Cancer cells respond to hypoxia by upregulating the hypoxia-inducible factor 1α (HIF1A) transcription factor, which drives survival mechanisms that include metabolic adaptation and induction of angiogenesis by VEGF. Pancreatic tumors are poorly vascularized and severely hypoxic. To study the angiogenic role of HIF1A, and specifically probe whether tumors are able to use alternative pathways in its absence, we created a xenograft mouse tumor model of pancreatic cancer lacking HIF1A. After an initial delay of about 30 days, the HIF1A-deficient tumors grew as rapidly as the wild-type tumors and had similar vascularization. These changes were maintained in subsequent passages of tumor xenografts in vivo and in cell lines ex vivo. There were many cancer cells with a "clear-cell" phenotype in the HIF1A-deficient tumors; this was the result of accumulation of glycogen. Single-cell RNA sequencing (scRNA-seq) of the tumors identified hypoxic cancer cells with inhibited glycogen breakdown, which promoted glycogen accumulation and the secretion of inflammatory cytokines, including interleukins 1β (IL1B) and 8 (IL8). scRNA-seq of the mouse tumor stroma showed enrichment of two subsets of myeloid dendritic cells (cDC), cDC1 and cDC2, that secreted proangiogenic cytokines. These results suggest that glycogen accumulation associated with a clear-cell phenotype in hypoxic cancer cells lacking HIF1A can initiate an alternate pathway of cytokine and DC-driven angiogenesis. Inhibiting glycogen accumulation may provide a treatment for cancers with the clear-cell phenotype.
These findings establish a novel mechanism by which tumors support angiogenesis in an HIF1α-independent manner.
Adaptation to hypoxia in pancreatic adenocarcinoma is mediated through the stabilization and activation of the hypoxia-inducible factor (HIF) family of transcription factors. HIF1α (hereafter HIF1A) drives the transcriptional activation of multiple pathways, including a metabolic switch to promote glycolysis and the expression of various proangiogenic cytokines including Vegf-A (VEGFA), which act on surrounding endothelial cells to induce the formation of new blood vessels (1).
A previous study explored the dependence of tumor growth on HIF1A by selective deletion in a genetic model of pancreatic cancer (2). In that model, deletion of HIF1A promoted the formation of pancreatic intraepithelial neoplasms (PanIN) precursor lesions, but the role of HIF1A in tumor maintenance or growth was not explored. As HIF1A is essential to adaptation of cells to low oxygen, we set out to investigate the adaptive mechanisms cells may use to circumvent loss of HIF1A. Furthermore, due to the role of HIF1A in promoting angiogenesis, and the dependence of VEGFA expression on HIF1A (3), we discovered alternative proangiogenic pathways that could be important in cancers that become resistant to therapies targeting VEGFA such as bevacizumab, and in other diseases where anti-VEGFA therapy is used, but likewise encounter high rates of resistance (4).
Here, we studied the role of HIF1A in tumor growth and maintenance in pancreatic cancer using a xenograft model of MiaPaCa-2 cells with stable knockdown of HIF1A. We explored the tumor growth kinetics and observed a long delay in the formation of tumors, followed by a rapid growth that was maintained through several tumor passages. Characterization of the tumors using multispecies single-cell RNA-seq (scRNA-seq), histology, and immunostaining indicated a dramatic accumulation of glycogen in tumors lacking HIF1A. This was found to be a result of the dependence on HIF1A for glycogen breakdown, leading to an abundance of intracellular glycogen. We associated this glycogen accumulation with a proinflammatory signature in pancreatic cancer cells, characterized by several immunoattractant cytokines, including IL1β and IL8. These tumor-derived cytokines attracted myeloid dendritic cells (cDC) of subtypes 1 and 2, which in the context of the tumor microenvironment, acquired a proangiogenic phenotype. This work indicates a novel nonmetabolic role for glycogen accumulation in driving a proinflammatory program, and sustaining solid tumor growth in the absence of HIF1A via cDC recruitment and proangiogenic cytokine release.
Materials and Methods
Generation of shHIF1A and EV lines
shRNA targeting HIF1A or glycogen synthase-1 GYS1 (Supplementary Fig. S1) were cloned into pSUPER backbone. MiaPaCa-2 cells (ATCC) with stable expression of hypoxia-response element (HRE)/luciferase, under neomycin selection, were transfected with either pSUPER-HIF1A or pSUPER empty vector (EV). Cells were selected using 2 μg/mL of puromycin for several passages, then single-cell sorted to establish clonal cultures. HIF1A knockdown was verified by Western blot and decreased expression of downstream HIF1A genes (Supplementary Fig. S2). Cell lines were routinely tested to be Mycoplasma free, and the identity of each line was authenticated at 2-month intervals while in culture by the Genomics Shared Resource at SBP. Single-cell separation for scRNAeq is described in Supplementary Methods S1. siRNA when used was Dharmacon SMARTpool used at 20–100 nmol/L and validated by protein knockdown.
Five- to six-week-old Nod-scid (NOD.CB17-Prkdcscid/J) and NOD-scid gamma (NOD-scid IL2Rgnull) mice were obtained from the Jackson Laboratories. Tumor cells (107) were injected into the flanks in 0.9% sterile saline. Tumor and body weight was measured twice weekly. All studies beyond primary tumors (Fig. 1A) were on tumors reimplanted once and cell lines derived therefrom. All animal studies were SBP ACUC approved.
Following tumor dissociation, CD45+ cells were magnetically positively selected using the Miltenyi Biotec LS column. Samples were stained using a panel of antibodies against tumor-infiltrating leukocytes. Following staining, samples were run on the MACSQuant (Miltenyi Biotec) flow cytometer.
Extracellular flux assay for glycolysis
The XF Glycolysis Stress test was used according to the manufacturer protocol on the Seahorse Bioscience XF96e instrument, to measure the rate of lactate formation and oxygen utilization.
HIF1A-deficient cancer cells form fast growing tumors after a delay period
EV and shHIF1A cells were injected in the flanks of Nod-scid mice. Tumors formed by EV cells grew rapidly to 1,000 mm3 within 30 days, while the shHIF1A tumors persisted for up to 50 days before growing with kinetics that matched the EV tumors (Fig. 1A). Only cells from EV tumors showed HRE promoter activation (Fig. 1B). Cells obtained from EV and shHIF1A tumors were propagated in vitro, and used to establish second-generation tumors. The second-generation shHIF1A tumors grew after only a short delay with growth kinetics that matched the EV tumors, indicating that HIF1A-independent growth was maintained ex vivo (Fig. 1C). IHC characterization of these tumors (Fig. 1D and E) showed no specific staining for HIF1A, thus excluding restoration of HIF1A as a mechanism of increased growth. There was no difference in apoptosis between EV and shHIF1A tumors, as measured by TUNEL staining and no difference in gross necrosis. Extracellular matrix by collagen I/III staining was increased in the shHIF1A tumors, as was staining for α-smooth muscle actin, a marker for cancer-associated fibroblasts, tumor hypoxia measured by pimonidazole staining was increased (Supplementary Fig. S3), while staining for endothelial cell marker CD31 showed that shHIF1A tumors were equally vascularized as the EV tumors. Overall, these data indicate that despite an initial delay in growth, HIF1A-deficient tumors adapt and grow rapidly having normal angiogenesis despite increased hypoxia, and no increase in apoptosis.
Tumors lacking HIF1A have a clear-cell phenotype characterized by intracellular glycogen accumulation
Gross histochemical analysis of the tumor samples stained with Masson's trichrome showed a marked difference between the EV and shHIF1A tumors, with shHIF1A tumors showing an abundance of clear cells throughout the whole tumor (Fig. 2A). Oil Red O staining of the shHIF1A tumors was negative, suggesting that the clear cells did not contain abundant lipids (Supplementary Fig. S3), while Periodic acid–Schiff (PAS) staining was positive, indicating that the cells contained accumulated polysaccharides (Fig. 2B). To differentiate between glycogen, glycoproteins, and mucins, PAS staining was performed in conjunction with diastase (PAS-D), an enzyme that specifically digests glycogen. Diastase caused only marginal lightning of PAS staining in EV tumors, but completely cleared the staining in shHIF1A tumors, indicating that the clear cells in shHIF1A tumors contain mainly glycogen. To validate these findings, glycogen content within the tumors was measured enzymatically, confirming a dramatic increase in glycogen content in the shHIF1A tumors (Fig. 2C). To explore a potential metabolic role for increased glycogen content, the glucose flux into the cells and its utilization by glycolysis was measured using cell lines established from EV and shHIF1A tumors. Glucose uptake was found to be slightly decreased in the shHIF1A cells compared with EV cells, although not statistically significant (P = 0.07; Fig. 2D). A Seahorse extracellular flux assay was used to measure glycolysis, showing that under low oxygen conditions (2% O2) the glycolytic capability of shHIF1A cells was dramatically decreased compared with EV cells (Fig. 2E) with an increase in mitochondrial respiration (Fig. 2F). These data indicate that an accumulation of glycogen is associated with slightly decreased glucose uptake, and a marked decrease in glucose utilization by glycolysis in the shHIF1A tumors, suggesting a nonmetabolic role of glycogen in these tumors. The source of energy metabolism in these cells remains to be determined.
scRNA-seq identifies hypoxic cell populations in tumors
In order to elucidate the mechanism underlying the accumulation of glycogen in shHIF1A tumors, and to gain insights into the pathways being utilized by the tumor to bypass the lack of HIF1A, scRNA-seq was utilized (Fig. 3A). Following sequencing and downstream analysis, 6 well-defined clusters were obtained from the bioinformatics pipeline (Fig. 3B). Three clusters were found to express hypoxic genes, based on a hypoxia signature that includes non-HIF1A genes (5), with three different levels of expression: “low,” “medium,” and “high.” For analysis purposes, these three clusters were grouped into a single “hypoxic” group, while all others were grouped “normoxic.” Comparison of the hypoxic and normoxic clusters using a GSEA signature for hypoxia validated the classification of the cells (Fig. 3C). Within these clusters, it was possible to compare the expression of transcripts between cells from EV and shHIF1A tumors. Expression of glycolytic enzymes, broadly reported to be HIF1A dependent, was enriched in the EV cells within the hypoxic clusters, and repressed in the shHIF1A cells (Fig. 3D).
HIF1A-deficient tumors show suppressed glycogen breakdown at the transcriptional level
To elucidate the mechanism of glycogen accumulation in the shHIF1A cells, clusters from the scRNA-seq were analyzed for expression of key glycogen synthesis and breakdown enzymes. A heat map was generated of the relative expression values within normoxic and hypoxic clusters in EV and shHIF1A tumors showing increased glycogen synthesis and breakdown enzymes in hypoxia, in both the EV and shHIF1A tumors (Fig. 3E). When analyzing the hypoxic groups only, and comparing the shHIF1A to EV cells, the synthetic glycogen enzyme GYS1 was found to remain elevated in the shHIF1A groups. However, glycogen branching by GBE1, debranching by AGL, and breakdown as regulated by the PYGL cofactors PHKA1, PHKA2, and PHKB, were decreased in the shHIF1A tumors (6). Figure 3F depicts the glycogen synthesis and breakdown in the shHIF1A tumors, in hypoxia. This transcriptional pattern suggests a mechanism by which, in the absence of HIF1A, glucose is converted into glycogen, resulting in the accumulation of glycogen in shHIF1A cells. We next investigated the effect of GYS1 knockdown on tumor formation in Nod-scid mice finding that by itself it had only a small effect delaying tumor growth, less than by HIF1A inhibition, while both GSY1 and HIF1A inhibition completely prevented tumor formation (Fig. 3G and Supplementary Fig. S4). Taken together, the results suggest that an increase in glycogen in HIF1A-deficient cells, which gives rise to a clear-cell phenotype, is a key factor in the ability of the tumors to overcome dependence on HIF1A for growth.
HIF1A-deficient cancer cells express a proinflammatory signature that is dependent on glycogen
In order to explore the mechanism by which increased glycogen accumulation could contribute to HIF1A-independent tumor growth, we analyzed the signaling pathways upregulated in shHIF1A cells. To this end, bulk RNA-seq was performed on two independently derived resistant shHIF1A tumors, and two EV tumors. Pathway analysis of the top transcripts indicated a significant upregulation of various inflammation-associated transcripts, including IL1A, IL1B, CXCL8 (IL8), and GCSF, as well as various MMP family members (Fig. 4A), consisting an inflammatory signature present solely in the shHIF1A tumors. Indeed, the most upregulated member of this signature, IL1B, was also shown by IHC to be only expressed at the protein level in the shHIF1A tumors (Fig. 4B). Importantly, IL1B remained upregulated during ex vivo propagation of the cells, and continued to be expressed in subsequent xenograft experiments (Supplementary Fig. S5). To understand the upstream driver of this transcriptional pattern in cancer cells, pathway analysis was used, which predicted that IL1B and NF-kB were activated (Fig. 4C).
Given a previous report of the mechanistic link between glycogen accumulation and a proinflammatory stress (7), the connection between the transcriptional regulation of the glycogen enzymes and the inflammatory signature was further explored. siRNA-mediated knockdown of members of the glycogen synthesis and breakdown pathway was performed, with IL1B used as a readout for the proinflammatory signature in cancer cells. Inhibition of the glycogen synthesis and branching enzymes GYS1 and GBE1 resulted in a dramatic decrease in IL1B expression, both in the EV and shHIF1A cells, while inhibition of the glycogen breakdown and debranching enzymes PYGL1 and AGL had no effect (Fig. 4D). However, IL1B antibody given to the mice had no effect on tumor growth, suggesting multiple cytokines may be involved. These data indicate that the shHIF1A tumors acquire a proinflammatory transcriptional signature, most likely as a result of the accumulation of intracellular glycogen.
Increased transcriptional myeloid signature correlates with DC infiltration
To explore the interplay between tumor and stroma, upstream regulator analysis was performed on the bulk RNA-seq of the mouse tumor stroma transcriptome. From this pathway analysis, a number of predicted positive regulators of mouse transcription included several top hits from the human transcriptome: IL1A, IL1B, CXCL8, and IL6 (Fig. 5A). This suggests that the transcriptional changes in the mouse stroma are in part caused by the human cytokines secreted by the tumor. Because a number of the proinflammatory cytokines that were expressed by the tumor are thought to play a role in immune cell recruitment, the mouse stroma transcripts were analyzed for evidence of immune pathway activation with the caveat that NOD-scid mice are deficient in a number of immune cells, including B and T cells. A number of pathways related to myeloid cells were found to be upregulated in the shHIF1A tumors, including chemotaxis and activation, and specifically myeloid DC were predicted to have migrated into shHIF1A tumors (Fig. 5B). To validate this finding, flow cytometry was used to quantify the percentage of a panel of tumor-infiltrating lymphocytes (TIL) within EV and shHIF1A tumors (Fig. 5C and Supplementary Table S1). Consistent with previous studies on the role of HIF1A in pancreatic cancer (2), the shHIF1A tumors displayed an increased percentage of CD45+ cells, and consistent with the predictions from the bulk RNA-seq and pathway analysis, CD11c+ cDC, identified as CD11b+ CD11c+ Class II+ Ly-6C/G− cells were identified. Next, we used Nod-scid gamma mice whose CD11c+ DCs are deficient in cytokine production compared with Nod-scid mice (8), and found that implanted siHIF1A MiaPaCa-2 cells took significantly longer to begin to form tumors, 54 days (Fig. 5D), than we had seen for Nod-scid mice, 28 days, P = < 0.01 (Figs. 1A and 3G). Taken together, these results suggest that human cytokines secreted by the HIF1A inhibited tumor cells lead to the recruitment of mouse immune cells into the tumor, the most significant being DCs that contribute to tumor formation in the absence of HIF1A.
Single-cell analysis of stroma unveils an angiogenic signature in cDC1 and cDC2 DC subtypes
Because of a previously reported role for cDC in promoting angiogenesis (9), and the absence of HIF1A/VEGF-driven angiogenesis in the shHIF1A tumors, the increase in cDC could be responsible for allowing the shHIF1A tumor to become vascularized. To understand if cDC were responsible for an angiogenic stimulus, the mouse component of the xenograft scRNA was annotated using a panel of transcript data for various cell types found within the stroma (Supplementary Table S2), and two clusters were identified as cDC. Further stratification based on transcription allowed for the identification of DC subtypes cDC1 and cDC2 (Fig. 5E and Supplementary Table S3; ref. 10). Angiogenic transcript expression was analyzed within these two subtypes and compared between the EV and shHIF1A tumors and indicated that both cDC1 and cDC2 subtypes expressed more angiogenic transcripts in the shHIF1A tumors versus the EV (Fig. 5F). A functional HUVEC assay was then performed to quantify the angiogenic potential of the CD11c+ cells found within the shHIF1A tumors. The CD11c+ fraction of the tumor was significantly more proangiogenic than either tumor cells alone, or the CD11c+ fraction from the spleen (Fig. 5G). These data indicate that host CD11c+ cells found in tumors can promote angiogenesis in the absence of cancer cell–driven angiogenesis.
Analysis of TCGA data for glycogen accumulation and patient survival
A search of The Cancer Genome Atlas (TCGA) was performed in order to explore if glycogen accumulation in cancer cells could be related to patient survival. The PANCAN database of 9,184 solid cancers was compared with the glycogen pathway members identified in this study as contributing to glycogen accumulation. As previously reported for hematologic malignancies (11), elevated expression of GYS1 and branching enzyme gene GBE1 expression was correlated with poor patient survival (Fig. 6A). In contrast, expression of the glycogen phosphorylase cofactor genes PHKA1, PHKA2, and PHKB, as well as the debranching enzyme gene AGL, was associated with improved survival. A signature for glycogen accumulation was therefore developed based on the increased glycogen accumulation observed in our model system, with upregulated GYS1/GBE and downregulated PHKA1/PHKA2/PHKB/AGL (Fig. 6B). By combining synthesis and breakdown, a signature that favors glycogen accumulation in pan-TCGA data were correlated with shorter patient survival, consistent with our findings implicating glycogen accumulation in tumor growth. However, for individual tumor types, only ovarian cancer showed a significant correlation and pancreatic cancer was nonsignificant (Fig. 6C and D). Finally, TCGA was used to show a negative correlation between the glycogen accumulation signature and an HIF1A activation signature (Fig. 6E).
HIF1A has long been thought to play an essential role in cancer cells’ tolerance and adaptation to hypoxic stress through activities such as promotion of glycolysis, and the formation of angiogenic factors (e.g., VEGFA; refs. 12, 13). We found that inhibition of HIF1A in a transplanted human pancreatic cancer cell mouse xenograft model, after initially showing growth, ultimately returned to the same growth rate of the parental cell line. This effect was maintained through subsequent implantation and in cell lines. The HIF1A inhibited tumor cells acquired a “clear-cell” phenotype due to accumulation of glycogen associated with loss of HIF1A-dependent glycogenolytic enzymes, while glycolysis was inhibited. At the same time, the formation of immunoattractant inflammatory cytokines led to accumulation in the tumor microenvironment of conventional DCs that through release of proangiogenic factors sustained the tumor vascularization necessary for tumor growth.
Glycogen accumulation is observed in various cancers, but is most well established in renal clear-cell carcinoma. Here, loss of the Von Hippel–Lindau tumor suppressor E3 ubiquitin ligase prevents both HIF1A and HIF2A degradation allowing elevated levels and activity in both air and hypoxia. However, the interaction between the two HIFs in driving renal clear-cell carcinoma growth, whether as tumor suppressors or activators remains unclear (14). In ovarian clear-cell carcinoma, HIF1A is thought to be responsible for the accumulation of glycogen by increasing the expression of the glycogen synthesis enzyme GYS1 (15). Although clear-cell carcinomas of the pancreas have been observed, this phenotype is not typically associated with increased intracellular levels of glycogen (16). However, glycogen is seen in serous microcystadenoma, a benign or indolent form representing 5% to 10% of pancreatic neoplasms (17, 18). As pancreatic cancer can express both HIF1A and HIF2A (19), it was for simplicity that we chose to study a cell line that expresses only HIF1A.
Currently, there is no definitive mechanistic connection between glycogen accumulation and the transcription of proinflammatory genes observed in our pancreatic cancer model. Hypoxia is known to activate the unfolded protein response (20), which can increase tumor growth through the formation of proinflammatory cytokines (21). It is thus possible that glycogen accumulation could further enhance this pathway (22). The inflammatory signature expressed by the shHIF1A tumor cells was correlated with an increase in myeloid DC accumulation, which our findings suggest results in a proangiogenic phenotype. Although NOD-scid mice used in our study of tumor growth lack a number of immune cells, including B and T cells, they do have DCs (23). This infiltration of DCs in tumors has been reported (24), and their role in promoting angiogenesis is established (25). Notably, in Nod-scid gamma mice that have dysfunctional DCs with decreased cytokine release (8), the appearance of shHI1A tumors was considerably delayed compared with Nod-scid mice, consistent the importance of DCs to tumor formation. In immune-competent animals, it is possible that other immune cells may play a similar role.
A significant finding from our studies is that of apparently normal angiogenesis in the absence of VEGFA expression in the shHIF1A xenograft. Although a number of agents targeting the VEGFA pathway are approved for use in cancer and in ocular disorders, tumor inhibition and ocular improvement is short-lived due to acquired resistance in the form of either VEGFA target and receptor overexpression, or a shift to other proangiogenic factors (26). In this work, we identified several proangiogenic cytokines secreted by DCs including PROK2/Bv8, whose secretion by CD11c+ cells was previously reported to be involved in VEGFA-independent angiogenesis, and speculated to drive resistance to bevacizumab (27). Thus, the tumor-inflammatory cytokine/stroma angiogenic cytokine mechanism we identified may help explain how VEGFA-independent angiogenesis is possible. Finally, an analysis of the TCGA database indicates that our findings may have clinical relevance. A pan-cancer analysis of more than 9,000 patients showed a correlation between patients with low HIF1A activation and a signature for glycogen accumulation.
Thus, key enzymes involved in the glycogen pathway may provide targets for drug discovery to modulate the elevated glycogen phenotype, which patient data suggest may be correlated with decreased survival, and as a way to overcome resistance to anti-VEGF therapy.
Disclosure of Potential Conflicts of Interest
M.Y. Koh has ownership interest (including patents) in Kuda Therapeutics, Inc. No potential conflicts of interest were disclosed by the other authors.
Conception and design: M. Maruggi, B.P. James, F. Soldevilla, P.R. de Jong, M.Y. Koh, G. Powis
Development of methodology: M. Maruggi, B.J. Baaten, M.Y. Koh, G. Powis
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Maruggi, F.I. Layng, R. Lemos Jr, B.P. James, F. Soldevilla, B.J. Baaten, G. Powis
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Maruggi, F.I. Layng, B.P. James, B.J. Baaten, P.R. de Jong, M.Y. Koh, G. Powis
Writing, review, and/or revision of the manuscript: M. Maruggi, F.I. Layng, P.R. de Jong, G. Powis
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R. Lemos Jr, G. Powis
Study supervision: P.R. de Jong, M.Y. Koh
Other (histology): G. Garcia
This study was supported by NIH grant 5F31CA203286 (M. Maruggi), CA216424 (G. Powis), and CCSG grant P30CA030199. The help of the SBP Cancer Center Animal, Genomics, Histology, and Flow Cytometry Services is gratefully acknowledged.
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.