Standardized Uptake Value for ¹⁸F-Fluorodeoxyglucose Is a Marker of Inflammatory State and Immune Infiltrate in Cervical Cancer

Despite significant advances in screening and prevention, cervical cancer remains a leading cause of cancer deaths in women worldwide (1). The standard treatment for locally advanced cervical cancer, definitive pelvic radiotherapy (RT) given with concurrent cisplatin chemotherapy, has not been improved upon for nearly 20 years, despite the fact that this therapy fails in approximately one third of patients (2–4). Treatments for recurrent and metastatic cervical cancer are limited, associated with significant morbidity, and rarely curative (5, 6). Therefore, innovative treatment strategies are needed.

A number of prognostic imaging markers have been developed for cervical cancer, including metrics derived from positron emission tomography with ¹⁸F-fluorodeoxyglucose (FDG-PET) and magnetic resonance imaging (MRI; refs. 7–11). We have previously demonstrated that persistent or progressive disease on posttreatment FDG-PET is a powerful negative prognostic marker, and that it is associated with PI3K/AKT pathway activation (8, 12). However, this is of limited value when making initial treatment decisions, thus a pretreatment marker is desirable. Pretreatment maximum standardized uptake value (SUV_max) of the primary tumor is particularly suited for this purpose. Other potentially more robust PET metrics have been proposed, but SUV_max is the most widely clinically reported, most widely studied, and we have previously reported its prognostic significance in patients with cervical cancer (7). A deeper knowledge of primary tumor biology associated with a high SUV_max would further our understanding of how high-risk cancers should be treated, and could transform FDG-PET signal from a prognostic marker to a marker predictive of response to specific therapies.

The purpose of this work was to identify targetable signaling pathways, genetic signatures, and immune cell populations enriched in cervical cancers with a high pretreatment SUV_max. We established an SUV_max cutoff most prognostic for freedom from progression (FFP) in a cohort of patients with cervical cancer treated with definitive RT using modern techniques to identify high-risk cancers, confirming our prior work that established SUV_max as prognostic in a new cohort of patients (7). We used whole-transcriptome RNA-seq data from a subcohort of patients to identify signaling pathways associated with high SUV_max using gene set enrichment analysis (GSEA). We likewise used the RNA-seq data to identify immune cell populations associated with high SUV_max using CIBERSORT, and confirmed these results using a tissue microarray (TMA). Finally, we performed benchtop studies to further explore the impact of macrophages on cervical cancer glucose metabolism and the signaling pathways identified in the GSEA.

All studies involving patient records and samples were conducted in accordance with the Declaration of Helsinki. An SUV_max cutoff with the most prognostic value was determined in a cohort of patients with cervical cancer treated with definitive RT at our institution between 2006 and 2015. To be included, patients had to have completed the entire prescribed course of definitive-intent RT. FIGO stage I–IVa tumors (FIGO 2009 staging) and all histologic subtypes were allowed, patients with metastatic disease were excluded, all patients were >18 years old at the time of their diagnosis, and all patients underwent pretreatment FDG-PET/computed tomography (FDG-PET/CT) or FDG-PET/magnetic resonance imaging (FDG-PET/MRI). Details of our institutional practice for treating cervical cancer with RT have previously been reported (13). Briefly, patients were treated with external beam RT to the pelvis using intensity-modulated RT and to the para-aortic lymph nodes if they were involved. The primary cervical tumor was also treated with image-guided brachytherapy. Nearly all (90.1%) patients were treated with concurrent chemotherapy. This cohort of patients was analyzed with institutional review board (IRB) approval with waiver of consent (IRB No. 201601052).

The second cohort of patients was used to identify overexpressed genes and pathways associated with high FDG uptake. This consisted of patients enrolled on a prospective tumor banking study with written informed consent (IRB No. 201105374). Overlap between this cohort and the SUV_max cohort is described in the results below and in Supplementary Fig. S1. Tumor samples with sufficient high-quality RNA for whole-transcriptome sequencing (RNA-seq), as defined by the criteria utilized for The Cancer Genome Atlas (TCGA), were included in this study (14). Tumors were not macrodissected and thus results are expected to include genes attributable to tumor and any infiltrating stromal or immune cells.

The third cohort of patients consisted of patients whose tumor samples were incorporated into a TMA, and were interpretable for IHC analysis. Overlap between this cohort and the SUV_max cohort is also described in the results below and in Supplementary Fig. S1. This was also done with a waiver of consent from the IRB for access to tissue specimens and clinical data (IRB No. 201807063). Cohort 3 was also composed of patients treated with definitive RT.

FDG-PET was performed as part of routine staging evaluation of patients with newly diagnosed cervical cancer according to our institution's imaging protocols, which have been developed to maximize visualization of cervical tumors (15). Briefly, patients were required to fast for at least 4 hours prior to FDG injection, and blood glucose levels were required to be less than 200 mg/dL at the time of tracer injection (5/331 patients had a blood glucose over 200 mg/dL). PET/CT images were acquired approximately 60 minutes after FDG injection. Images were acquired on 1 of 3 PET/CT scanners: Siemens Biograph 2 (n = 107), Siemens Biograph 40-HD (n = 187), and Siemens mCT (n = 32), and on one PET/MRI scanner (Siemens mMR, n = 5). All images were reconstructed using standard iterative reconstruction algorithms.

As part of routine clinical practice, the SUV_max of the primary cervical tumor was measured by determining the maximum single-voxel SUV within a volume of interest encompassing the primary tumor. A representative image of a large cervical tumor with an automatically generated metabolic tumor volume is shown in Supplementary Fig. S2. The SUV is defined as: SUV_max = (tissue radioactivity concentration (mCi/mL) × patient weight (g))/injected dose (mCi).

Given the significant overlap between the cohort of patients used in the SUV analysis and the cohort of patients used for the RNA-seq analysis, a comparison was made between patients included only in the SUV analysis cohort and patients included in both cohorts using the Wilcoxon two-sample test and Fisher exact test, as appropriate.

The optimal SUV_max for discriminating patients with good versus poor FFP was determined by outcome-oriented cutoff methodology utilizing an alpha-adjusted log-likelihood test (16). Once the optimal SUV_max was determined, the prognostic significance of this SUV_max for FFP, cancer-specific survival (CSS), and overall survival (OS) was evaluated with Kaplan–Meier/log-rank and Cox univariate and multivariate analysis. For FFP any recurrence was counted as an event and patients were censored at time of last follow-up. For CSS death from cervical cancer was counted as an event, and for OS, death from any cause was counted as an event. In either case, patients were censored at time of last follow-up or death.

A bootstrap Cox multivariate analysis was then performed to better illustrate the significance of SUV_max in the context of other known clinical prognostic markers. The bootstrap analysis was performed by taking 1,000 random samplings from the first cohort of patients, such that each sampling was half the size of the total cohort. A Cox multivariate analysis was then run on each sampling. The variables included in this analysis were SUV_max, FIGO stage, lymph node involvement by FDG-PET (none, pelvic, para-aortic), age, and histology (squamous vs. nonsquamous).

We performed RNA-seq analysis on the patients in cohort 2. Specific details regarding this are included in the Supplementary Methods. The expressed genes were ranked according to fold change between patients with high SUV_max (≥11.41) and low SUV_max (<11.41) values and included to perform GSEA using default parameters (17–21). The H: Hallmark gene sets (n = 50) and the C2: curated gene sets (n = 4,762) from the Molecular Signatures Database (MSigDB) were used to discover altered biological pathways between patients with cervical cancer with an SUV_max above or below the cutoff point established in the SUV_max analysis (21).

Using CIBERSORT, we estimated the relative abundances of hematopoietic cells in the patients from cohort 2 (22). For signature genes, we used the leukocyte gene signature matrix, LM22, which contains 547 genes that distinguish 22 human hematopoietic cell phenotypes, including seven T-cell types, naïve and memory B cells, plasma cells, natural killer (NK) cells, and myeloid subsets. Relative fractions of B cells, CD4 T cells, CD8 T cells, neutrophils (M0, M1, and M2), macrophages, monocytes, dendritic cells, and NK cells were compared between patients with high and low SUV_max using the Student t test.

To evaluate the prevalence of various immune cells in human cervical tumors, we performed IHC for markers of macrophages and neutrophils on a TMA of human cervical cancer tumors. Details regarding this TMA have previously been published (23), and are outlined in more detail in the Supplementary Methods. Briefly, areas of cancer on formalin-fixed paraffin-embedded (FFPE) tumor specimens were identified and triplicate 1.5-mm cores were placed in a TMA. Sections 6 μm thick were then mounted onto polarized slides for IHC staining. Staining for macrophages was done on a Ventana automated stainer (Ventana Medical Systems). Staining was performed with anti-CD68 (0.4 μg/mL, monoclonal mouse, Ventana) and anti-CD163 (0.3 μg/mL, monoclonal mouse, Cell Marque). Neutrophil staining was performed manually, and slides were costained with CD66b (1:600, BD Biosciences) and neutrophil elastase (1:200, Novus).

Stained slides were digitally scanned and analyzed using the Aperio eSlide Manager platform (Leica Biosystems). Samples were considered either negative (≤1% of cells stained) or positive (>1% of cells) for all IHC stains used. Categorizing immune infiltrate as high versus low has previously been described, including for CD68 (24–26). Samples were scored according to immune cell infiltrate into the tumor. All tumor samples were scored by two independent blinded readers. Readers evaluated cores at 20× magnification and scored each core as either negative (≤1% of cells stained) or positive (>1% of cells stained). Agreement between readers was assessed with Cohen's kappa statistic. The scores from both readers for all available cores from each patient were averaged to generate a composite score rounded to the nearest whole number (<0.5 to 0 for negative, ≥0.5 to 1 for positive). This score was matched to the SUV_max for that tumor, and the SUV_max values of tumors that stained positive versus negative were compared using the Wilcoxon two-sample test.

Two human cervical cancer cell lines obtained through the ATCC collection were evaluated: ME180 and SiHa. Experiments were performed on cell lines between passages 10 and 30. Mycoplasma testing was performed by the iPSC Center (ME180 and SiHa cells, February 20, 2019) at Washington University School of Medicine. All experiments were carried out under standard tissue culture conditions at 37°C and 5% CO₂, with cells maintained in IMDM with 10% fetal bovine serum and 0.1 mg/mL gentamycin.

Human monocytic THP-1 cells were differentiated into M0, M1, and M2 macrophages using methods outlined in the Supplementary Methods. The generation of macrophage-conditioned media is also outlined in the Supplementary Methods.

In the coculture experiments, SiHa and ME180 cells were plated onto 6- or 12-well plates, and THP-1 monocytes were differentiated into M0, M1, or M1 macrophages in 6 or 12 Transwell inserts (membrane pore size of 0.4 μm, Corning). Cells were cocultured for 48 hours. For the conditioned media experiments, cervical cancer cells were maintained as above and directly exposed to conditioned media collected from different macrophages. Cervical cancer cells from both conditioned media and coculture experiments were harvested after 48 hours for Western blot analysis or a glucose uptake assay.

Activation of STAT3 by coculture with macrophages was confirmed using Western blotting with primary antibodies against total STAT3 (1:1,000, room temperature for 1 hour, Santa Cruz Biotech) and phosphorylated STAT3 (pSTAT3 (Tyr705), 1:2,000, 4°C overnight, Cell Signaling Technology). Blots were probed with horseradish peroxidase–conjugated anti-rabbit (1:3,000, Cell Signaling Technology) for 1 hour at room temperature. Blotting was also performed for actin (1:1,000; Santa Cruz Biotech) at room temperature for 1 hour. The STAT3 and actin primary antibodies were both conjugated to HRP. Whenever necessary, blots were stripped by incubating membrane with restore Western blot stripping buffer (Thermo scientific) with intermittent shaking and reprobed with required antibodies. For detection, Amersham ECL select (GE Healthcare) was used according to the manufacturer's protocol. Images were acquired using Chemidoc Imaging systems (BioRad).

A glucose uptake assay was performed using 2-[N-(7-nitrobenz-2-oxa-1,3-diazol-4-yl)amino]-2-deoxy-D-glucose (2-NBDG; Cayman) by flow-cytometric detection of fluorescence produced by the cells. Briefly, cervical cancer cells were plated in a 12-well plate and cocultured with macrophages or conditioned media collected from macrophages as described above. After 48 hours of exposure, cells were trypsinized, washed in PBS and stained by 2-NBDG. For 2-NBDG uptake in the experimental settings of α-IL6, cells were exposed with α-IL6 (500 μg/mL). After 48 hours of exposure, cells were washed three times with PBS to remove remaining α-IL6 from the culture media. Subsequently, cells were trypsinized, washed in PBS, and stained by 2-NBDG. Flow cytometry was performed using a MACSQuant Analyzer10 flow cytometer (Miltenyi Biotec). For each measurement, data from 10,000 single-cell events were collected using FACScan flow cytometer.

Cytokines were analyzed by human Discovery Assay using an array of 13 cytokines (Human Cytokine Array/Chemokine Array 13-Plex, Eve Technologies Corporation). For each sample, conditioned or culture media were collected from macrophages cocultured with ME180 cells as a supernatant and processed according to the company's specifications and requirements.

The cohort of patients used in the SUV_max analysis (cohort 1) consisted of 318 patients, the cohort of patients used for the RNA-seq analysis (cohort 2) consisted of 68 patients, and the cohort used for the TMA consisted of 82 patients. All but eight of the patients in cohort 2 were also included in cohort 1, and all but 11 of the patients in cohort 3 were also included in cohort 1. Patient characteristics are summarized in Table 1, and the overlap between the cohorts is summarized in Supplementary Fig. S1. We compared the characteristics of the patients included only in the SUV_max analysis with the patients included in both the SUV_max analysis and RNA-seq analysis, and also with the patients included in both the SUV_max analysis and the TMA analysis. There were no significant differences between these groups in terms of age at diagnosis, FIGO stage, lymph node involvement by FDG-PET, histology, or the primary tumor SUV_max (Table 1).

Median follow-up was 3.2 years for cohort 1, 4.0 years for cohort 2, and 5.1 years for cohort 3. OS rates at 2 and 5 years were 76.1% and 66.8%, respectively, in cohort 1; 72.5% and 56.5%, respectively, in cohort 2; and 69.5% and 56.1%, respectively, in cohort 3. CSS rates at 2 and 5 years were 77.6% and 70.7%, respectively, in cohort 1; 64.7% at both 2 and 5 years in cohort 2; and 73.2% and 61.0%, respectively, in cohort 3.

Using outcome-oriented cutoff methodology, an SUV_max < 11.41 was found to be most prognostic of FFP. This is similar to the SUV_max of 13.3 previously reported as being associated with the worst survival (7). This SUV_max cutoff was also prognostic of CSS, but not OS (Fig. 1A–C), and was prognostic for local (cervical) recurrence and any pelvic recurrence as well (Supplementary Fig. S3). On Cox univariate analysis, SUV_max (≥ vs. <11.41), FIGO stage, and lymph node involvement on PET (para-aortic vs. none) were found to be prognostic for FFP. On multivariate analysis for FFP, SUV_max, FIGO stage, and lymph node involvement were all significant (Table 2).

We also performed a bootstrap multivariate analysis to further determine the relative prognostic significance of SUV_max (≥ vs. <11.41) in the context of other known clinical prognostic variables; namely, age at diagnosis, FIGO stage, nodal involvement on pretreatment FDG-PET/CT, and histology. Results summarized in Supplementary Table S1 demonstrate that in 1,000 random samplings from cohort 1, SUV_max proved to be a significant prognostic factor 63.3% of the time, FIGO stage 72.1% of the time with FIGO stage III portending a particularly poor prognosis, nodal involvement 96.5% of the time with para-aortic lymph node involvement being a powerful prognostic variable, and age and histology 12.6 and 12.0% of the time, respectively. Median hazard ratios and P values when each variable was included in the model are shown in Supplementary Table S1.

Once the association of SUV_max and prognosis was established, we performed a GSEA on the RNA-seq data to determine gene sets associated with a high SUV_max. The GSEA was performed treating SUV_max as a dichotomous variable (≥ or <11.41). We focused on Hallmark gene sets, which represent well-defined biological states that display coherent expression, and have been generated using a computational methodology (27). The top three Hallmark gene sets associated with high SUV_max were for inflammatory response, TNFα signaling via NF-κΒ, and JAK/STAT3 signaling (Fig. 2). There was overlap between many of the genes in these gene sets (i.e., gene present in >1 gene set). Furthermore, many of these genes encode cytokines and chemokines associated with specific immune cell subpopulations, including cells within the myeloid lineage, and macrophages in particular (Fig. 2).

Given the association between high SUV_max and inflammatory pathway genes related to myeloid cell populations, we performed a deconvolution of the bulk RNA-seq data using the CIBERSORT algorithm. The relative prevalence of neutrophils, monocytes, and NK cells was significantly different between the high and low SUV_max groups, with higher proportions of neutrophils and monocytes and lower proportions of NK cells in the high SUV_max group (Fig. 3A). The proportion of macrophages almost met the threshold for significance (P = 0.066; Fig. 3A). Given the number of genes related to macrophages overexpressed in the high SUV_max tumors in the GSEA, we also examined the predicted prevalence of macrophage subgroups, and found that a greater proportion of M2-polarized macrophages were predicted in high SUV_max tumors (Fig. 3B).

To validate the findings from the GSEA and CIBERSORT data suggesting that high SUV_max is associated with inflammatory pathways and immune cells in the myeloid lineage, we performed IHC staining on a TMA of human cervical cancer tumors to identify immune cell subpopulations. We specifically focused on surface markers for neutrophils and macrophages. CD68 expression was used as a general marker of macrophages, with good agreement between the two readers (kappa = 0.69). Tumor samples with minimal/negative CD68 staining (≤1%) showed a lower average SUV_max than samples that were positive (>1%) for CD68 (Fig. 3C). CD163 expression was used as a marker of M2-polarized macrophages. There was no significant difference in the SUV_max between samples that were negative for CD163 versus those that were positive (Supplementary Fig. S4). To evaluate neutrophil infiltrate, separate TMA slides were costained with CD66b and neutrophil elastase; there was good agreement between readers of neutrophil staining (kappa = 0.70). There was a nonsignificant positive correlation between SUV_max and neutrophil infiltrate in the tumor microenvironment (Fig. 3D).

To study the relationship between macrophages, glucose uptake, and JAK/STAT3 signaling, human monocytic THP-1 cells were differentiated into macrophages using phorbol 12-myristate 13-acetate (PMA), then further polarized into M1 macrophages using interferon-γ (INFγ) and lipopolysaccharide (LPS), or M2 macrophages using IL4 and IL13. Human cervical cancer cells (SiHa and ME180) were then exposed to macrophage-conditioned media or were cocultured with macrophages in transwell plates.

Glucose metabolism was measured using fluorescent 2-NBDG by flow cytometry. Exposing ME180 and SiHa cells to either M0, M1, or M2 macrophage-conditioned media, or coculturing ME180 and SiHa cells with M0, M1, or M2 macrophages, resulted in a significant increase in 2-NBDG uptake by the cervical cancer cells. This was seen in both cell lines with all subtypes of macrophages (Fig. 4A). We therefore evaluated an array of 13 proinflammatory cytokines and chemokines that were secreted in this transwell system (Eve Technologies). This demonstrated an increase in a wide array of cytokines secreted by macrophages. Two cytokines, IL1β and IL6, showed consistent increases when macrophages were cocultured with cervical cancer cells, compared with the macrophages alone (Fig. 4B). Results from four cytokines are not shown (INFγ, IL2, IL4, and IL13) because either the signal for most conditions was too low to be detected (INFγ and IL2), or the signal from ME180 cells alone outweighed that from the macrophages by several orders of magnitude (IL4 and IL13).

We further explored the impact of IL6 on glucose metabolism in ME180 and SiHa cells. IL6 is known to activate the JAK/STAT3 pathway, resulting in STAT3 phosphorylation. ME180 cells showed no baseline STAT3 phosphorylation, but showed STAT3 phosphorylation when cultured with macrophages or macrophage-conditioned media. SiHa cells showed baseline STAT3 phosphorylation, which was unchanged or possibly slightly diminished when cocultured with macrophages and macrophage-conditioned media (Fig. 4C). Treatment with neutralizing antibody to IL6 (500 ng/mL) resulted in a decrease in STAT3 expression in SiHa cells (D), as well as a decrease in glucose uptake, as measured by 2-NBDG when either SiHa or ME180 cells were cocultured with M2 macrophages (Fig. 4E).

Finally, to connect JAK/STAT3 signaling to glucose metabolism in human tumors, we repeated the GSEA, but ranking expressed genes according to STAT3 expression as opposed to SUV_max. The Hallmark Glycolysis pathway was significantly upregulated in tumors with high STAT3 expression [false discovery rate (FDR) q <0.0001, FWER P <0.0001).

In this work, we have made a connection between a high SUV_max derived from FDG-PET, an immune infiltrate characterized by myeloid-derived cells, particularly monocytes and M2 macrophages, and inflammatory signaling pathways. As high SUV_max is also prognostic, this could serve as an imaging biomarker to identify high-risk cancers with a specific inflammatory and immune profile, and guide use of novel targeted or immune therapies for these patients. To our knowledge, these are the first data to demonstrate a connection between tumor glucose metabolism, inflammatory signaling pathways, and immune cell infiltrate in cervical cancer.

Our data specifically suggest that a high SUV_max is associated with inflammatory pathways, including JAK/STAT3 and NF-κΒ signaling (Fig. 2), both of which have previously been implicated as important signaling pathways in cervical cancer. Activated STAT3 (pSTAT3) has been shown to be a poor prognostic marker in cervical squamous cell carcinoma (28, 29). Likewise, NF-κΒ signaling has been implicated in cervical cancer carcinogenesis, progression, and therapeutic resistance, though the exact role of NF-κΒ in cervical cancer appears to be complex (30, 31). In addition, many of the genes identified in these pathways in our cohort have been shown to have prognostic significance or important biological implications in cervical cancer, including IL6 (32).

The inflammatory pathway gene sets associated with high SUV_max on the GSEA share a number of genes, particularly cytokines and chemokines linked with myeloid-derived immune cells (Fig. 2). The association between SUV_max and myeloid-derived immune cells was further supported by our CIBERSORT and IHC analyses (Fig. 3). The CIBERSORT algorithm showed an association between SUV_max and neutrophils, monocytes, and M2-polarized macrophages, and a negative association with NK cells, and the IHC analysis of the TMA further supported the association between CD68-positive cells (i.e., monocytes and macrophages) and SUV_max. An association between neutrophils and SUV was not seen on the TMA, though this remains an active area of investigation (33). The negative association between SUV_max and NK cells seen in the CIBERSORT analysis is interesting given the known high glucose uptake of NK cells (34). It is possible that the tumor microenvironment is modulating expression of NK cell genes included in the LM22 signature used for CIBERSORT (35). This study therefore provides the important insight that the immune microenvironment and tumor glycolysis are interconnected.

We have further demonstrated a connection between glucose metabolism, macrophage infiltrate, and JAK/STAT3 signaling (Fig. 4). Specifically, we have shown that coculturing human cervical cancer cells with macrophages or growing cervical cancer cells with macrophage-conditioned media results in increased glucose uptake. Cytokine array profiling showed IL6 and IL1β concentrations were significantly increased in media from macrophages cocultured with ME180 cells, compared with media from the macrophages alone, suggesting an interplay between the cervical cancer cells and the macrophages that alters expression of these factors. Coculture with macrophages also increased STAT3 phosphorylation in ME180 cells, and treatment with neutralizing antibody to IL6 decreased glucose uptake in SiHa and ME180 cells. Finally, we demonstrated that the Hallmark Glycolysis gene set is significantly upregulated in human tumors with high STAT3 expression, once again utilizing GSEA.

In aggregate, these data demonstrate: (i) primary tumor SUV_max is associated with FFP and CSS in women with cervical cancer; (ii) high SUV_max is associated with inflammatory signaling pathways; (iii) high SUV_max is also associated with a myeloid-derived immune cell infiltrate, particularly macrophages. We propose that the relationship between cervical cancer glucose metabolism and macrophage immune infiltrate is mediated in part by cytokines and chemokines involved in JAK/STAT3 signaling. There is some other evidence in the literature that STAT3 can affect glucose metabolism through modifying hexokinase expression, FOXO1 and HIF1α signaling, and mitochondrial metabolism, though further study is needed to establish the exact mechanism of this connection in cervical cancer (36–38). Notably, we did not establish a connection between SUV_max and the Hallmark Hypoxia signature in our GSEA.

The clinical implications of inflammatory pathways and immune makeup in cervical cancers are actively being investigated and elucidated. Tumor-associated macrophages have been shown to be prevalent in cervical tumors, more so than regulatory T cells, suggesting they contribute to an immunosuppressive tumor microenvironment (24, 39, 40). Tumor-associated macrophages have also been associated with poor prognosis in cervical cancers (26, 41). JAK/STAT3 is also a targetable pathway, and ruxolitinib is clinically used in blood disorders, and is under investigation in a number of cancers (NCT02713386, NCT02876302, and NCT03120624; refs. 42, 43). Our data provide a link between a proinflammatory but immunosuppressive microenvironment in cervical cancer, and a routinely clinically measured imaging metric, the PET-derived SUV_max for FDG.

This study has important limitations. We have focused on the SUV_max of cervical tumors treated at a single institution with a robust PET imaging program over a recent period of time (since 2006). There are a number of potential critiques regarding using SUV_max as a quantitative marker. In this case, our PET data were derived from several different scanners with differing performance characteristics, and the discriminating value for high versus low SUV_max in this study reflects this specific data set. We have not validated the reproducibility of SUV_max in cervical cancer across a broad array of scanners, reconstruction methods, and different imaging centers, and it is therefore not clear how easily it could be adopted as a quantitative predictive biomarker across many institutions (44). Future studies could alternatively utilize SUV_peak, most typically defined as the average SUV within a small volume (e.g., 1 cm³) within the highest uptake region of the tumor, or an SUV ratio. These have been suggested as a more robust alternative to SUV_max less sensitive to differences in noise, scanner resolution, and reconstruction, and both have some evidence to support their use in cervical cancer (45–47). Although it has limitations, SUV_max is simple to measure, is one of the most widely reported FDG-PET metrics in standard clinical interpretations, and we have previously reported its prognostic significance in a separate cohort of patients with cervical cancer (7, 48, 49). Regardless, our data have outlined high FDG uptake as a poor prognostic factor that is correlated with the inflammatory state and immune infiltrate in cervical cancer tumors, even if our specific SUV_max cutoff may not hold true across a more diverse array of scanners and institutions. Because of its broad clinical availability, the connection between FDG uptake and inflammatory and immune state could have immediate clinical and therapeutic implications.

Future work should focus on further study and characterization of the interplay between immune infiltrate and cervical cancer metabolism, and on validating our connection between high SUV_max for FDG and a macrophage-predominant immune infiltrate. With regard to the connection between glucose metabolism and immune infiltrate, here we have broadly established an interplay between the presence of macrophages, or macrophage-conditioned growth medium, and cervical cancer glucose metabolism. We focused on factors associated with JAK/STAT3 signaling, and IL6 in particular, as our GSEA and cytokine array data suggested this as a potential candidate for mediating this effect. However, the interactions between macrophages and cervical cancer cells are certainly complex, and mediated by a host of other factors. We have also recently demonstrated a connection between FDG uptake, oxidative stress, and radioresistance in cervical cancer cell lines and animal models (50). Further work should explore the interplay between glucose utilization, oxidative stress, and the inflammatory and immune microenvironment in primary cervical tumors.

In sum, we have demonstrated that a high SUV_max for FDG is prognostic of outcomes in cervical cancer, and also associated with proinflammatory signaling pathways and a monocytic immune infiltrate. Ultimately, SUV_max may potentially be used as a surrogate marker for this microenvironment to identify cancers that can be treated with therapies targeting proinflammatory pathways (e.g., JAK/STAT3 signaling) or therapies that aim to reverse the infiltration of tumor-associated macrophages and other cells that suppress adaptive immunity.

J.M. Floberg reports grants from the Radiological Society of North America during the conduct of the study. B.A. Siegel reports grants from the NIH, ASTRO Resident Research Seed Grant, and RSNA Resident Research Grant during the conduct of the study; grants and personal fees from the American College of Radiology, Imaginab, Inc., and Blue Earth Diagnostics; and personal fees from the American Medical Foundation for Peer Review and Education, Avid Radiopharmaceuticals, Inc., Capella Imaging, LLC, Curium Pharma, Radiological Society of North America, and Siemens Healthineers outside the submitted work. S. Markovina reports grants from the NCI during the conduct of the study and grants and nonfinancial support from GlaskoSmithKline outside the submitted work. J.K. Schwarz reports grants from NIH, ASTRO, and RSNA during the conduct of the study; grants from Goldman Sachs Philanthropy Cancer Research Fund, Siteman Investment Program, 2019 AACR–Bristol Myers Squibb Mid-career Female Investigator Grant, NIH; and nonfinancial support from Calithera Biosciences outside the submitted work. No disclosures were reported by the other authors.

J.M. Floberg: Conceptualization, resources, data curation, formal analysis, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. J. Zhang: Conceptualization, data curation, formal analysis, validation, visualization, methodology, writing–original draft, writing–review and editing. N. Muhammad: Investigation, visualization, methodology, writing–original draft, writing–review and editing. T.A. DeWees: Formal analysis, writing–original draft, writing–review and editing. M. Inkman: Formal analysis, writing–review and editing. K. Chen: Formal analysis, investigation, writing–review and editing. A.J. Lin: Formal analysis, investigation, methodology, writing–review and editing. R. Rashmi: Investigation, methodology, writing–review and editing. K. Jayachandran: Data curation, formal analysis, investigation, methodology, writing–review and editing. B.T. Edelson: Methodology, writing–review and editing. B.A. Siegel: Formal analysis, investigation, writing–review and editing. F. Dehdashti: Formal analysis, investigation, writing–review and editing. P.W. Grigsby: Conceptualization, resources, data curation, formal analysis, supervision, writing–review and editing. S. Markovina: Conceptualization, resources, supervision, investigation, methodology, writing–review and editing. J.K. Schwarz: Conceptualization, resources, supervision, funding acquisition, methodology, project administration, writing–review and editing.

We would like to thank Marina Platik and the Anatomic and Molecular Pathology Core Laboratory at Washington University in St. Louis for their assistance in staining the TMA. We would also like to acknowledge Songyang Wang, MD, PhD, and Min Tan from Stephanie Markovina's laboratory for their assistance with IHC and for supplying supplies, reagents, and expertise for investigating the JAK/STAT3 pathway. This work was supported in part by NIH R01CA181745 and the AACR–Bristol Myers Squibb Mid-career Female Investigator Award to J.K. Schwarz; by NIH K08CA237822 to S. Markovina, by NCI K22CA237839 to J. Zhang, and by ASTRO Resident Research Seed Grant 531448 and the RSNA Resident Research Grant to J.M. Floberg.

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.