated in PAAD tissues compared with the level of expression in normal tissues (Figure S14). Such reduction in USP14 was also demonstrated in tumor cells under CAF ATG5 KD-CM (Figure 5E).
It has been well-established that STAT3 represents the most important downstream transcription factor of IL-6. Accordingly, we performed a CHIP assay to analyze the level of STAT3-bound potential binding site in the USP14 promotor. As shown in Figure 5F and G, one putative binding region in STAT3 was found and promoter constructs containing mutations in this region were generated to cause a STAT3-binding deficiency. To determine whether this STAT3 site was a transcriptionally active region, a dual luciferase assay was performed in 293T cell lines to test the expression of USP14 containing WT or MUT promoter elements for STAT3. STAT3-Flag cells exhibited a significantly higher level of USP14 expression compared to the negative CTRL in both SW1990 and PANC-1. Moreover, mutating the USP14 promoter reduced the expression in NC and STAT3-Flag cells. Similar results were observed after Tocilizumab treatment. Tocilizumab-treated cells had significantly lower USP14 expression than the CTRL group in SW1990 and PANC-1 cells. Mutating the USP14 promoter reduced the expression in the ctrl and Tocilizumab-treated cells (Figure 5H).
Furthermore, the results of the IP showed tan endogenous interaction between USP14 and PD-L1 in multiple pancreatic cell lines (Figure 5I). Moreover, a glutathione S-transferase (GST)-pull down assay showed that USP14 bound to PD-L1 directly (Figure 5J). A Duo-link assay consistently demonstrated the binding between USP14 and PD-L1 in the cells (Figure 5K and L). USP14 inhibition (IU1) (deubiquitylating enzyme activity of USP14 was determined by HA-Ub-VS) or depletion increased the level of PD-L1 expression in pancreatic cancer by immunoblotting analysis, flow cytometry, and immunofluorescence (Figure 5M-T and Figure S15A-S15C). Functionally, a USP14 knockdown in pancreatic cancer cells in vitro increased the resistance of tumor cells to activated CD8+T cells in a T cell-mediated tumor cell-killing assay (Figure S15D and S15E). Next, we inoculated USP14 KD KPC cells into immunodeficient mice to study the effects in vivo. The tumor volume was significantly reduced in USP14 KD group compared with the WT group. In addition, a prolonged survival time was observed in the USP14 KD groups compared with the WT group in immunodeficient mice; however, no significant differences were observed between the USP14 KD groups and the CTRL group in terms of tumor growth and survival in immunocompetent mice (Figure S16). To explore whether a USP14 deficiency enhanced the effect of αPD-1 therapy, αPD-1 was comminated with IU1 in an immune-competent orthotopic model. Importantly, a combination with αPD-1 and IU1 further decreased the tumor weight and prolonged the animal survival time compared with that of other groups (Figure S17). Together, our results suggest that the transcriptional activation of USP14 by STAT3 interacted with and negatively regulated PD-L1 in pancreatic cancer.
USP14 destabilizes PD-L1 through specifically removing K63-linked poly-ubiquitination of PD-L1 at the K280 residue.
In contrast to upregulated PD-L1 expression following the USP14 deletion, USP14 overexpression decreased the level of PD-L1 expression in pancreatic cancer, which was accumulated by treatment with the proteasome inhibitor, MG132, suggesting that USP14 regulated PD-L1 expression via the proteasome system (Figure 6A and B). Indeed, the USP14 deletion exhibited a longer half-life than the control groups, and elevated K63-linked ubiquitination of PD-L1, which was reduced by K48-linked ubiquitination (Figure 6C-E). Moreover, an in vitro deubiquitination assay in a cell-free system further confirmed that USP14 could directly remove K63-linked ubiquitin chains from PD-L1, but not cleave the canonical K48-linked ubiquitination of PD-L1, since this would be expected to negatively regulate the abundance of PD-L1 protein expression (Figure 6F). Data from the Duolink assay indicated that the binding between PD-L1 and USP14 (Duo: red) occurred in the ER (Calnexin; green). The number of PD-L1-USP14 PLA dots were positively correlated with the level of calnexin fluorescence intensity (Figure 6H). In addition, the cell fractionation results showed that Eer I treatment indeed rescued the level of PD-L1 in the ER (Figure 6I).
PD-L1 is a transmembrane protein, and the cytoplasmic domain of PD-L1 (PD-L1-ICD; 260-290 residues) is involved in multiple regulation pathways controlling PD-L1 protein ubiquitination and degradation (17). Therefore, we speculated that USP14 can remove ubiquitin chains from the ICD region of PD-L1. Two evolutionarily conserved USP14 ubiquitination-specific motifs were identified across multiple species, centering at the K271 and K280 residues, were further confirmed by a ubiquitinationmic analysis (Figure 6J and K). We constructed GFP-PD-L1 K271R and GFP-PD-L1 K280R mutants to examine whether USP14 could affect GFP-PD-L1 K271/280R expression and ubiquitination. In PD-L1 KO KPC cells, the K280R mutant displayed significantly decreased levels of PD-L1 expression, compared with GFP-WT PD-L1 and GFP-K271R PD-L1 (Figure 6L). Furthermore, in USP14 KD KPC cells, the ubiquitination of GFP-WT PD-L1 was significantly increased, whereas the K280R mutant displayed virtually no ubiquitination (Figure 6M). Thus, USP14 destabilizes PD-L1 through specifically removing K63-linked poly-ubiquitination of PD-L1 at the K280 residue, but not K271, as confirmed by a bioinformatics analysis (Figure S18).
Targeting CAF autophagy renders primary PDAC tumors eradicable by immunotherapy via engineering stem cell-derived biomimetic vesicles
To explore the efficacy of combination therapy between ICB and the inhibition of autophagy in PDAC, we found that inhibiting chemical autophagy (chloroquine) sensitizes PDAC tumours to ICB therapy (Figure S19), as previous reports(18). However, inhibition autophagy by chloroquine via an intraperitoneal administration orthotopic mouse model could not specifically target CAFs autophagy. Thus, to achieve targeted delivery of chloroquine diphosphate (CQ) to CAFs, a biomimetic drug delivery system, termed mesenchymal stem cell (MSC)-Lipo, was prepared according to the procedure illustrated in Figure 7A, as previously reported (19, 20).
The hydrodynamic size of MSC-Lipo was 74.267 nm ± 14.614 nm and the zeta potential of MSC-Lipo decreased to -3.980 mV ± 0.314 mV with the addition of negative charges of the MSC membrane (Figure S20A). The results of transmission electron microscopy (TEM) showed that MSC-Lipo exhibited a clear and complete bilayer structure with no adhesion (Figure 7B and Figure S20B). The colocalization of liposomes (labeled by DiD, red) and MSC membrane (labeled by DiO, green) illustrated that these two components were well integrated at a weight ratio of 1:0.5 (Figure S20C). The immunoblot analysis indicated that MSC-specific marker proteins (e.g., CD105, CD90, and CD44) were preserved after membrane integration using a sonication probe (Figure 7C). The total protein profile of MSC-Lipo was consistent with that of the MSC membrane, whereas liposomes did not exhibit any protein expression according to SDS-PAGE results (Figure 7D). The MSC-Lipo characterization results indicated that MSC-Lipo inherited some properties of specific proteins on the surface of the MSC was successfully prepared. Furthermore, MSC-Lipo also maintained 48.123% ± 0.696% drug encapsulation efficiency of CQ after sonication (Figure 7E). Therefore, stable drug loading capability lays the foundation for MSC-Lipo as a drug delivery carrier.
To investigate the delivery capacity of MSC-Lipo in an orthotopic pancreatic TME, the cellular uptake and tumor targeting ability of MSC-Lipo were evaluated both in vitro and in vivo. The flow cytometry and IF analysis and demonstrated that MSC-Lipo has a better uptake in mCAFs than KPC cells (Figure 7F-7H). Compared with free-DiD, MSC-Lipo presented a stronger tumor targeted ability in vivo imaging system, especially 12 h after intravenous injection (Figure 7I-J). Moreover, to further study the cellular uptake in tumor tissues, the result demonstrated that a substantial amount of MSC-Lipo was taken up by the mCAFs (orange fluorescence) (Figure 7K). Both the in vitro and in vivo results confirmed that MSC-Lipo could targeting deliver drugs to mCAFs rather than KPC.
Furthermore, the anti-tumor effect of αPD-1 combined with CQ-loaded MSC-Lipo was explored in mice with orthotopic pancreatic cancer. The outcomes proved that αPD-1 combined with CQ-MSC-Lipo most effectively suppressed tumor growth, which was especially reflected in the tumor volume and weight. Surprisingly, compared with an intraperitoneal injection of free CQ, an intravenous injection of CQ-MSC-Lipo exerted a better anti-tumor effect with the reduction of the dosage to 4.6% and a dosing frequency of 33.33% at the same time (Figure 7L and M). There was no significant difference in body weight between each group at the end of treatment (Figure 7N), whereas all groups were within the normal range for liver and kidney function (Figure S21A). The weight to body ratio of each organ and H&E staining did not reveal a significant difference between each group (Figure S21B and S21C), indicating that CQ-MSC-Lipo had substantial biocompatibility in the treatment of pancreatic cancer.
In conclusion, the findings of our study revealed that a deletion in CAFs autophagy reduced the level of IL-6 production, disrupting desmoplasia, and decreasing the level of USP14 expression transcription in pancreatic cancer cells. Indeed, we identified USP14 as the post-translational factor upregulating PD-L1 expression by removing K63 linked-ubiquitination at the K280 residue. Functionally, the autophagy-deficient CAFs improve the efficacy of PD-1/PD-L1 mAbs and gemcitabine treatment of pancreatic cancer in an immune-competent mouse model (Figure 8).