Immune checkpoint blockade (ICB) has become a standard therapy for several cancers, however, the response to ICB is inconsistent and a method for noninvasive assessment has not been established to date. To investigate the capability of multimodal imaging to evaluate treatment response to ICB therapy, hyperpolarized 13C MRI using [1–13C] pyruvate and [1,4–13C2] fumarate and dynamic contrast enhanced (DCE) MRI was evaluated to detect early changes in tumor glycolysis, necrosis, and intratumor perfusion/permeability, respectively. Mouse tumor models served as platforms for high (MC38 colon adenocarcinoma) and low (B16-F10 melanoma) sensitivity to dual ICB of PD-L1 and CTLA4. Glycolytic flux significantly decreased following treatment only in the less sensitive B16-F10 tumors. Imaging [1,4–13C2] fumarate conversion to [1,4–13C2] malate showed a significant increase in necrotic cell death following treatment in the ICB-sensitive MC38 tumors, with essentially no change in B16-F10 tumors. DCE-MRI showed significantly increased perfusion/permeability in MC38-treated tumors, whereas a similar, but statistically nonsignificant, trend was observed in B16-F10 tumors. When tumor volume was also taken into consideration, each imaging biomarker was linearly correlated with future survival in both models. These results suggest that hyperpolarized 13C MRI and DCE MRI may serve as useful noninvasive imaging markers to detect early response to ICB therapy.
Hyperpolarized 13C MRI and dynamic contrast enhanced MRI in murine tumor models provide useful insight into evaluating early response to immune checkpoint blockade therapy.
See related commentary by Cullen and Keshari, p. 3444
Immunotherapy using immune checkpoint blockade (ICB) has emerged as a promising cancer treatment (1). ICB therapy removes the inhibitory signals of T-cell activation by blocking costimulatory receptors or ligands such as PD-1, PD-L1, and CTLA4, enabling tumor-reactive T cells to override regulatory mechanisms and exhibit antitumor activity (2). A key feature of this therapy is the highly durable tumor response, resulting in a plateau in the tail of the survival curve unlike chemotherapy or genomically targeted therapy, the benefits of which tend to diminish as time progresses (3). Despite the potential of this type of therapy, the remarkable responses to ICB are limited to a fraction of patients (4). Thus, an accurate and reliable tool for assessment of early treatment response is needed (5). Although several biomarkers including PD-L1 expression (6), tumor mutation burden (7, 8), mismatch repair deficiency (7, 9), radiomic features (10, 11), and tumor-infiltrating lymphocytes (TIL) have been proposed (12), none of them has become a gold standard to predict treatment response to date (13). The unique response to immunotherapy also complicates the search for a biomarker. Unlike cytotoxic therapy, where the effectiveness can be frequently judged by the reduction in the size of the tumor, the treatment response in immunotherapy is often delayed; tumors often transiently enlarge or new lesions will appear in the early stages only to be followed by tumor shrinkage or long-term stability of tumor size (14, 15). The delayed response can pose a potential problem in designing treatment plans as the effectiveness of the therapy cannot be judged until the optimal treatment window has passed. To alleviate this problem, the immune-related response criteria (irRC), immune-related response criteria (irRECIST), and immune RECIST (iRECIST) standards were developed for imaging assessment of treatment response. These criteria require a consecutive scan at least 4 weeks apart for confirmation of progressive disease (15, 16). When the treatment is found ineffective, patients may lose the opportunity to receive other treatments during the assessing time. Thus, developing an early prediction of the treatment response to immunotherapy is of great importance (17).
Treatment response to cancer immunotherapy is closely related to the tumor microenvironment. Immunologically “hot” tumors are typically characterized by a high degree of CD8+ cytotoxic T cells infiltration (18). A lack of CD8+ T-cell infiltration and the presence of immune suppressor cells including regulatory T cells, myeloid derived suppressor cells, and type II macrophages are part of the extrinsic mechanisms of resistance to immunotherapy (19). Recently, increased metabolic competition between tumor cells and immune cells was also reported. Aerobic glycolysis (the “Warburg effect”), a metabolic effect characteristic of cancer cells (20), causes depletion of extracellular glucose and restricts glucose availability to T cells. The decrease in the availability of glucose leads to a suppression of glycolytic metabolism in T cells (21). Decreased glycolysis results in decreased effector function as phosphoenolpyruvate produced as a byproduct of glycolysis is essential in maintaining the high cytosolic Ca2+ levels necessary to support T-cell activation (22, 23). In this context, it is reported that PD-L1 promotes glycolysis in tumors via Akt/mTOR activation, and that PD-1 blockade can decrease glycolysis in tumors (21). The microenvironment of tumors are also a challenging environment for the immune system in other respects as it is characterized by hypoxia, low pH, high interstitial fluid pressure, and immune-inhibitory metabolites such as lactic acid (24, 25). The impaired perfusion of abnormal tumor blood vessels can create a highly hypoxic microenvironment susceptible to invasion by immune suppressant cells (26). Recent studies have demonstrated that tumor hypoxia is negatively associated with the efficacy of immunotherapy and that the stimulation of immune cell functions with PD-1 and CTLA4 blockade can also help to normalize tumor vessels (27, 28). In this regard, it is potentially feasible to assess the treatment efficacy of immunotherapy by evaluating the metabolism and microenvironment of the tumors before a reduction in the size of the tumor can be detected.
A few studies have reported that 18F-FDG-PET/CT can potentially predict early response to PD-1 blockade or CTLA4 blockade in patients with non–small cell lung cancer or melanoma (29, 30). However, several studies have also demonstrated the inability of 18F-FDG-PET/CT to distinguish patients with pseudoprogression from those with progressive disease in patients with melanoma (31, 32). Thus, to date, noninvasive imaging approaches to clarify the early response to effective immunotherapy in vivo have not been established.
The purpose of this study is to investigate the capability of noninvasive metabolic and physiologic imaging to evaluate early response to ICB therapy. New advancements in imaging using dynamic nuclear hyperpolarization (DNP) enhance the usually undetectable 13C signal by approximately four orders of magnitude in vivo (33). By using 13C pyruvate as a probe, it is possible to monitor the glycolytic profile of a tumor noninvasively, which has proven useful for monitoring the treatment response to several chemotherapeutics such as anti-angiogenic agent sunitinib and hypoxia-activated prodrug TH-302 (34, 35). Similarly, the [1,4-13C2] fumarate probe has been shown to detect necrotic tumor cell death in vivo (36, 37).
To test the applicability of these new imaging techniques along with conventional dynamic contrast enhanced (DCE) MRI in detecting ICB treatment response, we imaged mice bearing murine colon adenocarcinoma carcinoma (MC38) tumor or melanoma (B16-F10) tumor as ICB sensitive and less ICB sensitive models, respectively. Using the combination therapy of PD-L1/PD-1 and CTLA4 as a treatment model, which has an improved therapeutic effect compared with either monotherapy (38–40), multimodal metabolic and physiologic imaging showed substantial differences between the sensitive and less sensitive models and correlated with survival time within each model and therefore may serve as a guide for further development of imaging biomarkers.
Materials and Methods
Mice and tumor
Female C57BL/6 (B6) mice were supplied by the Frederick Cancer Research Center, Animal Production and housed in a specific pathogen-free environment and used at an age of 8 to 12 weeks. MC38 colon adenocarcinoma and B16-F10 melanoma were used as sensitive or less sensitive models to ICB therapy. MC38 was purchased from Kerafast. B16-F10 was used from the frozen stock in our lab and tested in February 2020 and authenticated by IDEXX RADIL using a panel of microsatellite markers. Molecular testing of cell lines for multiple pathogens, including Mycoplasma, was performed at the time of receipt and prior to in vivo studies. Both cell lines were maintained in DMEM containing 4.5 g/L glucose supplemented with 10% FCS and antibiotics.
In vivo checkpoint blockade treatment
For the in vivo treatment model, 1 × 105 MC38 tumor cells or 2 × 105 B16-F10 tumor cells were inoculated subcutaneously into the right leg of B6 mice. Tumor bearing mice were injected intraperitoneally with 200 μg of αPD-L1 (10F.9G2) or αCTLA4 (9H10) antibody or with isotype control antibodies (all from BioXcell) on days 9, 12, and 15 after tumor inoculation at the size of approximately 100 to 150 mm3. Imaging experiments were performed on day 13 or 14 after second injection of antibodies.
In vitro treatment
For in vitro treatment assay, tumor cells were cultured with 100 U/mL of recombinant murine IFNγ for 48 hours followed by 10 μg/mL αPD-L1 and αCTLA4 antibody treatment for an additional 24 hours before the assay.
To generate activated tumor-draining lymph node (TDLN) cells, B6 mice were inoculated subdermally with 1 × 105 MC38 tumor cells on both flanks to stimulate TDLNs. Thirteen days later, TDLNs (inguinal) were harvested, and single-cell suspensions were prepared mechanically. To analyze TILs, single-cell suspensions were prepared from solid tumors by digestion with a mixture of 0.1% collagenase, 0.01% DNase, and 2.5 U/mL hyaluronidase for 3 hours at 37°C. FITC-conjugated mAbs against CD4 (RM4–5); PE-conjugated mAbs against CD3; Cy-chrome-conjugated mAbs against CD8 (53–6.7) were purchased from BD Biosciences. The cell surface phenotypes were determined by direct immunofluorescence staining with conjugated mAbs and analyzed using FACS Calibur (BD Biosciences). Digested cells were stained with αCD3, αCD8, and αCD4 antibody. Data were collected on FACS Calibur flow cytometer. TILs were identified and gated on a forward scatter versus side scatter plot.
Western blot analysis
In vitro treated tumor cells were lysed on the dish with radioimmunoprecipitation assay buffer (RIPA buffer, Thermo Fisher Scientific) supplemented with protease and phosphatase inhibitors (Roche). In vivo treated tumor tissue was excised and homogenized in T-PER Tissue Protein Extraction Reagent (Thermo Fisher Scientific). Protein concentrations were measured by the bicinchoninic acid assay (BCA protein assay, Thermo Fisher Scientific). Glut-1, Hexokinase-2, lactate dehydrogenase A (LDHA), monocarboxylate transporter 1 (MCT1), MCT4 proteins were separated on 4% to 20% Tris-Glycine gel (Life Technologies) by SDS-PAGE and were transferred to nitrocellulose membrane. The membranes were blocked for 1 hours in blocking buffer (3% nonfat dry milk in 0.1% Tween 20/TBS), which was then replaced by the primary antibody (1:500–1:1,000), diluted in blocking buffer, and then incubated for 1 hours at room temperature. The membranes were then washed three times in washing buffer (0.1% Tween 20/TBS). The primary antibody was detected using the appropriate horseradish peroxidase conjugated secondary antibody and measured by the Fluor Chem HD2 chemiluminescent imaging system (Alpha Innotech Corp.). Density values for each protein were normalized to actin.
Seahorse metabolism assay
The extracellular acidification rate (ECAR) was analyzed on a XF96 Extracellular Flux Analyzer using XF Glycolysis Stress Test Kit (Agilent Technology) in accordance with the manufacturer's instructions. Briefly, in vitro treated cells on 96-well plates were first incubated in XF base medium without glucose or pyruvate. Glucose was injected and the ECAR is measured as the rate of glycolysis under basal conditions. Then, oligomycin, an ATP synthase inhibitor was injected and ECAR was monitored (maximum glycolytic capacity). Finally, 2-deoxy-glucose (2-DG), a glucose analog that inhibits glycolysis through competitive binding to glucose hexokinase, was injected to confirm that the ECAR produced in the experiment was due to glycolysis.
Tumor tissues were excised, frozen with Tissue-Tek O.C.T. compound (Sakura Finetek USA Inc.) by ultra-cold ethanol and sectioned (10 mm) using a cryostat, with the sections being thaw-mounted on glass slides. After fixing with 4% paraformaldehyde, sections were treated with cold acetone for 15 minutes. After blocking nonspecific-binding sites on sections with Protein Block Serum-Free reagent (Dako North America Inc.) for 30 minutes, the slides were covered by CD3 antibody (BD Biosciences; 1:250) combined with CD8 antibody (Abcam, Inc.; 1:250) overnight at 4°C. The sections were then incubated with Alexa Fluor 488 antimouse and the Alexa Fluor 546 F(ab′)2 fragment of goat antirabbit IgG (HþL; Invitrogen; 1:2,000) for 1 hours at room temperature, before being mounted with Prolong Gold antifade reagent with DAPI (Invitrogen). CD31 was stained with same procedure using CD31 antibody (BD Biosciences; 1:250) and Alexa Fluor 488 antirat (Invitrogen; 1:250). The sections were also stained with hematoxylin and eosin (H&E) and mounted on permount for histologic observation. The stained slides were scanned using a BZ-9000 microscope (Keyence), and the immunostain-positive area was quantified using ImageJ software (downloaded from https://imagej.nih.gov/ij/).
Hyperpolarized 13C MRI
Details of the hyperpolarization procedure were reported previously (34, 35). Briefly, [1–13C] pyruvic acid (30 μL) or [1,4–13C2] fumaric acid (2.5 M in 30 μL deuterated DMSO), containing 15 mmol/L Ox063 and 2.5 mmol/L of the gadolinium chelate ProHance (BraccoDiagnostics) was polarized in Hypersense DNP polarizer (Oxford Instruments). After the polarization reached 80% of the plateau value, the hyperpolarized sample was rapidly dissolved in 4.5 mL of a superheated alkaline buffer consisted of 40 mmol/L HEPES, NaOH, and 100 mg/L EDTA. Hyperpolarized [1–13C] pyruvate or [1,4–13C2] fumarate solution was rapidly injected intravenously through a catheter placed in the tail vein of each mouse (12 mL/g body weight). Hyperpolarized 13C MRI studies were performed on a 3 T dedicated MR Solutions animal scanner (MR SOLUTIONS Ltd.) using a 17 mm home-built 13C solenoid coil placed inside of a saddle coil tuned to 1H frequency. Both 1H and 13C were tuned and matched and anatomical image was taken after shimming on proton. 13C spectra was acquired every 1 second for 240 seconds from the whole leg including each tumor. The repetition time, spectral width, flip angle, and number of average were 1,000 milliseconds, 3,300 Hz, 10°, and 1, respectively.
Fumarase activity assay
A serum sample was obtained from isotype control antibody treated mice and from αPD-L1 and αCTLA4 antibodies treated mice one day after second injection. Fumarase activity was measured with the Fumarase Assay Kit (BioAssay Systems) in accordance with the manufacturer's instruction manual.
DCE-MRI of gd-DTPA
DCE-MRI studies were performed on a 1 T scanner (Bruker BioSpin MRI GmbH). T1-weighted fast low-angle shot (FLASH) images were obtained with TR = 156 milliseconds; TE = 4 milliseconds; flip angle = 45˚; four slices; 0.44 × 0.44 mm resolution; 15-second acquisition time per image; and 45 repetitions. Gd-DTPA solution (4 mL/g of body weight of 50 mmol/L Gd-DTPA) was injected through a tail vein cannula 1 minute after the start of the dynamic FLASH sequence. To determine the local concentrations of Gd-DTPA, T1 maps were calculated from three sets of Rapid Imaging with Refocused Echoes (RARE) images obtained with TR = 300, 600, 1,000, and 2,000 ms, with the acquisitions being made before running the FLASH sequence.
The significance of the differences between groups was analyzed using the Student t test for two groups and Tukey–Kramer test for more than two groups. Cohen d was calculated to assess the effect size of tumor reduction by the treatment. Cohen d = (Mean (Treatment) – Mean (Control)) / (SD (Treatment)2 + SD (Control)2) / 2)1/2. Kaplan–Meier curves were constructed for the survival of mice; the differences between groups were identified using the log-rank test. A two-tailed P value 0.05 was considered significant.
The animal experiments were conducted according to a protocol approved by the Animal Research Advisory Committee of the NIH (RBB-159-3E) in accordance with the National Institutes of Health Guidelines for Animal Research.
In vivo treatment effect of immune checkpoint inhibitors in a murine tumor model
As a model for the response to ICB in vivo, we evaluated the response of two xenograft models to either anti-PD-L1 mAb monotherapy or anti-PD-L1 mAb therapy in combination with anti-CTLA4 mAb. Mice were inoculated subcutaneously with either MC38 (colon adenocarcinoma) or B16-F10 (melanoma) cells into the right leg and treated with either or both immune checkpoint inhibitors by intraperitoneal injection every 3 days for a total of three injections.
By itself, anti-PD-L1 Ab therapy inhibited tumor growth in MC38 colon adenocarcinoma tumors (Fig. 1A) and extended survival times (Fig. 1B and C) compared with the control group (median survival time 29.5 days compared with 24 days without treatment, P = 0.0003). Combining anti-PD-L1 with anti-CTLA4 Ab treatment resulted in a stronger inhibition of tumor growth and a statistically significant increase in survival time relative to anti-PD-L1 Ab alone or the control group (33.5–29.5 or 24 days, P = 0.047, P < 0.0001, Fig. 1B). Anti-PD-L1 Ab + anti-CTLA4 Ab also delayed tumor growth (Fig. 1D) and extended survival in B16-F10 tumors (median survival time 24 days for the combination treatment compared with 19.5 days without treatment, P = 0.0018; Fig. 1E and F). However, the cytoreductive effect of anti-PD-L1 Ab + anti-CTLA4 Ab was less in B16-F10 than in MC38 tumors (Cohen d: −2.88 to −3.88; Supplementary Fig. S1). The survival benefit from anti-PD-L1 Ab + anti-CTLA4 Ab was also less in B16-F10 compared with the MC38 cell line (mean survival benefit 4.2–12.1 days) These results are consistent with previous reports showing that, while both cell lines respond to a statistically measurable degree to ICB treatment, MC38 displays a greater response (more sensitive) than B16-F10 (41).
This difference in sensitivity to ICB therapy was not reflected in T-cell counts (Fig. 2A). Intratumor levels of CD3+CD8+ T cells increased substantially in both tumor models after 4 days of anti-PD-L1 Ab + anti-CTLA4 Ab treatment, suggesting effective infiltration of lymphocytes into the tumors. ICB treatment raised levels of CD3+CD4+ and CD3+CD8+ T cells in the tumor draining lymph nodes in MC38 tumors, but the populations of both types of cells was already at a similar high level in the absence of treatment in B16-F10 tumors. Because no change was observed upon treatment in the B16-F10 model, the fraction of CD3+CD4+ and CD3+CD8+ cells in the tumor draining lymph nodes is not likely to serve as a strictly generalizable marker for treatment response. To more quantitatively measure total and activated T cells in situ, histologic assessment was performed on tumor slices to confirm T cells proliferation. Immunofluorescence staining of CD3 and CD8 showed that infiltration of CD3+ T cells and CD8+ cytotoxic T cells significantly increased in ICB treated MC38 tumors compared with isotype control Ab treated MC38 tumors (P < 0.0001 and 0.0001, Fig. 2B). In B16-F10 tumors, the absolute number of lymphocytes was generally smaller than in MC38 tumors. The infiltration of CD3+ T cells and CD8+ cytotoxic T cells was significantly higher in both tumor models treated with ICB combination therapy compared with the corresponding tumors treated with the isotype control Ab treated (P < 0.0001 and 0.0001, Fig. 2C). These findings confirmed an active immune response as early as day 4 after the ICB treatment in both tumor models. The CD8 to CD3 ratio was similar between groups in both tumors, suggesting a similar level of activation between CD4 and CD8 T cells (Fig. 2B and C).
Changes in tumor glycolytic metabolism induced by ICB
Activation, proliferation, and differentiation of T cells all require the synthesis of numerous macromolecules. To generate synthetic intermediates to support these processes, naïve T cells often undergo metabolic reprogramming to enhance glycolytic flux (42). Decreases in glycolytic flux therefore could serve as a potential biomarker to predict ICB efficacy. To investigate whether glycolytic metabolism of tumor cells is altered directly by PD-L1 blockade, we first measured the in vitro extracellular acidification rate (ECAR) of tumor cells by the Seahorse assay to determine the extent of lactic acid fermentation (Fig. 3A). Anti-CTLA4 Ab was added to anti-PD-L1 Ab so that the treatment regimen, although not the system itself, is comparable with the in vivo experimental model. The ECAR of MC38 cells in vitro was not altered after ICB treatment. In contrast, the ECAR of the less ICB sensitive B16-F10 cell line was significantly reduced after ICB treatment (Fig. 3A), suggesting that the glycolytic metabolism of B16-F10 is more dependent on the PD-L1/PD-1 pathway than that of MC38. To investigate whether expression of glycolytic enzymes and transporters are affected by PD-L1 blockade, PD-L1 inhibitor treated tumor cells were grown and treated in vitro and analyzed by Western blotting. No significant changes in the expression of glycolytic enzymes and transporters were found after PD-L1 blockade in either cell line (Fig. 3B; Supplementary Fig. S2). Because the tumor cells themselves only form part of the tumor microenvironment, this experiment was repeated in vivo with tumors treated with anti-PD-L1 Ab + anti-CTLA4 Ab and excised from mice (Supplementary Fig. S3). Mirroring the in vitro experiment, no significant changes were observed glycolytic enzyme levels in MC38 tumors. In B16-F10, the expression of specific glycolytic enzymes including lactate dehydrogenase A (LDHA), Glut1, hexokinase 2, and MCT4 was reduced after ICB treatment. However, these differences were not statistically significant at the α = 0.05 level. Therefore, downregulation of glycolytic enzymes in response to PD-L1 treatment is a possible, but unproven, factor in the decrease in glycolytic flux in B16-F10.
To evaluate changes in glycolysis upon ICB treatment as a potential imaging biomarker for ICB response, we measured the flux through the lactate dehydrogenase pathway in vivo by hyperpolarized 13C MRI using [1–13C] pyruvate. Nonlocalized one-dimensional spectra encompassing the entire tumor implanted leg were acquired continuously for 240 seconds after the injection of hyperpolarized [1–13C] pyruvate and the conversion of pyruvate to lactate was evaluated quantitatively (Fig. 3C). The lactate/pyruvate ratio is a direct measure of flux through the lactate dehydrogenase pathway (43) and an indirect measure of glycolysis. In the ICB sensitive MC38 tumors, ICB treated tumors showed lower Lac/Pyr ratios compared with control tumors, although the difference was not statistically significant (median Lac/Pyr ratio 0.80 for the ICB treatment compared with 0.90 for the control, P = 0.075, Fig. 3D). A cluster of large tumors with high Lac/Pyr ratios likely undergoing anaerobic fermentation is evident in the control group (Fig. 3E). No correlation of the Lac/Pyr ratio with the treatment group was found when controlled for tumor size to remove this effect (P = 0.13; Supplementary Table S1A). The decrease in glycolytic flux after ICB treatment was more marked in B16-F10 tumors. The Lac/Pyr ratio was significantly lower in the ICB treatment group relative to the control group (median Lac/Pyr ratio 0.61 for the ICB treatment compared with 0.82 for the control, P = 0.0073, Fig. 3F). Even when controlled for tumor size, the Lac/Pyr ratio remained negatively correlated with the treatment with statistical significance (P = 0.018, Supplementary Table S1B and Fig. 3G). The bicarbonate/pyruvate ratio was similar between treated and untreated control groups in both tumors (Supplementary Fig. S4). Taken together, although these results suggested that Lac/Pyr tended to be lower in anti-PD-L1 Ab + anti-CTLA4 Ab treated tumor compared with control tumors in both tumor models, the Lac/Pyr ratio did not reflect the effect size of ICB therapy (larger in MC38 compared with B16-F10).
Necrotic change induced by ICB
Tumor cell necrosis is potentially a more direct biomarker for ICB efficacy. To evaluate necrotic cell death in tumors, hyperpolarized 13C MRI using [1,4–13C2] fumarate was performed on the same sets of tumors. Fumarate is not actively transported into the cell and the plasma membrane is impermeable to fumarate when intact. A loss of membrane integrity during necrosis allows entry of fumarate and the conversion to malate by cytosolic fumarate hydratase, which can be detected as a doublet distinct from fumarate at 182 to 183 ppm. Figure 4A shows the representative dynamic 13C spectra of ICB sensitive MC38. Figure 4B shows the malate to fumarate (Mal/Fum) ratio of MC38 tumor bearing mice sorted by treatment group. Mal/Fum ratios were significantly higher in ICB treated MC38 tumors relative to the control group (median Fum/Mal ratio 0.091 for the ICB treatment compared with 0.055 for the control, P = 0.0006, Fig. 4B). The Mal/Fum ratio remained a significant variable even when tumor size was controlled (P = 0.013, Supplementary Table S1C; Fig. 4C), suggesting that tumor cell necrosis is induced as a treatment effect of ICB therapy in vivo. Mal/Fum was slightly higher in the ICB treated B16-F10 tumors (Fig. 4D), but the difference was not significant (median Fum/Mal ratio 0.096 for the ICB treatment compared with 0.081 for the control, P = 0.21, Fig. 4E). Multiple regression analysis confirmed that Mal/Fum ratio was not a significant predictor of the treatment group (P = 0.42, Supplementary Table S1D; Fig. 4F). No difference in serum fumarase activity was found between ICB treated MC38 mice and the control (P = 0.72, Fig. 4G), suggesting that serum fumarase activity cannot be used as a surrogate marker to monitor ICB efficacy.
To confirm the necrotic change by histology, tumor sections were stained with H&E according to standard procedures. In MC38 tumors, infiltration of lymphocytes was confirmed by focal areas of hematoxylin negative spaces (dot-line area, Fig. 5A). Nucleic enlargement in tumor cells (white arrow) indicating necrotic cell damage was also observed (Fig. 5A). The overall decrease in hematoxylin positive area, indicating the replacement with interstitial tissue after tumor cell death, was quantified using ImageJ software, and found to be significantly increased in ICB treated MC38 tumors compared with the control (P = 0.027, Fig. 5B; H&E staining of all the samples in Supplementary Fig. S5). These histologic findings support the necrotic cell death as a consequence of antitumor effect of lymphocytes induced by ICB therapy. The increase in necrotic area and lymphocyte infiltration was also evident in the less ICB sensitive B16-F10 tumors. Such changes were observed more diffusely in B16-F10 tumors. However, the decrease in hematoxin positive level was milder in B16-F10 tumors, and the difference between treatment and control groups did not meet the test for statistical significance (P = 0.13, Fig. 5B). The histologic difference in the level of tumor cell death by the cell line is consistent with the difference in Mal/Fum ratio detected with hyperpolarized 13C MRI using [1,4–13C2] fumarate.
Changes in intratumor blood perfusion induced by immune checkpoint inhibitors
Tumor perfusion/permeability changes in response to ICB treatments were investigated by DCE-MRI using the Toft model. In the Toft model, permeability changes are characterized by the influx forward volume transfer constant (Ktrans) from plasma into the extravascular-extracellular space (EES), which reflects the sum of all processes (predominantly blood flow and capillary leakage) that determine the rate of gadolinium influx from plasma into the EES. We examined isotype control Ab treated and anti-PD-L1 Ab + anti-CTLA4 Ab treated tumor bearing mice after the second Ab injection. Relative to the control, Gd-DTPA uptake was increased (Fig. 6A and B) and Ktrans was significantly higher in ICB treated MC38 tumors (P = 0.0025, Fig. 6C). The separate effects of perfusion and an increase in the volume of the EES through tissue clearance can be distinguished by the AUC of the calculated Gd concentration at short (1 minute) and long (10 minutes), respectively. Significant increases in both perfusion and permeability were observed (P = 0.0037, P = 0.0029, Fig. 6D). Changes in the ICB less sensitive B16-F10 tumors were relatively less (Fig. 6A). ICB treated tumors showed higher Ktrans but did not meet the strict criterion for statistical significance (P = 0.052, Fig. 6E). Perfusion (AUC 1 minute) was significantly higher in ICB treated B16-F10 tumors, whereas no difference in the equilibrium (AUC 10 minutes) was detected, suggesting a change in the perfusion without a significant volume increase in the EES (P = 0.021 and 0.25, respectively, Fig. 6F). These findings suggest that ICB treatment improved blood perfusion in MC38 tumors, but caused only limited changes in B16-F10 tumors.
To examine the changes in blood vessel density, IHC was performed with tumor sections. Supplementary Figure S6 shows the staining of CD31, which is a marker of vascular endothelial cells, as well as the staining of H&E in same section. Compared with the control tumor, CD31 staining was increased in ICB treated MC38 tumor (Supplementary Fig. S6A). CD31 positive area increased notably in hematoxylin less staining area where tumor cells are less (solid-line area), whereas CD31 positive area were poor in hematoxylin dense area where tumor cells are rich (dot-line area) in both control Ab and ICB treated tumors (Supplementary Fig. S6A). Quantification of CD31 positive area showed a significant increase in ICB treated tumor (P < 0.0001, Supplementary Fig. S6B). In contrast, no significant difference was observed in CD31 staining for B16-F10 tumors (P = 0.81, Supplementary Figs. S6C and S6D). These findings suggest a difference exists between MC38 and B16-F10 tumors in the improvement of vascular flow by ICB therapy, which supports the findings of tumor perfusion/permeability differences in DCE MRI.
Correlation between imaging biomarkers and clinical outcome
The previous experiments have the implicit assumption that after treatment a valid biomarker will change significantly in the ICB sensitive model (MC38) but not in the less sensitive model and control groups. An invalid biomarker for ICB, by contrast, either not change significantly after treatment in model known to be ICB sensitive model or the magnitude of the change will be similar in the nontreated control groups and less sensitive models. Although useful for probing the underlying metabolic and physiologic changes that occur during treatment, this paradigm does not strictly correspond to the clinical situation where it is usually the ability of biomarkers to predict progression within similar tumor types that is primarily of interest. To investigate each imaging biomarker's prognostic capability, a part of the mice scanned in hyperpolarized 13C MRI or DCE-MRI experiments in Fig. 3C–G, Fig. 4A–F, or Fig. 6 were followed until their clinical endpoint. As expected from Fig. 1, there was strong separation for both B16-F10 and MC38 from the control groups (Fig. 7A). No correlation of survival was observed for any biomarker with the control group. Correlations were relatively strong but with significant scatter (Supplementary Table S2B). Single regression analysis showed that the Lac/Pyr ratio had a significant negative correlation with survival in B16-F10 tumors (R2 = 0.59 ± 0.06 and P = 0.02, Supplementary Table S2B). Ktrans was also significantly positively correlated with survival in MC38 tumor and B16-F10 tumors (Fig. 7A, R2 = 0.76 ± 0.01 and P = 0.02, Supplementary Table S2B).
Some of the single time point studies suggested tumor volume may be a complicating factor in the relationship between metabolic markers and ICB response (Supplementary Table S1B). Because tumor volume is easily obtainable from an anatomical MRI, we constructed multilinear regression prediction models for survival using the imaging biomarker, the tumor volume, and the cell line as a categorical variable. Both Ktrans and ve were included in the DCE model (Fig. 7B). Correlation significantly improved in each case. The least accurate predictor, the Lac/Pyr ratio, still obtained an R2 of 0.6, although the Lac/Pyr variable itself was not statistically significant (P = 0.13, β = −18.6). The Mal/Fum ratio (R2 = 0.81) and the DCE Ktrans/ve (R2 = 0.84) appear to be better predictors of survival than the Lac/Pyr ratio in this model.
In this study, we investigated the capability of noninvasive metabolic and physiologic imaging to evaluate the early response to ICB therapy in tumor bearing animal models. To our knowledge, this is the first report of metabolic imaging assessment to detect the early response to ICB therapy using hyperpolarized 13C MRI.
Expression of PD-L1 on cancer cells promotes glucose uptake and production of lactate as part of the metabolic programming that enhances the survival of cells in hypoxic conditions and provides fuel for further growth. The exact mechanism by which this occurs is not fully understood. Previous reports have shown that blocking PD-L1 may directly dampen tumor glycolysis by inhibiting mTOR activity (24). In vivo tumor metabolism when evaluated by 13C pyruvate MRI showed a significant decrease in glycolysis in B16-F10 tumors after ICB treatment yet had little independent effect on MC38 after adjustment for tumor size, despite the greater impact of ICB on tumor volumes and survival time on this cell line (Fig. 3C–G). Within the same cell line, Lac/Pyr emerged as the least efficient of the imaging biomarkers chosen for predicting treatment outcome (Fig. 7B). The reasons for the relative insensitivity of Lac/Pyr as biomarker for treatment are not entirely clear. The in vivo tumor microenvironment consists of tumor cells with substantial contributions from the tumor stroma and immune cells. The expression of PD-1 on T cells is reported to inhibit glycolysis or amino acid metabolism and upregulate fatty acid oxidation, leading to impaired energy generation, which compromises proliferation and effector functions (44). CTLA4 can also cause decreased expression of GLUT1, increased mitochondrial oxidation and fatty acid uptake, and decreased biosynthesis on T cells (45, 46). Therefore, inhibition of PD-1/PD-L1 and CTLA4 would preferentially promote T-cell function. ICB treatment might restore the metabolic balance in the competition between tumor and immune cells in favor of T cells, leading to a less glycolytic profile when the tumor is considered as a whole. In our study, glycolytic flux in MC38 tumors was not affected by ICB when only tumor cells are considered whereas the glycolytic flux in B16-F10 tumors was significantly altered (Fig. 3A), in line with previous reports that show the higher glycolytic activity in B16-F10 tumors results in high acidification of the tumor microenvironment when compared with MC38 tumors (47). These suggest that B16-F10 is more dependent on glycolytic metabolism for energy production than MC38. Moreover, a greater decrease in lactate production (Fig. 3F) and in the expression levels of glycolytic enzymes and transporters in tumors after ICB treatment (Supplementary Fig. S3) suggest that the glycolytic metabolism in B16-F10 may be more susceptible to ICB therapy. In this respect, 13C pyruvate MRI may detect the metabolic shift by ICB therapy in selected cancers whose metabolism strongly relies on glycolysis.
13C fumarate MRI showed significantly increased Mal/Fum in ICB treated tumors relative to control tumors, suggesting necrotic tumor cell damage was induced in response to ICB therapy (Fig. 4). The Mal/Fum ratio also emerged as a strong predictor of survival time within the same cell line when combined with the tumor volume (Fig. 7B). The utility of the malate production from hyperpolarized [1,4–13C2] fumarate probe as a sensitive indicator of tumor cell death has been previously shown in several tumor types treated with either a cytotoxic agent, anti-VEGF drug, or multikinase inhibitor (36, 48, 49). In contrast to chemotherapy, which directly shows cytotoxicity on tumor cells, the effect of ICB is induced via activation of immune system. Blockade of immune checkpoints such as PD-L1, PD-1, CTLA4, whose signals help keep T cells from killing cancer cells, release the “brakes” on the immune system and enable T cells to activate and show cytotoxicity. There are two main mechanisms involved in cytotoxic T cell-mediated tumor cell death: one elicited by granule exocytosis (perforin and granzymes) and the other via the death ligand/death receptor system like Fas ligand (FasL) and TRAIL. It has been assumed that the final consequence of granule exocytosis is the induction of cell death by apoptosis. However, recent experimental evidence indicating that perforin and granzymes of cytotoxic lymphocytes can activate non-apoptotic pathways of cell death (50). Recently, it was reported that death receptor can induce a mixed mechanism of cell death called “necroapoptosis” (51). Necroapoptosis is a programmed form of necrosis, or inflammatory cell death, whereas necrosis is associated with unprogrammed cell death resulting from cellular damage. In necroapoptosis, rupture of the cell membrane and excretion of intracellular substrates occur as in necrosis. Thus, we assume that mixed mechanisms of tumor cell death including a mechanism similar to necrosis allows 13C fumarate MRI to detect tumor cell death in response to ICB therapy in our study. Mal/Fum showed a significant increase in the treated group of ICB sensitive MC38 model (Fig. 4B), and a much smaller change in the less ICB sensitive B16-F10 model (Fig. 4E). The results suggest that 13C fumarate MRI detects treatment effect of ICB therapy more directly and accurately than 13C pyruvate MRI.
Treatment effects were also detectable by measurements of intratumor perfusion/permeability, by DCE MRI (Fig. 6). Previous studies have shown that abnormal tumor vasculature contributes to immune suppression through multiple mechanisms. Hypoxia promoted by impaired vessel perfusion not only increases the recruitment of immunosuppressive cells through the secretion of cytokines such as TGFβ, VEGF, and IL10, but also upregulates the expression of CTLA4 or LAG3 on regulatory T cells and the expression of PD-L1 on myeloid-derived suppressor cells, tumor-associated macrophages, and tumor cells (26, 52, 53). Treatment with dual anti-CTLA4 and anti-PD1 therapy induces tumor vessel normalization through a dynamic process that involves various immune populations at different stages of tumor development (26, 28). Although the precise mechanism remains unknown, it is reported that type 1 helper T cells play a crucial role in vascular normalization by ICB (28). An increase in tumor vascularity during immune rejection has previously been detected by DCE-MRI in a E.G7-OVA tumor-bearing mice model, which undergoes spontaneous regression (54). As for ICB therapy, there was one report showing that DCE-MRI showed no alteration in the relative blood distribution volume after anti-PD-L1 and anti-CTLA4 AB treatment in a CT26 tumor (55). In contrast, another report gave evidence of vascular normalization after ICB therapy as they showed the number of functional vessels covered by pericytes increased after dual blockade of PD-1 and CTLA4 (28). In this study, Ktrans increased in treated MC38 and B16-F10 tumors relative to control tumors and the increase for each cell line was parallel in magnitude to the corresponding treatment effect (Fig. 6C and E).
Quantitative analyses of DCE-MRI has been used in previous clinical studies and found to be useful in the assessment of anticancer therapies as well as in the prediction of eventual response in a variety of cancers (56, 57). Routine clinical use of DCE is widespread (58, 59). However, many factors determine the accuracy and precision of the parameter values obtained from DCE MRI including the estimation of the arterial input function, spatial and temporal resolutions, pharmacokinetic models, and curve fitting strategies (60). Thus, further investigation is warranted for determining the efficacy of ICB therapy in a clinical setting.
Several limitations of this study should also be acknowledged. First, we could not monitor changes in individual mice over time due to the potential invasiveness of repeated probe injections. The reaction to ICB is not instantaneous and it takes at least a few days for immune system to be activated. During this period, the tumor was still growing, which may complicate the interpretation of metabolic or physiologic changes. Next, in this study, we used vehicle treated mice as a negative control for ICB treatment. The use of vehicle treated mice allows the comparison of immune-activated models with non-activated ones without considering other confounding factors. However, by not using a completely refractory tumor model as a control, treatment effects that cause changes in imaging biomarkers may be overlooked if they are caused by mechanisms unrelated to immune activation. Further study on ICB refractory models would be needed to examine the possibility. Finally, the subcutaneous tumors used are not ideal because the unnatural tumor microenvironment may result in different metabolic profiles in comparison to more orthotopic models. Further investigation using orthotopic xenograft with multiple cell lines or a spontaneous tumor model is needed to further validate the biomarkers proposed here.
No disclosures were reported.
Y. Saida: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft. J.R. Brender: Data curation, validation, visualization, methodology, writing–review and editing. K. Yamamoto: Validation, visualization. J.B. Mitchell: Writing–review and editing. M.C. Krishna: resources, supervision, funding acquisition, writing–review and editing. S. Kishimoto: Conceptualization, data curation, formal analysis, investigation, methodology, writing–review and editing.
This research was supported by the Intramural Research Program, Center for Cancer Research, NCI, NIH (grant no. Z01BC010477). Y. Saida is a Japan Society for the Promotion of Science (JSPS)-NIH Research Fellow.
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.