Macrophage HIF-1α Is an Independent Prognostic Indicator in Kidney Cancer

Clear cell renal cell carcinoma (ccRCC) is the most common and aggressive form of kidney cancer and is also one of the most resistant of solid tumors to conventional chemotherapy (1). The etiology of ccRCC is uniquely linked to loss of the von Hippel–Lindau (VHL) tumor suppressor, which occurs in more than 90% of ccRCC (2). pVHL loss results in the constitutive stabilization of HIF-1α and HIF-2α that drive the expression of hundreds of genes associated with tumor progression (3–6). Despite sharing many transcriptional targets, the HIFs also drive unique transcriptional targets. For example, anerobic glycolysis appears to be predominantly controlled by HIF-1α, whereas erythropoietin synthesis is largely HIF-2α–regulated (7, 8). Hypoxia- or pVHL-independent HIF-α stabilization can also occur through increased mRNA transcription and/or translation, such as through cytokine- and growth factor receptor–mediated signaling, particularly within inflammatory mediators such as macrophages (9). We have previously proposed the notion of a HIF switch, which orchestrates the complementary roles of HIF-1α and HIF-2α during acute versus chronic hypoxia, and during vascular and bone development in the embryo (10). The hypoxia-associated factor (HAF), acts as a mediator of the HIF switch, by ubiquitinating and degrading HIF-1α, while enhancing HIF-2α transactivation through hypoxia-dependent HAF SUMOylation (11–13).

The current paradigm for the HIFs in ccRCC describes a driving role for HIF-2α and a tumor suppressor role for HIF-1α that is unique to ccRCC. This is based on studies showing that overexpression of HIF-2α promotes, whereas overexpression of HIF-1α inhibits the growth of ccRCC xenografts in mice (14). Consistent with this view, homozygous deletion of HIF1A has been reported in approximately 50% of high-grade ccRCC, implicating HIF-1A as a ccRCC tumor suppressor gene (15). Although lost in high-grade ccRCC, HIF-1α is believed to play an important role during ccRCC initiation and is elevated in the earliest preneoplastic lesions in VHL patients. The appearance of HIF-2α protein in these early lesions is associated with increased dysplasia and cellular atypia (16, 17). The HIF-2 bias observed in ccRCC has been attributed to the increased potency of HIF-2α compared with HIF-1α in promoting expression of pro-tumorigenic factors such as cyclin D1, TGFα, and VEGFA (14, 18). HIF-2α also potentiates protumorigenic c-Myc and β-catenin activity, which are inhibited by HIF-1α (19, 20). Consequently, there has been increasing enthusiasm for the selective targeting of HIF-2α in ccRCC (21–24). However, HIF-1α protein has been detected via IHC in most ccRCC cases, which is inconsistent with its proposed tumor suppressor role (15, 25, 26). In view of these conflicting observations, we recognized a need for additional studies to clarify the involvement of the HIFs in ccRCC progression (14, 15, 19, 27). Here, we report a previously unanticipated role of HIF-1α in tumor-associated macrophages (TAMs), particularly in M2-polarized TAMs, as a marker of ccRCC morbidity, whereas ccRCC cell-specific HIF-2α and HAF levels indicate better patient prognosis.

Studies were conducted in accordance with the Declaration of Helsinki. Human ccRCC and uninvolved kidney tissue were obtained from archival samples from patients who had provided written informed consent according to protocols approval by Institutional Review Boards at their respective institutions. Fresh tumor and uninvolved kidney for single-cell sequencing mRNA-sequencing (scRNA-seq) were obtained after informed consent from patients undergoing nephrectomies at Huntsman Cancer Institute (HCI, Salt Lake City, UT), approved under IRB No. 67518 and No. 10924. A separate cohort of 57 patients with metastatic ccRCC from HCI who were treated with antiangiogenic therapy were retrospectively analyzed for HIF-1/2α protein levels and correlated to radiographic disease progression as determined by the local radiologist. Brief tissue descriptions: Analytic cohort: Multistage TMA [M.D. Anderson Cancer Center (MDACC, Houston, TX, 380 patients), stage I tissue (MDACC, 12 patients)]. Validation cohort: Multigrade TMA (Mayo Clinic, 24 patients), multigrade sections (Mayo Clinic, 25 patients). Immune analysis: multigrade sections (HCI, 23 patients). More detailed descriptions are provided in Supplementary Methods.

TMAs and tissue were stained for HAF, HIF-2α, and HIF-1α after initial antibody workup and validation (Supplementary Figs. S2 and S8). mAb for HAF has been previously described, whereas all other antibody details, staining and statistical methods are described in Supplementary Methods (13).

Tumors were dissociated using the Human Tumor Dissociation Kit (Macs Miltenyi Biotec) according to the manufacturer's guidelines. Three ccRCC and paired uninvolved kidney samples showing >40% cell viability were selected for sequencing. Assessment of gene expression in single cells was achieved using the 10X Genomics Chromium single-cell gene expression solution (10x Genomics). Sequencing data were processed using the 10X Genomics Cell Ranger pipeline, which uses the Spliced Transcripts Alignment to a Reference (STAR) aligner for genome and transcriptome alignment. Further processing was conducted using the Seurat (3.0.0) package for R (3.5.3). Further detail is provided in Supplementary Methods.

This was performed to assess RNA levels of HIF1A, NDUFA4L2, CD163, and SLC2A1 using RNAscopeMultiplex Fluorescent Assay V2 (Advanced Cell Diagnostics). Staining of tissue using Opal 520, Opal 570, Opal 620, Opal 690 fluorescent dyes (Akoya Biosciences) was performed according to the manufacturer's guidelines. Confocal images were captured on a Leica SP8 confocal microscope equipped with UV, 488 nm, and white lasers and using an oil emersion 40× objective. For quantification, the average number of HIF1A or SLC2A1 mRNA molecules present within a 5 μmol/L radius of NDUFA4L2 and CD163 molecules was determined. At least 200 NDUFA4L2 and CD163 molecules were assessed across all patients within multiple fields of view. Statistical significance was determined by t test using GraphPad Prism.

THP-1, ACHN, Caki-1, A498, and 786-0 cells were from ATCC, whereas RCC4, RCC10, UMRC2, and UOK101 were kind gifts from M. Celeste Simon (University of Pennsylvania, Philadelphia, PA). All commercially available cell lines were authenticated using STR fingerprinting upon receipt and stored in frozen aliquots. Fresh vials were thawed for use in experiments and discarded after 30 to 35 passages or approximately 2 months. Mycoplasma testing using the MycoAlert Detection Kit (Lonza) was performed every 2 months with the last test performed on February 2020. Cells were maintained at 37°C 5% CO₂ in DMEM (Life Technologies, Thermo Fisher Scientific) or RPMI1640 (THP-1 cells) with 10% FBS. Hypoxia incubations were performed using a Whitley H35 Hypoxystation (HypOxygen). THP-1 cells were activated to M0 by incubation with 50 nmol/L Phorbol 12-myristate 13-acetate (Sigma-Aldrich) for 24 hours. Differentiation from M0 to M1 cells was performed by treatment for 48 hours with 15 ng/mL lipopolysaccharides (Sigma-Aldrich) and 50 ng/mL IFNγ (R&D Systems, Bio-Techne). For M2 differentiation, cells were incubated with 5 ng/mL IL13 and 5 ng/mL IL4 (both from R&D Systems) for 48 hours.

Membranes were probed with primary antibodies for HIF-1α (BD Biosciences, 610959), HIF-2α (Cell Signaling Technology, 70862), CD206 (Cell Signaling Technology No. 91992), HLA-DR (Abcam No. ab92511), and HAF as described in refs. 11–13. Images were acquired using Fluorochem M imaging system (Protein Simple). Blot quantification was performed using ImageJ.

For all other methods, please refer to Supplementary Methods.

We stained TMAs containing tumor core samples from 380 patients for HAF, HIF-1α, or HIF-2α. Patients had a median age of 59 years, were primarily white (75%) and male (69%) with good performance status (ECOG 0 = 74%). Representative images of cores after initial antibody validation and workup are shown in Supplementary Figs. S1 and S2. Patient characteristics overall and by TMA block are shown in Supplementary Table S1. Total HIF-1α staining, HIF-1α (total) was associated with higher T stage (Table 1A; Supplementary Table S2A; ref. 15). Similarly, cores from patients with metastatic tumors had higher HIF-1α levels than those without, and showed more variability within patients (given by higher SE of staining within patients), which may indicate increased HIF-1α tumor heterogeneity within metastatic tumors. Higher levels of HIF-1α were also associated with higher grade and showed more variability within patients. Thus, high HIF-1α levels were associated with high tumor grade, T stage, and increased metastasis.

Summaries for HIF-2α and HAF in relation to tumor stage, grade, and metastasis are shown in Table 1A with full tables in Supplementary Table S2B and S2C. HIF-2α levels were higher among patients with ECOG scores of 0, that is, the patients with the best performance status, compared with those with ECOG ≥ 1. HIF-2α was also highly associated with T stage, but with an unexpected trend of decreasing levels with increasing tumor stage from stage T1-T4. Significantly, we noted a clear decrease in HIF-2α levels with increasing grade, tumor size, the presence of metastasis or necrosis, with greater variability within the same patient in smaller tumors. Thus, high HIF-2α levels were associated with lower tumor grade and T stage and were inversely correlated with metastasis.

HAF, like HIF-2α, decreased as T stage increased. HAF was inversely correlated with metastasis, tumor grade, tumor size, and lymphovascular invasion (LVI; Table 1A; Supplementary Table S2C). Thus, HAF levels closely tracked with HIF-2α in that high HAF levels were associated with lower tumor grade and T stage and were inversely correlated with metastasis.

When examining the staining patterns for HIF-1α, we noted that unlike HIF-2α, which was mainly expressed in tumor cells, HIF-1α staining was almost exclusively observed in infiltrating nontumor cells (tumor stroma: S1). Indeed, we observed weak HIF-1α intensity within ccRCC cells, whereas most patients showed intense HIF-1α staining within the tumor stroma (Supplementary Figs. S1 and S3). Thus, by setting low- and high-intensity detection thresholds, we were able differentially quantify tumor and stroma HIF-1α positivity (Supplementary Fig. S3). When HIF-1α positivity using low (tumor) and high thresholds (stroma) were assessed, only nine patients (2.4%) showed positivity for low threshold (tumor-specific) HIF-1α, which under-powered further statistical analysis, whereas all the original samples showed expression of high threshold (stroma-specific) HIF-1α. High threshold HIF-1α staining - HIF-1α (stroma) summaries in relation to tumor grade, stage, and metastasis are shown in Table 1B with all measured parameters shown in Supplementary Table S2D. Similar to overall HIF-1α staining, HIF-1α (stroma), was significantly associated with higher grade and with increased heterogeneity in grade 4, larger tumors or tumors with LVI.

The univariable overall survival (OS) estimates by markers are shown in Table 1C, with full table with other clinical parameters shown in Supplementary Table S3. In this patient set, higher HIF-1α (stroma), and lower HIF-2α or HAF expression were significantly associated with decreased OS. Older age, ECOG scores >0, higher T stage, metastases, higher tumor grade, larger tumor size, LVI, or necrosis were each associated with decreased OS, as expected (Supplementary Table S3). Similar details for progression-free survival are shown in Supplementary Table S4. With no consistent cutoffs identified using recursive-partitioning analysis (RPA, Supplementary Tables S5 and S6), HIF-1α, HIF-2α, and HIF-1α (stroma) are presented with divisions of 33rd and 67th percentiles. An increase of 0.10 in the expression of HIF-1α was associated with 28% increase in the risk of death (HR, 1.28; 1.04–1.58; P = 0.02), whereas a corresponding increase in the expression of HIF-1α (stroma) was associated with 39% increase in risk of death (1.39; 1.06–1.83; P = 0.02; Fig. 1A). In contrast, an increase of 0.10 in HIF-2α levels were associated with a 13% decrease in risk of death (0.87; 0.80–0.93; P < 0.001; Fig. 1B). Similarly, each increase of 0.10 in HAF expression was associated with a 12% reduction in risk of death (HR, 0.88; 95% CI, 0.80–0.98; P = 0.02; Fig. 1C). We observed similar correlations of HIF-1α, HIF-2α, and HAF with progression-free survival as with OS (Supplementary Fig. S4). Thus, in the univariable setting, high HIF-1α or high HIF-1α (stroma) was significantly correlated with decreased OS, whereas high HAF or high HIF-2α was significantly correlated with increased OS.

Figure 1D presents the multivariable OS model with clinically relevant variables together with levels of the HIFs and HAF to determine whether these markers could contribute additional information once the common clinical measures were accounted for. Similar details for progression-free survival are shown in Supplementary Table S7. After accounting for all the variables in the model, higher ECOG, T and M stages, and HIF-1α (stroma) were all associated with a higher risk of death. Neither HAF nor HIF-2α levels contributed additional information about OS after accounting for the standard clinical predictors. Most importantly, HIF-1α (stroma) remained significantly associated with OS, with a 40% increased risk of death for each increase of 0.10 in expression [1.4 (1.1–1.9); P = 0.02]. A likelihood ratio test between the two multivariable models [either with or without HIF-1α (stroma)] indicates that the model that includes HIF-1α (stroma) is significantly better at predicting survival (P = 0.04). Thus, HIF-1α (stroma) is a significant predictor of poor overall survival in the multivariable setting.

To determine the relative levels of HAF and the HIFs in tumor versus paired uninvolved normal kidney tissue, we used large sections of additional tissue (containing both tumor and uninvolved kidney) obtained from 12 patients with early-stage tumors (T stage I comprising grades 2–3) quantitated using Aperio (Supplementary Figs. S5–S8). HIF-1α and HIF-2α were expressed only at low levels in uninvolved kidney, although HIF-1α levels were slightly higher within the glomerulus compared with the tubules (Fig. 2A; Supplementary Fig. S6). HIF-2α levels were significantly elevated in ccRCC (compared with paired uninvolved tissue; Fig. 2B; Supplementary Fig. S7), whereas HIF-1α levels were not significantly different in ccRCC compared with paired uninvolved tissue. Intriguingly, we observed that HIF-1α was expressed within both tumor and stromal cells within these T stage I tumors (Supplementary Fig. S6). HAF was highly expressed within tubular epithelial cells and was decreased in paired ccRCC tissue (Fig. 2A and B; Supplementary Fig. S8). Thus, HIF-2α is increased, HAF is decreased, and HIF-1α (tumor and stroma) is not significantly different in ccRCC compared with uninvolved normal kidney in T stage I ccRCC tissue.

To validate the relationships of HAF/HIFs to tumor grade, we analyzed a TMA comprising uninvolved, low-grade and high-grade ccRCC tissue. HAF levels were significantly decreased in low-grade ccRCC compared with uninvolved kidney but was not significantly different between high-grade and low-grade ccRCC (Fig. 2C). HIF-2α was significantly increased in low-grade ccRCC compared with uninvolved kidney but was then significantly decreased in high-grade compared with low-grade ccRCC (Fig. 2C). HIF-1α levels were almost exclusively stromal [referred to as HIF-1α (stroma)] and was significantly increased in low-grade ccRCC compared with uninvolved kidney, but was not significantly different between high-grade and low-grade ccRCC (Fig. 2C). In a second cohort, we observed a significant decrease in HAF levels in high-grade compared with low-grade tumors, and a significant increase in HIF-1α (stroma) in high-grade compared with low-grade tumor, whereas HIF-2α levels were not significantly different but trended lower in high-grade tumors (Fig. 2D; Supplementary Fig. S9). Taken together, a validation cohort from 49 additional patients confirm an association of decreased HAF and HIF-2α levels with increasing tumor grade, and an association of increased HIF-1α (stroma) with tumor grade.

To address the issue of HIF-1α localization, we enlisted GU pathologists (N. Kandula and D. Sirohi) to examine additional surgical specimens obtained from Huntsman Cancer Institute. These were comprised of low-grade (10 cases), high-grade (13 cases), and uninvolved normal tissue obtained from separate blocks and distal from the tumor (15 cases). Within tumor cell nuclei, HIF-1α and HIF-2α staining was very heterogeneous and were primarily localized to tumor margins and bordering regions of necrosis. HIF-1α expression was also detected in approximately 5% of immune cells in both uninvolved and ccRCC tissue, whereas HIF-2α was not detectable in immune cells (median expression <1% of immune cells (Fig. 3B–C). These immune cells were heterogeneously distributed throughout the tumor, particularly within local regions of inflammation but did not have any specific localization to tumor margins or necrotic regions. Although we observed a significant increase in the percentage of immune cells within high-grade ccRCC tissue compared with uninvolved kidney, the percentage of immune cells that were HIF-1α positive was not significantly different (Fig. 3A and C). Representative images show consistent HIF-1α staining in immune infiltrating cells but rarely in tumor cells, and frequent HIF-2α staining in tumor cells, but never in immune infiltrating cells (Fig. 3D and E). Thus, we confirm expression of only HIF-1α in immune infiltrating cells.

To more definitively identify the HIF-1α–expressing cell types, we performed scRNA-seq on three grade 3/4 ccRCC cases with paired uninvolved kidney. Quality control and data filtering performed produced a total of 7,569 ccRCC and 2,464 uninvolved cells for analysis. Clustering was performed using the Seurat package with cell identities assigned manually using previously described cell-specific markers in renal cancer (Fig. 3F; Supplementary Fig. S10A). Tumor clusters were identified by NDUFA4L2 expression, which has been characterized as a robust expression marker for ccRCC with high expression localized to two clusters in our scRNA-seq data (Tu^a, Tu^b; Supplementary Fig. S10B; refs. 28, 29). In comparison, CAIX, a commonly used protein maker of ccRCC, was not localized to any specific cluster [apart from a few cells in Tu^a (Supplementary Fig. S10B)], reducing its utility as a tumor-specific marker in this study. Analysis of average gene expression within each cluster showed that HIF1A and EPAS1 transcript were strongly expressed within a minority subset of tumor cells (Tu^b), and within the vasculature, but this was not associated with increased transcription of HIF target genes. However, classical HIF target genes (eg, CA9, EPO, VEGFA, and GLUT1) were highly expressed in the majority subset of tumor cells (Tu^a), where HIF1A and EPAS1 transcript were poorly expressed, consistent with HIF-α posttranscriptional protein stabilization by hypoxia or pVHL loss (Fig. 3G; Supplementary Fig. S10C). A cluster heatmap with a more exhaustive list of validated HIF target genes is shown in Supplementary Fig. S11. HIF1A (but not EPAS1) transcript was also elevated in the macrophage M^a and M^b clusters, and this was co-incident with the activation of HIF targets, VEGFA or CXCL8 (Fig. 3G; Supplementary Fig. S10C), suggesting a transcriptional mechanism of HIF activation in these macrophages In contrast, the M^c cluster also showed upregulation of HIF targets, but had low HIF1A transcript levels, suggesting that this cluster may activate HIF through the canonical hypoxia-mediated pathway. All three macrophage clusters expressed a spectrum of M1 and M2 activation markers and did not clearly correspond to any specific macrophage activation subtype (Fig. 3H). RNA in situ hybridization (RNA-ISH) confirmed expression of HIF1A transcript in CD163 expressing macrophages (a marker of alternatively activated, M2 polarized macrophages), whereas expression of the HIF target, SLC2A1 (also known as glucose transporter 1, GLUT1) was mainly localized in NDUFA4L2-expressing ccRCC tumor cells, in the absence of HIF1A transcript expression (Fig. 4A). Manual quantitation showed significantly higher levels of HIF1A mRNA molecules in close proximity to CD163 than to NDUFA4L2, and significantly increased levels of GLUT1 mRNA molecules in close proximity to NDUFA4L2 rather than to CD163 (Fig. 4B). Besides tumor cells, cells of the vasculature and macrophages, no other cell type showed clear upregulation of HIF1A or HIF target genes.

The data suggest that increased HIF1A transcript may translate to increased HIF-1α protein within macrophages but not within tumor cells, where activation of HIF target genes is likely to be primarily HIF-2α dependent. Intriguingly, HIF-1α protein was detected in both the nucleus and cytoplasm in macrophages, unlike in tumor cells, where HIF-1α, when expressed, was primarily localized within the nucleus (Supplementary Fig. S3). Double staining experiments with the endothelial cell marker CD31 confirmed that while many HIF-1α expressing cells resided within blood vessels, the endothelial cells were not themselves HIF-1α positive (Fig. 4C). Similarly, double staining with CD3 indicated that although HIF-1α–expressing cells were within proximity with T cells, T cells themselves did not express HIF-1α (Fig. 4D). However, HIF-1α colocalized with the monocyte/macrophage markers CD68 and CD14 and M2 marker CD163 (Fig. 4E–G). Despite expression of the classically activated, M1 macrophage marker CD86 via scRNA-Seq, we could not detect more than a few CD86-positive cells in ccRCC tissue (10 cases examined), and none colocalized with HIF-1α (not shown). Costaining of HIF-1α with other markers such as B-cell markers CD19, CD20, and CD21 and dendritic cell marker S100 was also attempted but only a few positive cells were observed and again, none colocalized with HIF-1α (not shown). Taken together, our scRNA-seq, RNA-ISH, and IHC data suggest that the highest HIF-1α–expressing cells are TAMs, primarily those that also expressed the M2 marker, CD163.

To further characterize the HIF1A-expressing macrophages, we performed differential gene expression analysis on HIF1A+ (HIF1A expression > 0) and HIF1A− (HIF1A ≤ 0) macrophages, followed by gene set enrichment analysis (GSEA; ref. 30). We identified 361 upregulated and 19 downregulated genes in HIF1A+ compared with HIF1A− macrophages. Of the top 10 upregulated genes, five were previously known to be regulated by HIF-1α (Fig. 4H) (31–35). GSEA identified complement, TNFα signaling via NF-κB, and inflammatory response as positively enriched pathways (P < 0.05) in HIF1A+ macrophages (Fig. 4I). Intriguingly, the complement pathway has previously been linked to ccRCC progression, whereas TNFα has been implicated in ccRCC metastasis and treatment resistance, while HIF-1 has long been implicated in both inflammation and tumorigenesis (36–38).

To investigate the mechanism(s) by which the tumor microenvironment results in HIF-1α upregulation within infiltrating TAMs, but loss of HIF-1α expression in the tumor, we used the THP-1 monocytic cell line and a panel of human RCC cell lines. We found that HIF-1α protein was strongly induced only in hypoxic M2-polarized macrophages (Fig. 5A–C; ref. 39). When we investigated the impact of hypoxia on RCC cells, we found that while cells with wild-type VHL showed robust HIF-1α and HIF-2α accumulation under hypoxic conditions, pVHL deficient cells showed constitutively low levels of HIF-1α in both normoxia and hypoxia, supporting the previously reported notion of HIF-1α loss in ccRCC (Fig. 5D–F; ref. 15). In contrast, HIF-2α levels remained relatively constant in pVHL mutant cells irrespective of oxygen tension, whereas HAF levels generally decreased in hypoxia similar to our previous observations (Fig. 5F–G; ref. 40). Thus, the data suggest that the combination of hypoxia and M2 polarizing conditions within the tumor microenvironment promotes the upregulation of HIF-1α protein in ccRCC TAMs, whereas intrinsic factors associated with pVHL loss inhibits the production of HIF-1α protein in tumor cells regardless of oxygen tension.

To explore the clinical utility of HIF-1α, we measured baseline HIF-1α levels in primary tumors from 57 patients with advanced/metastatic RCC who were subsequently treated with antiangiogenic therapy and assessed the relationship between HIF-1α levels and response (Supplementary Fig. S12A). As observed previously in prior datasets, HIF-1α staining was primarily stromal (Supplementary Fig. S12B–S12C). Importantly, HIF-1α (stroma) levels were significantly higher (P = 0.03) in patients who derived no clinical benefit (CB) from antiangiogenic therapy (median PFS, 3 months; median OS, 18.5 months) compared with those that derived CB (median PFS, 15 months; median OS, 42 months; Fig. 5H). In contrast, no significant associations were observed between HIF-2α and CB (Fig. 5I). Patient with stable disease, partial response, or complete response following therapy were considered to have derived CB, whereas those with progressive disease were considered to have derived no CB. As expected, patients deriving no CB had significantly increased risk of progression or death compared with patients who derived CB (Supplementary Fig. S12D–S12E). Thus, HIF-1α (stroma) is significantly associated with resistance to antiangiogenic therapy.

Here, we have undertaken a comprehensive study to delineate the relationship between the HIFs and HAF during ccRCC progression. Although HIF-1α had been proposed as a tumor suppressor in ccRCC, our study has revealed an unexpected role of HIF-1α in TAMs as an independent indicator of poor outcome. Our data suggest that the exposure to a combination of hypoxia and M2 polarizing cytokines, conditions frequently present within the tumor microenvironment, induce HIF-1α protein expression in TAMs. Our study highlights the potential contribution of TAM HIF-1α to ccRCC aggressiveness in the context of HIF-1α loss within tumor cells, suggests its utility as a prognostic marker, and potential therapeutic target. Surprisingly, HIF-2α, which is widely considered to be a tumor driver, was significantly decreased with tumor grade, and associated with better patient prognosis, similar to its coactivator, HAF (Table 1A, Fig. 2C and D; refs. 13, 15). However, we note that these findings do not bely the contribution of HIF-2α to the ccRCC phenotype, and its potential as a therapeutic target, as has been previously outlined (21–24).

HIF-2α is widely recognized as the major driving factor and a promising therapeutic target in ccRCC (21–24). Intriguingly, a recent meta-analysis found that although HIF-2α was negatively associated with overall survival in a variety of solid tumor types, HIF-2α was a positive prognostic biomarker for metastasis-free survival in kidney cancer (41). Another study showed a significant negative correlation of cytoplasmic HIF-2α with overall survival, but a significant positive correlation of nuclear HIF-2α with increased overall survival, which is in agreement with our findings (42). Only HIF-2α nuclear staining was considered in the analyses performed throughout our study due to the paucity of cytoplasmic HIF-2α detected in our tissue sections.

Although HIF-1α has previously been associated with classically activated, proinflammatory M1 macrophage polarization, we found that HIF-1α primarily colocalized with the M2 marker, CD163, at both the RNA and protein levels, whereas M1 markers such as CD86 were poorly expressed in the ccRCC cases we examined. We note that HIF1A was transcriptionally induced in TAMs but not in THP-1 cells which may reflect context-specific differences between the in vivo (TAMs) and in vitro (THP-1) settings. No differences were observed between HIF-1α protein levels within TAMs compared with macrophages derived from uninvolved kidney suggesting that increased stromal HIF-1α in high-grade ccRCC may be due to increased numbers of immune cells rather than increased HIF-1α expression (Fig. 3A). It is also possible that due to the proximity of the uninvolved kidney to the tumor site, macrophages identified within uninvolved kidney may also consist of TAMs that have migrated away from the tumor site. We note that although HIF-2α has been linked to M2 polarization, we could not detect macrophage-specific HIF-2α expression in any of the ccRCC tissues examined (43).

When we compared the transcriptomes of HIF1A+ TAMs with that of HIF1A− TAMs, we found that the most highly enriched gene sets were those associated with the complement pathway, TNFα signaling via NF-κB, and the inflammatory response (Fig. 4I). However, it is both possible and likely that other HIF-1α–driven pathways may contribute to the protumorigenic roles of TAMs, as described previously (44). Indeed, a known limitation of scRNA-seq technology is the relatively low mRNA capture efficiency and sequencing depth compared to bulk mRNA sequencing. In addition, the proportions of tumor and stromal populations characterized through scRNA-seq may vary from the proportions found in situ due to differences in efficiency of tissue disaggregation.

Although our study is designed to address limitations of previous investigations, there are limitations of our own work that should be considered. First, the primary evaluation of the prognostic significance of HAF and the HIFs was performed using TMAs. Related to this, intratumor heterogeneity (ITH) occurs in ccRCC, and prognostic biomarkers may be influenced by tumor sampling or interrater variability. In addition, there is potential for technical batch effects associated with the tissue distribution over multiple paraffin blocks (Supplementary Table S1). We addressed these issues by the inclusion of additional TMAs and large tissue sections from different institutions (the validation cohort), which confirmed the association of low HIF-2α or HAF with high tumor grade, and high HIF-1α with high tumor grade (Fig. 3B and C). Thus, despite these limitations, our data suggest that mean HIF-1α levels contribute additional value using fairly straightforward quantitative measurement techniques. Second, although our scRNA-seq studies have revealed potential downstream targets of HIF-1α in TAMs, lack of sequencing depth has precluded a more granular analysis of the HIF-1α TAM transcriptome, and certainly those of other nonneoplastic cells within the tumor mass. Finally, we also note the use of OS instead of cancer-specific survival due to lack of data specifically confirming cancer as cause of death.

Our studies build on our current understanding of the role of the HIFs in ccRCC (Fig. 5J). We show that HIF-2α but not HIF-1α is increased in ccRCC tumor cells compared with paired uninvolved kidney. HAF is decreased during ccRCC progression, and both HAF and HIF-2α are associated with lower tumor grade/stage and better outcome. Most notably, HIF-1α–expressing TAMs increase with tumor progression and is an independent negative prognostic indicator, suggesting that HIF-1α inhibition such as by small-molecule inhibitors may be a promising strategy for targeting the tumor-promoting properties of TAMs in ccRCC (Fig. 5I; refs. 45, 46). We also demonstrate the potential utility of HIF-1α as a predictor of response to antiangiogenic therapy. Although future prospective studies will be required to confirm the prognostic role of HIF-1α and its predictive value, our work has established TAM HIF-1α as a potential contributor to ccRCC progression in the context of HIF-1α loss in tumor cells, and identified downstream targets of TAM HIF-1α that warrant further investigation.

R.S.S. Tidwell reports grants from NIH/NCI during the conduct of the study and other support from Galera (salary support provided to institution for effort on a clinical trial) outside the submitted work. J.A. Karam reports personal fees from Pfizer and Roche/Genentech (consulting) and grants to their institution from Roche/Genentech, Merck, and Mirati outside the submitted work. B.L. Maughan reports grants and personal fees from Exelixis, Peloton, and Clovis, grants from Bristol-Myers Squibb and Bavarian-Nordic, and personal fees from Bayer, Merck, Tempus, and Astellas outside the submitted work. T.H. Ho reports other from Ipsen, Genentech, Exelixis, Cardinal Health, Sanofi, Surface Therapeutics, Pfizer, and EMD (advisory board) outside the submitted work. N. Agarwal reports other from Astellas, AstraZeneca, Bayer, Bristol-Myers Squibb, Clovis, Eisai, Eli Lilly, EMD Serono, Exelixis, Foundation Medicine, Genentech, Janssen, Merck, Nektar, Novartis, Pfizer, Pharmacyclics, and Seattle Genetics (consultancy), and reports research funding to their institution from AstraZeneca, Bavarian Nordic, Bayer, Bristol-Myers Squibb, Calithera, Celldex, Clovis, Eisai, Eli Lilly, EMD Serono, Exelixis, Genentech, GlaxoSmithKline, Immunomedics, Janssen, Medivation, Merck, Nektar, New Link Genetics, Novarits, Pfizer, Prometheus, Rexahn, Roche, Sanofi, Seattle Genetics, Takeda, and Tracon outside of the submitted work. M.Y. Koh reports personal fees from Kuda Therapeutics (founder and owner of equity) outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

S.J. Cowman: Software, formal analysis, investigation, visualization, methodology, writing–review and editing. D.G. Fuja: Investigation, methodology. X.-D. Liu: Formal analysis, investigation, visualization. R.S.S. Tidwell: Software, formal analysis, investigation, methodology. N. Kandula: Formal analysis, validation, investigation, methodology. D. Sirohi: Formal analysis, validation, investigation, methodology. A.M. Agarwal: Formal analysis, investigation. L.L. Emerson: Investigation. S.R. Tripp: Validation, investigation, methodology. J.S. Mohlman: Investigation. M. Stonhill: Investigation. G. Garcia: Investigation, methodology. C.J. Conley: Software, formal analysis. A.A. Olalde: Investigation. T. Sargis: Investigation. A. Ramirez-Torres: Investigation. J.A. Karam: Investigation. C.G. Wood: Investigation. K. Sircar: Investigation. P. Tamboli: Investigation. K. Boucher: Software, formal analysis, validation, methodology. B. Maughan: Formal analysis, validation, investigation, methodology, writing–review and editing. B.T. Spike: Formal analysis, validation, investigation, methodology. T.H. Ho: Validation, investigation, methodology. N. Agarwal: Resources, validation, investigation. E. Jonasch: Resources, validation, investigation, methodology, writing–review and editing. M.Y. Koh: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

This work was supported by NIH/NCI grants CA181106 and CA217905 to M.Y. Koh, and NIH/NCI R01CA224917 to T.H. Ho. Research in this publication utilized the High Throughput Genomics and Bioinformatics Resource and the Cancer Biostatistics Resource funded by NIH/NCI grant P30CA042014 awarded to the Huntsman Cancer Institute at the University of Utah, and the Biostatistics Resource Group at the U.T. M.D. Anderson Cancer Center funded by NIH/NCI P30CA016672. We acknowledge financial support by the Huntsman Cancer Foundation and the Experimental Therapeutics and Cell Response and Regulation Programs at HCI. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. This manuscript is dedicated to the memory of Adam A. Olalde.

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.