Unfolded protein response mediator PERK governs myeloid cell-driven immunosuppression in tumors through inhibition of STING signaling

TME promotes immunoregulatory activity and UPR activation in MDSC

To elucidate the impact of the UPR priming in the regulatory activity of MDSC in tumors, we first monitored the crosstalk between the T cell inhibitory potential and the activation of UPR sensors in CD11b⁺Gr1⁺ cells from tumors (T-MDSC) and spleens (S-MDSC) of mice bearing s.c. Lewis lung carcinoma (LLC), s.c. B16 melanoma, or orthotopic ID8-Defb29-Vegfa ovarian tumors; and in splenic myeloid counterparts (iMC) from tumor-free mice. A superior capacity to impair T cell proliferation and an increased expression of total and phosphorylated PERK and IRE1a were detected in tumor-MDSC compared to splenic-MDSC or iMC (Figure 1A–B), without a distinct accumulation of MDSC subsets (Figure S1A). Also, consistent with a preferential ER enlargement in tumor-MDSC, elevated fluorescence of ER tracker (Hambrock et al., 2002) (Figure 1C, Figure S1B–C), and augmented ER dilation detected by transmission electron microscopy (Figure 1D) were found in tumor-MDSC compared to splenic-MDSC or iMC. Next, we tested whether the exposure of bone marrow-derived MDSC (BM-MDSC) to tumor explants (TES) triggered activation of the UPR and immunoregulatory potential. Pre-treatment of murine BM-MDSC with LLC-TES (Thevenot et al., 2014), or human BM-MDSC with supernatants from the renal cell carcinoma cell line 786–0 (hTES) (Rodriguez et al., 2009), resulted in upregulation of total and phosphorylated PERK and IRE1a and enhanced ability to block T cell proliferation, which were similar to the effects induced by the UPR inducer, Thapsigargin (Thaps) (Figure 1E–H). Furthermore, administration of Thaps into LLC-bearing mice promoted tumor growth, splenic-MDSC expansion, and tumor-MDSC immunosuppressive function (Figure S1D–F). These results suggest the potential link between UPR activation and immunoinhibitory activity in tumor-MDSC.

Mitigation of UPR blunts MDSC activity and boosts cancer immunotherapy

We determined whether the inhibition of the UPR activation impacted tumor growth and MDSC-linked T cell dysfunction using Tauroursodeoxycholic acid (TUDCA), a molecular chaperone that alleviates ER stress and UPR priming (Nakagawa et al., 2014; Ozcan et al., 2006). Treatment of mice bearing established LLC or B16 tumors with TUDCA delayed tumor growth (Figure 2A), which correlated with a decreased ability of tumor-MDSC to block T cell proliferation (Figure 2B, Figure S2A), and reduced expression of the MDSC-inhibitory factor Arginase I, but not NOS2 (Figure 2C). Also, in support of the mitigation of UPR by TUDCA, lower expression of phospho-PERK and phospho-IRE1a was noticed in tumor-MDSC from TUDCA-treated mice, compared to MDSC from controls (Figure S2B). Because the activity of MDSC depends on tumor burden, we then evaluated the effect of TUDCA in cultured MDSC. TUDCA reduced the ability of human BM-MDSC conditioned with hTES, murine BM-MDSC developed after exposing BM-iMC to TES, and LLC tumor-MDSC to block T cell proliferation (Figure S2C–E). Next, we sought to establish the role of T cells in the anti-tumor effects induced by TUDCA. Higher expansion of recently activated CD69⁺CD44⁺CD8⁺ T cells was detected in tumors from B16 or LLC-bearing mice treated with TUDCA (Figure 2D). Moreover, the anti-tumor actions triggered by TUDCA were not observed in immunodeficient Rag1^−/− mice (Figure 2E), suggesting the key role of T cells in the anti-tumor effects induced by TUDCA.

Next, we studied whether treatment of tumor-bearing mice with TUDCA switched MDSC into cells that elicited anti-tumor responses. Depletion of MDSC-like cells upon injections with anti-Gr1 antibody delayed B16 and LLC tumor progression in vehicle-treated mice, while restoring tumor growth in TUDCA-treated counterparts (Figure 2F). Moreover, anti-Gr1 or TUDCA treatments promoted the spontaneous expansion of recently primed CD69⁺CD44⁺CD8⁺ T cells and EGSRNQDWL-H-2D^b-tetramer⁺CD8⁺ T cells recognizing the gp100 melanoma antigen in B16-bearing mice (Figure 2G–H). Conversely, the increase of these tumor-reactive T cell groups was prevented in TUDCA-treated mice after co-administration with anti-Gr1 (Figure 2G–H), showing that TUDCA treatment transforms tumor-MDSC into T cell-stimulating cells.

Furthermore, we tested whether TUDCA amplified the effects of various forms of immunotherapy. Treatment with TUDCA plus anti-PD-L1 induced synergistic anti-tumor effects in B16-bearing mice, compared to mice treated with the single agents (Figure 2I). Also, we evaluated the effect of TUDCA in an adoptive T cell transfer (ACT) model against the experimental antigen ovalbumin (OVA), in which pre-activated anti-OVA_257–264 (SIINFEKL) OT-I cells were transferred into mice bearing OVA⁺ EG7 tumors. Higher anti-tumor responses were noted in mice treated with TUDCA plus ACT, compared to those receiving TUDCA or ACT single treatments or untreated controls (Figure 2J), which correlated with a higher frequency of IFNg-producing T cells upon activation of splenocytes with SIINFEKL (Figure 2K). Moreover, some of the mice receiving TUDCA plus ACT rejected the EG7 tumors and were resistant to re-challenge with the same tumor cells, but not against non-related tumors (Figure 2L), indicative of specific anti-tumor protective memory. Thus, our results suggest that TUDCA enhances the effects of immunotherapy in tumor-bearing mice.

PERK regulates the immunosuppressive activity of tumor-MDSC

To monitor the UPR activation status in human MDSC infiltrating tumors, we tested the expression of phospho-PERK in tumor-linked MDSC from a TMA created from 46 patients with metastatic non-small cell lung carcinoma (Met-NSCLC) through automated multispectral imaging. Higher phospho-PERK was detected in cells resembling PMN-MDSC (Pan-Cytokeratin^negCD11b⁺HLA-DR^negCD14^negCD15⁺) and M-MDSC (Pan-Cytokeratin^negCD11b⁺HLA-DR^negCD14⁺CD15^neg) from Met-NSCLC tumors, compared to counterparts from control tissues (Figure 3A). In addition, we validated the activation of PERK in MDSC from a TMA from 84 patients with high-grade serous ovarian cancer. Elevated frequency of phospho-PERK⁺ M-MDSC-like cells was found in ovarian tumors, compared to same subsets in healthy ovaries (Figure S3A). However, we could not consistently analyze phospho-PERK expression in PMN-MDSC from ovarian tumors as limited infiltration of cells resembling PMN-MDSC was noted (Figure S3A). Together, our results indicate the elevated PERK activation in tumor-MDSC from human malignancies.

To study the impact of PERK in MDSC on tumor growth, we generated myeloid cell-conditional Eif2ak3^−/− mice, after breeding Eif2ak3^fl/fl mice with mice expressing lysozyme (Lyz2)-driven Cre recombinase (Eif2ak3^fl/flLyz2-cre), which enabled gene excision in tumor-associated M-MDSC, PMN-MDSC, and macrophages, and to a lower extent in myeloid DCs (Figure S3B–C). Delayed growth of LLC and B16 tumors and prolonged survival after injection with ID8-Defb29-Vegfa ovarian tumors were observed in Eif2ak3^fl/flLyz2-cre mice compared to controls (Figure 3B, S3D). To expand our data in a spontaneous tumor model, lethally-irradiated LSL-Kras^G12D/+Trp53^fl/fl mice received bone marrow cells from Eif2ak3^fl/flLyz2-cre or Eif2ak3^fl/fl mice. Seven weeks later, mice were injected with intra-muscular adenovirus coding for Cre-recombinase and followed for growth of autochthonous flank sarcomas, which depend on MDSC activity (Rutkowski et al., 2015). Delayed tumor progression was found in LSL-Kras^G12D/+Trp53^−/− mice reconstituted with bone marrows from Eif2ak3^fl/flLyz2-cre mice, compared controls (Figure 3C). Also, a reduction in the ability to blunt T cell proliferation and in the expression of PERK and Arginase I, but not in IRE1α or NOS2, were detected in tumor-MDSC from Eif2ak3^fl/flLyz2-cre mice, compared to MDSC from Eif2ak3^fl/fl mice (Figure 3D–E). Moreover, higher expansion of cells resembling M-MDSC, but not PMN-MDSC, was observed within tumors from Eif2ak3^fl/flLyz2-cre mice (Figure S3E), which did not correlate with changes in the apoptosis marker Annexin V, or the unmitigated ER stress driver of apoptosis, DR5 (Lu et al., 2014) (Figure S3F–G). To rule out the effect of the tumor burden differences in the actions induced by PERK deletion, we obtained BM-MDSC from hematopoietic-endothelial specific PERK-null mice (Eif2ak3^fl/flTek-cre), as Eif2ak3 excision was inefficient in BM-MDSC from Eif2ak3^fl/flLyz2-cre mice (Figure S3H–K). Supporting the intrinsic role of PERK in UPR-primed MDSC, excision of Eif2ak3 abolished the amplified T cell-inhibitory effect found in BM-MDSC treated with Thaps, without altering the regulatory potential of vehicle-treated BM-MDSC (Figure 3F). Furthermore, elimination of PERK did not impact the frequency of common myeloid progenitors (CMP) or granulocyte-macrophage progenitors (GMP) (Trento et al., 2017) in the bone marrow of tumor-bearing mice nor affected MDSC development from hematopoietic precursors (Figure S3L–M). Because of the role of MDSC in T cell dysfunction, we then studied the effect of T cells in the anti-tumor responses observed in Eif2ak3^−/− mice. Higher frequency of total, recently activated, and IFNg-expressing CD8⁺ T cells was found in tumors from Eif2ak3^fl/flLyz2-cre mice, compared to controls (Figure 3G–I). Also, depletion of CD8⁺ T cells restored tumor growth in Eif2ak3^fl/flLyz2-cre mice (Figure 3J), indicating the protective CD8⁺ T cell immunity status in tumor-bearing mice lacking PERK in myeloid cells.

To recapitulate our findings in therapeutic models, we next evaluated the anti-tumor effects of two small-molecule PERK inhibitors, GSK-2606414, a dual PERK and RIPK1 inhibitor (Rojas-Rivera et al., 2017), and AMG-44, a potent and selective PERK inhibitor (Smith et al., 2015). Delayed tumor growth was noticed in B16-bearing mice treated with GSK-2606414, which correlated with a lower immunoinhibitory activity of MDSC (Figure S3N–O). A primary limitation for the use of GSK-2606414 is the induction of pancreatic toxicity (Yu et al., 2015). Therefore, we tested the therapeutic action of two different doses of AMG-44. A similar delay in tumor growth, without alterations in the blood glucose concentration or in the size or insulin expression in the pancreatic islets were found in B16-bearing mice treated with both AMG-44 doses (Figure 3K, Figure S3P–R). Also, illustrating the therapeutic potential of AMG-44, treatment of B16-bearing mice with AMG-44 impaired the immunoregulatory activity of tumor-MDSC, induced expansion of tumor-infiltrating CD8⁺ T cells (TILs) expressing IFNγ and synergized with anti-PD-L1 therapy (Figure 3L–N). Moreover, the anti-tumor action of AMG-44 was detected in PERK-competent mice, but not in Eif2ak3^fl/flLyz2-cre counterparts (Figure S3S), suggesting the role of myeloid cell-PERK in the effects induced by AMG-44. Next, we tested the intrinsic effect of AMG-44 on Thaps-treated MDSC. Similar to Eif2ak3^−/− MDSC (Figure 3F), AMG-44 overcame the ability of Thaps-treated BM-MDSC to impair T cell proliferation, without affecting MDSC baseline activity (Figure 3O). Overall, these results demonstrate the key role of PERK in MDSC activity and the therapeutic potential of PERK inhibitors to overcome MDSC-linked T cell dysfunction.

PERK deletion reprograms tumor-MDSC into immune-stimulatory cells.

We evaluated whether the elimination of PERK transformed MDSC into cells that promoted anti-tumor T cell responses. Deletion of MDSC-like cells after continuous injections with anti-Gr1 or anti-CCL2 antibodies restored LLC tumor growth in Eif2ak3^fl/flLyz2-cre mice, while delaying tumor growth in Eif2ak3^fl/fl mice (Figure 4A–B). Also, co-injection of tumor-infiltrating Eif2ak3^−/− CD11b⁺ Gr1⁺ cells plus LLC cells into C57BL/6 mice (WT), but not into Rag1^−/− mice, resulted in slower tumor growth, compared to LLC cells implanted alone or mixed with control tumor-MDSC (Figure 4C). Next, we compared the expression of major inflammatory drivers in tumor M-MDSC and PMN-MDSC from LLC-bearing Eif2ak3^fl/flLyz2-cre and Eif2ak3^fl/fl mice. Higher expression of MHC-I (H-2K^b), CD40, and CD80; but not MHC-II, CD11c, and CD86 was found in PERK-null M-MDSC compared to controls, whereas no changes were noted on PMN-MDSC (Figure 4D). Moreover, elevated expression of anti-tumor cytokines, interleukin-12 (IL-12) and TNFα, was detected in tumor M-MDSC and PMN-MDSC from Eif2ak3^fl/flLyz2-cre mice (Figure 4E). Next, we tested whether the deletion of PERK allowed tumor-MDSC to process and present exogenous antigens. PERK-deficient tumor-MDSC pulsed with complete OVA showed higher ability to prime OT-I T cell proliferation, though lower than DCs, compared to control MDSC (Figure 4F). To further study the ability of Eif2ak3^−/− MDSC to engulf tumor products and to cross-present antigens in the TME, we used injections with Pan02-OVA-ZsGreen tumors, engineered to express fluorescence protein ZsGreen and OVA. Elevated ZsGreen and SIINFEKL-bound H-2K^b were detected on tumor-MDSC from Eif2ak3^fl/flLyz2-cre mice compared to controls (Figure 4G–H, Figure S4A), which correlated with an extension of OVA-reacting SIINFEKL-H-2K^b-tetramer⁺ CD8⁺ TILs (Figure 4I). PERK deletion promoted the expression of ZsGreen and SIINFEKL-bound H-2K^b in tumor M-MDSC and PMN-MDSC, but not in macrophages or myeloid DCs, indicating a preferential effect on MDSC subsets (Figure 4J–K). Also, in agreement with the ability of PERK-ablated MDSC to present engulfed tumor antigens to CD8⁺ T cells, tumor-CD11b⁺Gr1⁺ cells from Eif2ak3^fl/flLyz2-cre mice bearing Pan02-OVA-ZsGreen tumors, but not from control mice, induced ex vivo proliferation of OT-I CD8⁺ T cells (Figure 4L), but not of OT-II CD4⁺ T cells (Figure S4B). Consistently, heightened expression of SIINFEKL-bound H-2K^b and higher capacity to prime OT-I T cell proliferation were found in tumor-MDSC from TUDCA-treated mice bearing Pan02-OVA-ZsGreen tumors (Figure S4C–D). Thus, our data suggest that PERK deletion or UPR mitigation transform MDSC into cells that prime CD8⁺ T cell immunity in tumor beds.

PERK deletion transforms MDSC through NRF2 inhibition

To find the mediators by which PERK drives MDSC activity, we focused on CHOP, a downstream target of unmitigated cellular stress regulated by PERK, and previously linked to MDSC function (Thevenot et al., 2014). Reduced CHOP was detected in tumor-MDSC from Eif2ak3^fl/flLyz2-cre mice compared to MDSC from controls (Figure 5A). However, the decreased immunosuppressive activity found in PERK-null BM-MDSC treated with Thaps was not noted in Ddit3^−/− BM-MDSC (Figure 5B). Also, ectopic expression of Ddit3 (Figure S5A) failed to restore immunosuppressive function of Eif2ak3^−/− BM-MDSC after treatment with Thaps, while it enhanced the regulatory potential of vehicle-treated MDSC (Figure 5C). These results ruled out the impact of a reduced CHOP expression in the ablated immunosuppressive activity of PERK-deficient MDSC.

In light of the described phosphorylation of NRF2 by PERK in tumor cells (Cullinan et al., 2003), we next studied the mechanistic interaction between PERK and NRF2 in MDSC. Similar to the PERK upregulation in tumor-MDSC (Figure 1B), we noted higher nuclear NRF2 expression in tumor-MDSC from LLC-bearing mice, compared to splenic MDSC or iMC (Figure S5B). In addition, deletion of Nfe2l2 reduced the suppressive activity of tumor-MDSC to a similar extent as that induced by PERK elimination (Figure S5C). Also, consistent with the upstream role of PERK on NRF2 signaling, reduced expression of nuclear NRF2, lower binding of NRF2 to a consensus DNA sequence, and elevated ROS were found in tumor-MDSC and Thaps-treated BM-MDSC lacking PERK, compared to controls (Figure 5D–F, Figure S5D–E). Next, we validated the relevance of a thwarted NRF2 signaling in the lower regulatory function of Thaps-treated Eif2ak3^−/− MDSC. Treatment with the indirect NRF2-inducing agents, Resveratrol or Sulforaphane (SFRN), partially restored the potential of Thaps-treated Eif2ak3^−/− BM-MDSC to impair T cell proliferation (Figure 5G). Consistently, enforced expression of NRF2-ΔNeh2, a dominant active form of NRF2 (Shin et al., 2007) (Figure S5F), restored the immunosuppressive activity and lowered ROS content in Eif2ak3^−/− BM-MDSC treated with Thaps, compared to the same cells carrying a control vector (Figure 5H, Figure S5G). Also, treatment with SFRN overcame the anti-tumor effects found in Eif2ak3^fl/flLyz2-cre mice, while blunting tumor growth in Eif2ak3^fl/fl mice (Figure 5I). Moreover, treatment of LLC-bearing Eif2ak3^fl/flLyz2-cre mice with SFRN reduced ROS amounts and restored the immunoregulatory potential of tumor-MDSC, without affecting controls (Figure 5J–K). These results support the role of a reduced NRF2 activity in the effects found in PERK-deficient MDSC.

PERK regulates mitochondrial homeostasis in MDSC through NRF2

We elucidated whether the deletion of PERK impacted mitochondrial function in tumor-MDSC. Altered mitochondria morphology, diminished mitochondrial mass, and disrupted mitochondrial membrane potential were found in tumor-MDSC from Eif2ak3^fl/flLyz2-cre mice, compared to controls (Figure 6A–C). Also, in comparison to wild type MDSC, tumor-MDSC or Thaps-treated BM-MDSC from PERK-null mice had a lower mitochondrial respiratory activity, as suggested by a reduced oxygen consumption rate (OCR) upon mitochondrial stress studies (Figure 6D, Figure S6A). However, no changes in the extracellular acidification rate (ECAR) were found between tumor-MDSC from Eif2ak3^fl/flLyz2-cre and control mice after glycolysis stress profiles (Figure S6B), indicating that PERK deletion altered mitochondrial function rather than glycolysis. Next, we identified the role of a decreased NRF2 in the mitochondrial alterations found in PERK-null MDSC. Enforced NRF2-ΔNeh2 expression restored mitochondrial respiratory activity, as tested by OCR, in Eif2ak3^−/− BM-MDSC treated with Thaps (Figure 6E), without inducing variations in vehicle-treated MDSC (Figure S6C). Moreover, tumor-MDSC from LLC-bearing Eif2ak3^fl/flLyz2-cre mice treated with SFRN had elevated OCR, compared to tumor-MDSC from untreated counterparts or from treated or untreated Eif2ak3^fl/fl mice (Figure 6F). Thus, lower NRF2 signaling promotes mitochondrial respiratory dysfunction in PERK-ablated MDSC.

Because of the emerging role of the cytosolic mtDNA as a mediator of innate immunity (West et al., 2015) and the contribution of NRF2 signaling in maintaining mitochondrial integrity, we then assessed the presence of the mtDNA genes, Cox1 and Nd1, in mitochondria enriched and depleted cellular extracts from tumor-MDSC. Elevated mtDNA in mitochondrial-free cytosolic extracts and reduced mtDNA in mitochondrial enriched extracts were detected in tumor-MDSC from Eif2ak3^fl/flLyz2-cre mice compared to controls (Figure 6G–H). Similarly, PERK deficiency induced mtDNA accumulation in mitochondrial-free cytosolic extracts from BM-MDSC treated with Thaps, but not from non-treated BM-MDSC (Figure S6D–E). Moreover, depletion of mtDNA with ethidium bromide (EtBr) (Wang et al., 2018) partially restored suppressive activity of Eif2ak3^−/− BM-MDSC treated with Thaps (Figure 6I), supporting the role of the amplified cytosolic mtDNA in the reprogramming of UPR-active PERK-deficient MDSC.

STING-dependent induction of Type I IFN drives effects of PERK ablation

The presence of mtDNA in the cytosol promotes STING pathway stimulation (Sliter et al., 2018). Consistently, heightened STING signaling, as indicated by a higher phosphorylation of TBK1 and nuclear IRF3, was detected in tumor-MDSC from Eif2ak3^fl/flLyz2-cre mice compared to controls (Figure 7A). Next, we identified the role of STING in the anti-tumor effects induced by PERK deletion, using myeloid cell PERK and STING null mice (Eif2ak3^fl/flTmem173^fl/flLyz2-cre). Dual elimination of PERK and STING in myeloid cells restored tumor growth and ability of tumor-MDSC to blunt T cell proliferation, compared to Eif2ak3^fl/flLyz2-cre or Tmem173^fl/flLyz2-cre mice (Figure 7B–C). Also, PERK-STING-deficient mice showed lower infiltration of total and IFNγ-producing CD8⁺ TILs compared to Eif2ak3^fl/flLyz2-cre mice (Figure 7D, Figure S7A). Moreover, deletion of STING in PERK-null mice bearing Pan02-OVA-ZsGreen tumors limited the expansion of anti-SIINFEKL T cell responses and ablated the capacity of tumor-MDSC to induce ex vivo proliferation of OT-I CD8⁺ T cells (Figure 7E–F). Furthermore, consistent with STING activation, we found elevated expression of Type I IFN-related transcripts, Ifnb1, Isg15, Ifit3, and Cxcl10, in tumor-MDSC from Eif2ak3^fl/flLyz2-cre mice, which was attenuated after STING deletion (Figure 7G). The increased Ifnb1 mRNA noticed in tumor-MDSC from Eif2ak3^fl/flLyz2-cre mice was not found in macrophages or myeloid DCs, indicating a preferential effect on MDSC (Figure S7B). Next, we tested the role of the Type I IFN in the anti-tumor effects noted in Eif2ak3^fl/flLyz2-cre mice. Antibody-based blockade of IFNα/β receptor subunit 1 (IFNAR1) restored tumor growth in Eif2ak3^fl/flLyz2-cre mice compared to IgG-treated counterparts (Figure 7H). Thus, activation of STING-driven Type I IFN drives the anti-tumor effects induced by myeloid cell-PERK deletion.

To define the mediators by which PERK deletion triggered the production of Type I IFN, we tested the impact of the amplified cytosolic mtDNA induced by reduced NRF2 signaling. Depletion of cytosolic mtDNA with EtBr prevented the Ifnb1 mRNA induction in Eif2ak3^−/− BM-MDSC treated with Thaps (Figure 7I). Moreover, similar to the effects found in Eif2ak3^−/− MDSC, higher cytosolic mtDNA and Ifnb1 mRNA were detected in tumor-MDSC from Nfe2l2^−/− mice compared to those from wild type mice (Figure S7C–D). Consistently, NRF2-ΔNeh2 overexpression prevented the induction of Ifnb1 mRNA in Thaps-treated Eif2ak3^−/− BM-MDSC, compared to controls (Figure 7J). Also, tumor-MDSC from Eif2ak3^fl/flLyz2-cre mice treated with SFRN showed reduced Ifnb1 mRNA, compared to those from untreated counterparts (Figure 7K). Overall, the results show the crosstalk between reduced NRF2 signaling, cytosolic mtDNA, and STING-dependent production of Type I IFN as mediators for the effects induced by PERK deletion in MDSC.

CONTACT FOR REAGENT AND RESOURCE SHARING

The authors declare that all the results supporting the findings of this study are available within the paper and its Supplemental Figures. Further information and requests for supporting data and resources should be directed to and will be fulfilled upon reasonable request by the Lead Contact, Paulo C. Rodriguez, PhD ([email protected]).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

METHOD DETAILS

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analysis was performed using GraphPad Prism 7 (San Diego, CA). For two groups, means were compared by two-tailed unpaired Student’s t-test. For multiple groups and tumor growth studies, analysis was done by ANOVA with Bonferroni correction for multiple comparisons. P value of <0.05 was considered statistically significant. Specific statistical test results are indicated in each figure: *, p<0.05; **, p<0.01 ***, p<0.001.