Induced Tumor Heterogeneity Reveals Factors Informing Radiation and Immunotherapy Combinations

Immune checkpoint therapies are transforming treatment paradigms in many cancers with positive clinical trials not only in metastatic disease but also in the curative intent setting. For example, the Pacific trial demonstrated a significant survival benefit for adjuvant PD-L1 blockade after chemoradiation in locally advanced lung cancer (1). Although focal radiation could provide dramatic results for metastatic disease when combined with immunotherapy, in humans there is limited evidence this occurs beyond a small proportion of patients with immunogenic tumors (2–5). However, with more studies evaluating immunotherapy in the definitive setting, the combination of radiation, chemotherapy, and immunotherapy is underway (1, 6). Therefore, understanding responses to combination therapy is critical to optimize local and systemic control of disease in the nonmetastatic setting. Identification of factors that enhance immune responses after radiation and/or chemotherapy may reveal combination therapies that license immune responses of immunologically cold tumors. Because radiotherapy is utilized in over half of all solid tumors for definitive treatment, radiotherapy and optimal immunotherapy combinations may be impactful. To this end, characterization of the responses of the complex tumor microenvironment (TME) is essential to drive advancements in the treatment of immunologically unresponsive tumors using radiotherapy.

Radiotherapy can induce adaptive immune responses and result in enhanced efficacy of immune checkpoint blockade; unfortunately, many tumors are not immunologically responsive to radiotherapy at baseline (3, 7–9). Dendritic cell activation is key for effective adaptive immune responses through presenting tumor antigens, detecting pattern recognition molecules that activate innate pathways and sensing cytosolic DNA (10–12). However, understanding how radiotherapy induces productive immune responses through this critical step is often bypassed in preclinical studies that use tumor cells that express model antigens or innate ligands that trigger dominant adaptive immune responses (12–15). Furthermore, since radiotherapy can induce tolerogenic cell death instead of immunogenic cell death, neoantigens may not be dominant and dendritic cell activation may not subsequently occur (16). Thus, a better understanding of different competing signals in the microenvironment is critical.

In this study, we modeled intrinsic intratumoral immune heterogeneity by mixing clonal tumor cells that induce different immune microenvironments prior to tumor implantation. With this approach we determined how different mixtures of tumor cells can induce an immune licensing phenotype over a dominant suppressive phenotype. Analysis of the microenvironmental factors that induced a licensing effect indicated that cell-intrinsic factors such as induced antigen presentation and infiltrating myeloid cell phenotype were associated with immune licensing. These findings led us to hypothesize that the CD40 agonist (anti-CD40) antibody in combination with radiotherapy can transform immunologically unresponsive tumors. Interestingly, we found that CD40 activation was robust and was associated with increased dendritic cells in the draining lymph node and subsequent T-cell infiltration in the tumor. In immunologically unresponsive tumors, our data suggests that an immune licensing agent such as anti-CD40 could overcome the limited responses of checkpoint therapy alone when combined with radiotherapy and efforts are underway to validate this approach prospectively in patients.

Py8119 and Py117 clonal cell lines derived from transgenic MMTV-PyMT mice congenic in the C57Bl/6 background were used between passage numbers 15–25. Cells were confirmed to be mycoplasma free and cultured in Hams F12K media, 5% FBS, MITO (1:1,000 dilution, Corning), 50 μg/ml gentamicin, and 2.5 μg/ml amphotericin B. Cells were derived as described previously and were not selected in any specific way beyond clonal dilution (17, 18). 4T1 cells expressing GFP and luciferase were attained by the Contag Lab (Stanford) and grown in RPMI 10% FBS.

Orthotopic implantation of 1 × 10⁶ PyMT cells in the mammary fat pad #4 in 50% PBS and 50% Matrigel without phenol red and low growth factor (Corning). For mixed ratio experiments a total of 1 × 10⁶ cells were injected with ratios of cells mixed into Matrigel or as described in the text and figures prior to implantation. A total of 2 × 10⁵ 4T1 cells were used for orthotopic implantation in PBS. Luciferase imaging of 4T1 tumors was performed using an IVIS-200 10 minutes after intraperitoneal injection of 3.3 mg of d-luciferin (Biosynth Chemistry & Biology)

Irradiation of tumors 7–10 days after implantation at the identified dose was performed using a 225 kVp cabinet X-ray irradiator with a 0.5 mm Cu filter (IC-250, Kimtron Inc.). Mice were anesthetized using 80 mg/kg ketamine hydrochloride and 5 mg/kg xylazine intraperitoneally and the tumors were shielded with a 3.2 mm lead shield with a 15 × 20 mm aperture. Anti-CD40 agonist Ab (FGK4.5, BioXCell) was administered 100 μg i.p. the starting the day of radiotherapy and repeated for 4–5 doses every 3 days. PD-1 blocking Ab (RMP1-14, BioXCell) was administered 200 μg i.p. starting the day of treatment and repeated for 4–5 doses every 4 days. Isotype control IgG2a Ab (2A3, BioXCell) was administered 100 μg i.p. starting the day of treatment and repeated for 4–5 doses every 3 days. CSF-1 blocking antibody (5A1, BioXCell) was administered 100 μg i.p. starting the day of treatment and repeated for 4–5 doses every 4 days.

Tumors were harvested after treatment and dissociated by chopping tissue finely and suspended in 1:1 F12K media and DMEM. Tumors were digested with Liberase TL at 20 μg/ml (Roche) and DNAse at 400 μg/ml (Roche) for 40 minutes at 37°C then filtered through a 40-μm mesh after diluting in media with 5% FBS. Tumor digestions for Fig. 1 were performed using collagenase type I at 200 U/mL (Worthington) and Dispase at 0.5 U/mL (StemCell Technologies). Cells were resuspended in RBC lysis buffer for 10 minutes at room temperature then counted and diluted in PBS. Cells were then treated with Zombie NIR (BioLegend), CD16/32 Ab (BioLegend) for FC blocking and cell surface antibody panel for 20 minutes on ice. Cells were washed two times then fixed and permeabilized using eBiosciences Foxp3/Transcription factor Staining Buffer set then labeled with intracellular antibodies including Arg1, iNOS, Ki67, and FoxP3 for 20 minutes.

Flow cytometry was performed on a four laser LSRII (BD Biosciences) or a LSRFortessa X-20 (BD Biosciences). In all analyses, at least 100,000 events of 2–4 million cells that were stained were recorded to the determine the relative abundance of respective cells per total live cells or CD45⁺ cells. Analysis was performed using FlowJo v10 Software (Tree Star). Rat/Hamster compensation beads (Life Technologies) and cells stained with live dead stain were used for compensations. The following antibody clones targeting mouse epitopes were used for analysis. BioLegend: CD45 (1:160, 103133, 30-F11), CD3 (1:80, 100237, 17A2), CD4 (1:400, 100406/100432, GK1.5), CD8 (1:160, 100708, 53-6.7), CD69 (1:100, 104509, H1.2F3), CD11b (1:400, 101243, M1/70), CD11c (1:200, 117310, N418), F4/80 (1:60, 123110/123133, BM8), MHCII IA I-e (1:160, 107606, M5/114.15.2), GR-1 (1:160, 108408, RG6-8C5), and PD-L1 (1:160, 124310, 10 F.9G2). BD Biosciences: MHCI H2Kb (1:160, 553570, AF6-88.5), Arg-1 (3 μL per sample, IC5868A, and Ki-67 (4 μL per sample, 561126, B56). eBiosciences: FoxP3 (1:100, 61-5773, FJK165), TCR-β (1:200, 45−5961, H57−597), and NOS2 (1:500, 25-5920, CXNFT). Gating strategy for immune cells, myeloid cells in Py117 and Py8119, and T cells in Py117 tumors are shown in Supplementary Fig. S1A–S1C. Gating strategy for leukocytes in the draining lymph nodes are show in Supplementary Fig. S2A–S2C.

Py117 and Py8119 tumors implanted into C57Bl/6 mice and one group was irradiated with 12 Gy or a second group was not irradiated when tumors were 75–125 mm³. Tumors were harvested 10 days after radiation and tumor tissue was homogenized in RIPA lysis buffer with protease inhibitor 100 mg of tissue per 500 μL of buffer. Supernatants were collected after centrifugation and frozen at −80°C. On day of assay samples were thawed on ice and diluted to 1 μg/μL after measurement by BCA assay.

Mouse 38-plex Luminex immunoassays (Affymetrix) were used according to manufacturer's recommendations and processed by the Stanford Human Immune Monitoring Core. Samples were further diluted 3-fold to a final volume of 60 μL and incubated with antibody linked beads in a 96-well plate the incubated for 1 hour at room temperature followed by overnight incubation at 4°C. Samples were further process as previously described (18).

For tissue culture microarray Py117 cells and Py8119 cells were plated in 6-cm dishes with 3.33 × 10⁵ cells, media were changed after 24 hours and cells were lysed on ice in RIPA buffer at 48 hours. For homogenates, tumors were harvested from mice on the day radiotherapy would typically occur for the Py117, Py8119, and the Py5050 mix tumors then tissue was homogenized 100 mg per 500 μL centrifuged, and RNA was extracted from the supernatants. Samples were measured by the Bioanalyzer QC to verify the quality, underwent in vitro transcription, fragmentation, labeling, hybridization, and scanning performed by the Stanford Protein and Nucleic Acid Facility. The Affymetrix mouse gene 2.0 ST whole transcriptome array was used for all samples. All microarray data is uploaded to GEO accession ID GSE118181.

The array data was analyzed using R\Bioconductor package oligo and limma version 1.42.0 and 3.36.1, respectively, RMA for normalization (19–21). All differential gene expression was reported at corrected P value equal or less than 0.01 and fold change equal or greater than 2. Annotation and collapse dataset to gene symbols were determined using annotation package specifically for Mouse gene 2.0 ST array mogene20sttranscriptcluster.db in R. Hierarchical clustering and heat map visualization was generated using ComplexHeatmap version 1.17.1 in R (22). Cell-intrinsic genes and genes induced in microenvironment were determined as following. First, we compared 117 and 8119 cells in culture and list of genes that were differentially expressed named as differentially expressed (DE) genes at cell level, and all other genes named as non-DE at cell level. Then, we compared tissues from Py117, Py5050, and Py8119 and differential gene expression was calculated as we mentioned above. After that, we filtered these genes by DE genes at cell level and shared genes were named cell-intrinsic genes. We then filtered by non-DE at cell level and shared genes were named genes induced in microenvironment. Gene pathways enriched in cells and tumors were determined using the Gene Set Enrichment Analysis (GSEA) v3.0 (23), and enriched terms were selected with an FDR cutoff of P ≤ 0.25. Principal component analysis (PCA) was calculated in R and visualized using pca3d package version 3.4.4. Gene ontology networks were generated and visualized using the ClueGO v2.5.1-CluePedia v.1.5.1 Cytoscape (v3.3.6) plugin (24, 25).

Transcriptomic data from PD-1 responsive and nonresponsive melanoma from Hugo and colleagues, 2016 was obtained from accession number GEO: GSE78220. Data were reanalyzed using STAR and DEseq2 pipeline and GRCh38 as a reference genome (26). Gene enrichment was obtained using GSEA Desktop v3.0 as we described above. TCGA transcriptomic data of patients with Breast (BRCA), Skin (SCKM), and pancreatic (PAAD) cancer along with clinical data were obtained using TCGAbiolinks R package (version 2.9.5) for harmonized data (27). Before calculating survival curves, data were cleaned from any censored incident before first death. Samples from BRCA male patients were removed. Patients with neuroendocrine carcinoma were removed from PAAD data. Multivariate Cox regressions was performed with the coxph function from the R survival library, Kaplan–Meier plot, a log-rank P value were plotted using survfit and ggsurvplot R packages. Correlation between CD40, CD40LG, and MAP3K14 expression level and infiltration of immune cells and tumor purity were calculated and plotted using TIMER (28, 29). Oncolnc data for CD40LG was obtained from TCGA BRCA, SKCM, and PAAD from oncolnc.org (30). Chen and colleagues normalized NanoString values were obtained from Supplementary tables (31). Riaz and colleagues normalized RNA sequencing data (RNA-seq) data was obtained from GEO: GSE91061 and significance was measured by a two-sided t test (32).

Flow cytometry and Luminex multiplex cytokine arrays were quantified using an ANOVA model and post hoc comparison with a Tukey adjustment using GraphPad6 (GraphPad Software Inc). Tumor growth curves were analyzed in a repeated measure mixed model allowing for errors to be correlated within mice using R (R Core Team).

All animal studies were approved by the Stanford Institutional Animal Care and Use Committee protocol number 9984.

All relevant data is available from the authors. All microarray data is uploaded to GEO accession ID GSE118181. PyMT cells are available through ATCC or MTA with UC San Diego.

We previously described differences in the epithelial to mesenchymal transition (EMT) phenotype and AXL receptor tyrosine kinase expression between the more immunogenic Py117 and immune resistant Py8119 PyMT tumors derived from the MMTV-PyMT mouse model (18). Using this model, we investigated how mixed tumor clones influence immune infiltrates in the microenvironment and studied the impact of immunologic heterogeneity on therapeutic responses. We hypothesized that the Py117 cells would cause a sensitizing or licensing effect over the suppressive effect of Py8119 cells. Orthotopic tumors with varying ratios of immunologically resistant Py8119 and sensitive Py117 tumor clones were implanted as depicted in Fig. 1A. Seven days after tumor implantation, tumors were treated with 12 Gy and exhibited tumor growth delay by radiotherapy that increased proportionally to the number of Py117 cells. The expected tumor growth curves were modeled using responses derived from single clone growth curves. Interestingly, the response of tumors with 50% Py117 cells were greater than predicted from modeling and the addition of Py117 cells to tumors with the same number of Py8119 lead to greater response compared with Py8119 alone, supporting a dominant sensitizing effect (Fig. 1B and C). In concordance with our past study, when mixed tumors were implanted into athymic nude mice, tumor growth increased and radiation response diminished in the mixed tumors, indicating the importance of the immune response (Supplementary Fig. S3A; ref. 18).

To evaluate the immunologic response of these tumors in more detail, tissues were dissociated and analyzed by flow cytometry. The proportion of CD45⁺ leukocytes was unchanged across the 20%–100% Py8119-containing tumors (Supplementary Fig. S3B). However, tumors with the greatest radiotherapy response displayed significantly increased numbers of CD8⁺ T cells and reduced CD45⁺CD11b⁺F480⁺ macrophages (Fig. 1D and E). No differences in CD4⁺ T cells or immature myeloid cells (CD45⁺CD11b⁺Gr1^hi) were observed in the mixed tumors (Supplementary Fig. S3C and S3D).

In mixed tumor populations, we found increased expression of PD-L1 that tracked with increased proportion of Py117 cells (Fig. 1F). These findings suggest that specific subclones within a heterogeneous tumor can support a more robust immunogenic microenvironment leading to induced PD-L1 expression. A similar observation was made with MHC-I, where mixed tumor populations had higher MHC-I expression compared with tumors with 100% Py8119 cells. Moreover, as low as 20% of Py117 cells in the mixture was enough to induce a homogeneous increase in MHC-I (Fig. 1G, right), which can lead to enhanced antigen presentation in the microenvironment.

We hypothesized that the greater than expected reduction in tumor volume after radiotherapy in mixed tumors is due to an initial innate response triggered by Py117 tumor cells followed by an adaptive T-cell response. Although NK-cell depletion did not result in a reduced response in Py117 tumors (18), we did observe differences in CD11b myeloid populations, as highlighted in Fig. 1E. Therefore, we assessed the different M1-like macrophage subtypes (M1: CD45⁺CD11b⁺F4/80⁺MHCII⁺iNOS⁺), M2-like macrophages (M2: CD45⁺CD11b⁺F4/80⁺Arg1⁺), and dendritic cells (DCs: CD45⁺CD11b⁻CD11c⁺MHCII⁺). Ten days after treatment with 12 Gy radiotherapy, tumors were harvested, dissociated, and analyzed by flow cytometry. Regardless of radiotherapy, the proportion of infiltrating CD45⁺ leukocytes were greater in the Py117 tumors, and the number of macrophages as a proportion of CD45⁺ cells was higher in Py8119 tumors (Fig. 2A and B). Within the Py8119 tumors, we observed similar proportions of M1-like macrophage cells expressing MHC-II and iNOS that decreases after radiotherapy (Fig. 2C). In contrast, we observed significantly more Arg1⁺ M2-like macrophages in the Py8119 tumors regardless of radiotherapy (Fig. 2D). Thus, Py117 tumors possess a greater innate inflammatory M1-like response than a protumor M2-like response. In addition to the proinflammatory M1 environment, increased numbers of DCs were found in the Py117 tumors (Fig. 2E).

Because macrophage infiltration supports an immunosuppressive microenvironment that can impede adaptive immune responses by various mechanisms, we sought to determine how cytokine and chemokine profiles correlated with these differences. Py8119 tumors had significantly higher CSF1, CCL2, CCL5, and CCL7 levels that were further elevated after radiotherapy compared with immune-responsive Py117 tumors (Fig. 2F). We also observed that CCL3 another substrate for CCR5 and CXCL10 that is often elevated in cancer. In contrast, Py117 tumors express reduced levels of these factors, have greater antigen presentation, increased DC infiltration, and generate immune responses that can be enhanced with radiotherapy and immunotherapy. The cytokine and chemokine profile differences between these tumors is similar to the profile of individual cultured tumor cells, suggesting that most of the factors present in vivo are in part generated endogenously by the tumor cells (18).

A range of single fraction radiation doses of 8 Gy up to 30 Gy have been observed to induce adaptive immune responses which depend on immunogenicity, and the immune microenvironment among other factors (33–35). Tumors containing 50% Py8119 cells and 50% Py117 cells (Py5050) were treated with either 12 Gy or 20 Gy in a single fraction which resulted in no significant difference in tumor growth between the two radiotherapy doses (Fig. 3A). However, there was a trend toward greater infiltrating CD45, DCs, and CD8 T cells in the mixed Py5050 mixed tumors 10 days after 20 Gy radiotherapy compared with 12 Gy. Because all cell infiltrates in the Py5050 mixed tumors were greater than in Py8119 tumors alone after 20 Gy, we focused on 20 Gy for the subsequent experiments (Fig. 3B–D). In comparing different tumor populations, the abundance of myeloid cells was identified as a key difference 10 days after radiotherapy, we next evaluated the response 3 days after radiotherapy. We choose 3 days since previous studies indicate that mice can develop robust responses within 3 days after effective immunotherapy that are greatly dampened 8 to 10 days after treatment despite change in immune cell such as T-cell influx after an adaptive response (36). We found that levels of CD45⁺ cells and macrophages were significantly elevated in Py8119 and Py5050 tumors 3 days after treatment (Fig. 3E and F). However, M1-like cells were significantly reduced in Py8119 tumors but not in Py5050 tumors despite an increase in M2-like macrophages in both tumors (Fig. 3G and H). No differences in T cells or DCs were observed in either tumor type 3 days after treatment (Supplementary Fig. S4A and S4B). We then assessed adaptive responses in the lymph node 3 days after radiotherapy because there were greater numbers of infiltrating DCs and CD8⁺ T cells in mixed tumors 10 days after radiotherapy. We observed increased proportion of Ki67⁺ DCs and CD8⁺ T-cells in Py5050 lymph nodes after radiotherapy compared with Py8119 tumors. While the DCs were unchanged in the nodes, the relative proportions of Ki67⁺ CD8⁺ T cells were increased, suggesting a greater adaptive immune response by Py5050 tumors (Fig. 3I and J).

To better characterize the tumor microenvironment of mixed tumors, we evaluated changes in gene expression of cells and tumors individually (Fig. 4A). Whole transcriptome microarrays for both cultured Py117 and Py8119 tumor cells and Py117 and Py8119 tumors revealed 1079 and 1760 differentially expressed (DE) genes at adjusted P < 0.001. GSEA showed that Py8119 cells are enriched for cell proliferation and DNA repair, whereas Py117 cells are enriched for cell adhesion, endothelial cell development, and innate immune signaling. GSEA of the tumors further confirmed the tumor cell findings in Py8119 tumors and revealed enrichment of extracellular matrix components, antigen presentation, and adaptive immune responses in Py117 tumors, which are all consistent with our biologic findings (Fig. 4B). A full list of gene set enrichment analysis could be found in Supplementary Tables S1–S4.

PCA was performed from microarrays of Py8119, Py117, and Py5050 tumors to determine whether Py5050 global gene expression was more similar to Py8119 or Py117 tumors. PCA revealed that Py5050 tumors are more similar to Py8119 tumors (Fig. 4C). To understand why Py5050 tumors are more immunologically sensitive than Py8119 tumors and their transcriptomics cluster more closely, we filtered all tumor sets by DE and non-DE genes at the cellular level to identify immune sensitivity enhancers inherited from cell-intrinsic gene or by tumor microenvironment interactions, as depicted in Fig. 4A. This analysis revealed that by using DE cell-intrinsic genes, Py5050 tumors clustered with Py8119 tumors with few overlapping genes induced in Py5050 and Py117. However, by using the tumor microenvironment–induced genes or non-DE genes at the cellular level, or cells not differentially expressed between the cells in culture, Py5050 tumors clustered with the Py117 tumors and we found a greater number of overlapping genes between the Py5050 and Py117 tumors (Fig. 4D). We conclude that most of the immunologic sensitivity of Py5050 tumors result from tumor microenvironment induced genes controlled by intrinsic characteristics of the Py117 cells.

A gene ontology network map was generated using the ClueGo-CluePedia with significant GO terms (circles), cell-intrinsic genes (triangle-DE genes in cells and tumors) and tumor microenvironment genes (hexagon- non-DE genes at the cellular level induced in the tumor microenvironment). The green network represents the major cell-intrinsic gene significant GO terms represented in the microenvironment cluster, where the red network represents the induced tumor microenvironment gene significant GO terms map (Fig. 4E). The significant GO terms are shown in the pie charts to depict the relevant cellular processes associated with the intrinsic genes and then induced microenvironment genes. These data demonstrate cell migration, epithelial differentiation, cell adhesion, and junction organization are key intrinsic differences between tumor populations, and then the TME differences are associated with innate and adaptive immune responses consistent with our findings described above (Fig. 4F).

Multiple features in the Py117 and the Py5050 tumors contribute to the observed adaptive immune response. Py117 cells at baseline seem to have an adaptive immune response that is downstream of cell-intrinsic innate immune pathways and antigen presentation as suggested by our microarray data (Fig. 4). Based upon the differences in the immune microenvironment including increased M2-like macrophages, fewer M1-like macrophages, lower antigen presentation, fewer DCs, and minimal innate and adaptive transcripts in the Py8119 tumors, we hypothesized that targeting myeloid and antigen-presenting cells using a CD40 agonist antibody (anti-CD40) would generate an improved immune response. Anti-CD40 could address all the immunosuppressive factors observed from the Py8119 tumors by enhancing antigen presentation, activating M1 versus M2-like macrophages and mature dendritic cells in the tumor and draining nodes (37). We tested this hypothesis by treating Py8119 tumors with isotype control antibody, anti-CD40, 20 Gy radiotherapy, or a combination of these treatments. The combination of radiotherapy and anti-CD40 therapy revealed a more robust response than either treatment alone (Fig. 5A). We also examined whether targeting myeloid cells through a different mechanism by blocking CSF-1 would also result in an increased response. Targeting CSF-1 to decrease macrophage numbers also led to enhanced radiation response. However, the treatment was less effective than anti-CD40, so we focused on the anti-CD40 agonist (Fig. 5A).

Tumors were harvested 3 days after initiating each treatment alone or the combination of anti-CD40 + radiotherapy to investigate the impact on innate immune cells. Anti-CD40 and anti-CD40 + radiotherapy resulted in a greater immediate and cytotoxic response to that which was previously reported for anti-CD40 therapy (Fig. 5B; ref. 38). All treatments induced a significant influx of CD45⁺ leucocytes and macrophages (Fig. 5C). The macrophage change consisted of an increase in Arg1⁺ cells in all treatment groups, while the M1-like MHCII⁺iNOS⁺ phenotype specifically increased in the anti-CD40–containing groups (Fig. 5D). Tumors were then harvested 10 days after anti-CD40 and radiotherapy treatment and infiltrating leukocytes remained elevated in all treatment groups as a proportion of live cells, being 40% in anti-CD40 alone and radiotherapy alone groups, compared with 60% after anti-CD40 + radiotherapy (Fig. 5E). The relative proportion of macrophages remained high after treatment but significantly decreased in the anti-CD40 and combination treatment groups (Fig. 5F). The group treated with radiotherapy alone had the most unfavorable macrophage profile, showing the highest M2-like and lowest M1-like phenotypes. Even 10 days after starting treatment, the groups receiving anti-CD40 maintained a relatively low proportion Arg1⁺ M2-like and higher proportion of MHCII⁺iNOS⁺ M1-like macrophages (Fig. 5G).

In addition to the favorable macrophage response, we next evaluated the lymph node 3 days after therapy initiation and observed a greater proportion of DCs in the draining lymph node only in the anti-CD40 and radiotherapy combination (Fig. 5H). At this early time point, anti-CD40 treatment resulted in an increased proportion of CD8⁺ T cells that were expressing CD69, a marker for lymphoid activation, and was significantly greater in combination with radiotherapy (Fig. 5I). When we evaluated the DCs within the microenvironment, we found a greater number of DCs after CD40 administration alone at 3 days, and a similar level at 10 days (Fig. 5J). This result suggests that in these tumors the anti-CD40 agonist is effective in enhancing DCs in the TME, but that the combination is required for more efficient lymph node migratory capacity and activation of T cells.

Our data suggest that in resistant Py8119 tumors, anti-CD40 in combination with radiotherapy induces a rapid innate and early adaptive immune response 3 days after initiating treatment. Tumors harvested 10 days after anti-CD40 and radiotherapy treatment were evaluated for evidence of an adaptive immune response in the lymph node after 3 days of treatment. There was no significant increase in infiltrating T cells detected after 3 days, but at day 10, a 30-fold increase in infiltrating CD8⁺ T cells was detected with the combination treatment compared with control-treated mice. The number of CD4⁺ T cells only increased after radiotherapy treatment alone (Fig. 6A). Comparing CD40, radiotherapy, and the combination we observed greater MHCI levels on CD45-negative cells after combination treatment (Fig. 6B). In addition, CD45⁻ cells exhibited increased expression of PD-L1 after combination therapy, suggesting that IFN induction of PD-L1 was induced after the antitumor immune response (Fig. 6C).

Given the induction of PD-L1 and improved T-cell infiltration, we aimed to evaluate the addition of a PD-1–blocking antibody but observed a nonsignificant improvement over radiotherapy + anti-CD40 in the therapeutic response in mice (Fig. 6D). However, there was a greater response compared with combination therapy with and without radiation in nude mice supporting an adaptive component to the therapeutic response (Fig. 6E). Because fractionated radiotherapy at lower doses per fraction has been suggested to be more effective for generating adaptive responses, mice were treated with 10 Gy × 3 and 15 Gy × 2 with anti-CD40 and PD-1 antibody to determine whether cures could be obtained. This combination resulted in no obvious additive or synergistic effects when adding anti-CD40 and PD-1 in these tumors though the overall response was robust (Fig. 6F). Despite evidence of an adaptive immune response with triple combination therapy, complete regression was still not observed. Interestingly, PD-L1 expression was increased by anti-CD40 + radiotherapy treatment, but the addition of PD-1 antibody resulted in an increased population of PD-L1⁻ CD45⁻ cells that may support outgrowth of PD-L1–negative cells (Fig. 6G).

Complete responses were unable to be obtained in Py8119 tumors, so we next tested the triple combination in 4T1 mammary carcinoma tumors that are highly metastatic, more sensitive to radiation, and more immunogenic. Although these tumors were resistant to combination immunotherapy with anti-CD40 and PD-1 antibody inhibition, complete responses were observed when immunotherapy was combined with a single dose of 20 Gy radiotherapy. Mice in all treatment groups had to be euthanized due to the development of lung metastases, confirmed by bioluminescence imaging and/or at necropsy. However, in the combination therapy group, 40% of 4T1 tumor–bearing animals exhibited a complete response and no evidence of metastases occurred (Fig. 6H and I).

In this work, we describe that a radiotherapy and anti-CD40 combination therapy could generate robust adaptive immune responses in immunologically unresponsive tumors even if complete eradication was not accomplished. We next sought to determine whether the transcriptomic phenotypes found in murine tumors are representative of human tumors that are immunologically unresponsive and if CD40 could be a good target in this setting. In the PyMT tumors, GSEA revealed enrichment in IFN signaling and antigen presentation suggesting a greater adaptive immune response in the Py117 tumors. We found a similar pattern of enrichment differentiating patients with melanoma that were PD-1 responders and nonresponders after analyzing RNA-seq data from Hugo and colleagues using the STAR-DESeq2 pipeline (Fig. 6J; ref. 26). Because anti-CD40 can enhance adaptive immune responses and tumor-infiltrating leukocytes, we sought to evaluate whether CD40LG that is primarily presented on activated Th cells, CD40, and downstream MAP3K14 kinase expression correlated with immune cell infiltration into the TME (39). We plotted RNA expression of these three genes from breast cancer and melanoma TCGA datasets against relative immune cell infiltration as determined by the TIMER deconvolution algorithm showing a positive correlation with CD8, CD4, and DC populations in both tumor types (Supplementary Fig. S5A and S5B; ref. 29).

Given that key CD40 pathway genes correlate with immune cell infiltration in melanoma and breast cancer, we next investigated whether the CD40 pathway was enriched in patients with breast cancer with longer survival. We compared survival of patients from the TCGA that died within 3 years of diagnosis (66 patients) versus patients that lived at least 3 years (429 patients) and found significant enrichment of the CD40 pathway in the latter group confirming favorable association with outcome (Supplementary Fig. S6A). Because CD40LG is expressed on Th cells and is the only known ligand for CD40, we hypothesized its expression in the TME could predict survival. We observed a correlation with CD40LG with outcome in the OncoLnc pipeline using a multivariate Cox regression model (breast P = 0.015, melanoma P = 0.010, and pancreatic P = 0.42). Of the differentially expressed genes CD40LG was high in the rank for breast and melanoma, but not pancreatic adenocarcinoma (Supplementary Fig. S6B; ref. 30). We also investigated how expression levels of CD40LG related to survival in breast, melanoma, and pancreatic cancer in the TCGA datasets (Supplementary Fig. S6C). Representative Kaplan–Meier curves with cutoffs at the median CD40LG expression levels are shown for each cancer type and the number of patients at risk with log-rank P value over a broad range of cutoffs show that high expression level is positively correlated with improved survival except in pancreatic cancer (Supplementary Fig. S6D). Having found that CD40 pathway expression correlates with myeloid cell infiltration and CD40LG correlates with survival, we hypothesized that a high CD40LG/CD40 ratio correlates with immunotherapy response and that high ligand would favor an overall antigen presentation phenotype leading to improved adaptive immunity. We found that before therapy the CD40LG/CD40 ratio was not significantly higher in patients with melanoma who ultimately responded to CTLA-4 therapy or PD-1 therapy, but CTLA-4–unresponsive patients had a lower ratio before than after and a trend for higher ratio remained in patients that responded to PD-1 therapy (31). A significant difference in the ratio was observed for on-treatment biopsies and the difference in the ratio was even greater in patients progressing on PD-1 therapy (Supplementary Fig. S4E). This observation is consistent comparing pretreatment and on-treatment melanoma patients treated with PD-1 therapy from Riaz and colleagues as there was a higher CD40LG/CD40 ratio from on-treatment biopsies, not observed from pretreatment biopsies, in patients that obtained a CR, PR, or SD compared with progressive disease (Supplementary Fig. S6F; ref. 32). Although these data are only hypothesis generating, immune resistance and the role of CD40 in antigen-presenting cells could play an important role in enhancing immune responses in tumors not responsive to checkpoint immunotherapy.

Ionizing radiation can function as an immune stimulator because it can release tumor antigens, HMGB1, calreticulin, and activate the cGAS/STING pathway (40). However, evidence that radiotherapy triggers local and systemic immune responses in patients either alone or combined with immunotherapy is minimal except for dramatic responses in a limited number of patients (3, 5, 41–43). In this work, using a model of induced heterogeneity by mixing ratios of immunologically cold tumor cells with responsive cells, we show tumor-derived factors impact the immune response to radiotherapy and that anti-CD40, which targets myeloid cells, antigen presentation, and innate immune responses is sufficient to overcome limitations of Py8119 cold tumors. The Py8119 tumors have a resistant EMT phenotype, elevated expression of immunosuppressive AXL tyrosine kinase, heavy M2 macrophage infiltration, low levels of dendritic cells, and few infiltrating T cells (18). The immune stimulating activity of Py117 cells altered the TME, recapitulating many factors targeted by anti-CD40, thereby converting the unresponsive Py8119 tumors to an immune responsive tumor. The combination of CD40 agonist and radiotherapy resulted in an induction of DCs in the draining lymph node 3 days after treatment, as compared with either no response for radiotherapy. This further supports that a constellation of immunosuppressive and immune-activating factors can tip the balance toward a better antitumor immune response. Thus, targeting the myeloid compartment will depend greatly upon the specific TME of a given tumor.

There is increasing support for combining radiotherapy and anti-CD40 to treat immune unresponsive tumors. In our poorly immunogenic tumors, radiotherapy did not provide immune-stimulating activity, but targeting the myeloid compartment increased tumor response. These data are consistent with the results from the recently reported Tonic trial that aimed to detect enhanced immunotherapy responses in unirradiated tumors in triple-negative breast cancer that showed very little evidence of focal radiation improving systemic responses to nivolumab (44). We show that the immune-stimulating effect is robust in the draining lymph node days after treatment despite only modest shifts in the balance of the macrophage phenotype in the TME. Previous studies suggest that total body irradiation with anti-CD40 can lead to enhanced infiltration of adoptively transferred T cells into pancreatic tumors (45), that single-fraction radiotherapy with intratumoral anti-CD40 could lead to T-cell responses in unirradiated tumors (46), and that anti-CD40 in combination with radiotherapy and immune checkpoint therapy complement each other in resistant pancreatic cancer models (47). Our data complements prior work and proposes a clinical approach to combine anti-CD40 with radiotherapy for locally advanced disease to enhance direct antitumor responses that may provide systemic protection if sufficient immunogenicity is generated.

We hypothesized that combining radiotherapy with immunotherapy has the greatest potential to treat local disease at high risk for developing metastases, especially when undetectable micrometastatic disease likely exists at the time of treatment. To test this concept, we used the mouse 4T1 syngeneic breast cancer tumor model, which is highly metastatic. A significant portion of mice with primary 4T1 tumors treated with radiotherapy, anti-CD40, and PD-1 antibody never developed metastatic disease compared with control mice. This approach could have promise in breast cancer as high-dose stereotactic radiation is currently proving to be safe in partial breast irradiation and a single-fraction dose escalated neoadjuvant trial have completed and more neoadjuvant studies are being planned or have begun (NCT03366844, NCT0116220, NCT02685332 and NCT03872505; ref. 48). Testing combination therapies this setting would provide an opportunity to evaluate the efficacy of radiation and immunotherapy combinations to improve progression-free survival with a potential impact on local and distant disease. Two additional approaches for combination therapy are to use radiotherapy to consolidate minimal residual metastatic sites after initial systemic therapy in metastatic patients or as adjuvant treatment for locally advanced definitive therapy, as reported in the Pacific trial using durvalumab, a PD-L1 Ab after chemoradiation in stage III lung cancer (1, 49, 50).

T.A. Aguilera reports receiving other commercial research support from Galera Therapeutics (sponsored trial), Apexigen Inc (trial funding), and iTeos Therapeutics (sponsored research), and has ownership interest (including patents) in a Stanford patent and UC San Diego patent. A. Koong has ownership interest (including patents) in Aravive, Inc. Q.-T. Le is a consultant at Grail, and is a consultant/advisory board member at Merck. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T.A. Aguilera, H. Shehade, M. Rafat, M. Kariolis, A. Koong, E.E. Graves, A.J. Giaccia

Development of methodology: T.A. Aguilera, M. Rafat, L. Castellini, M. Kariolis, L.G. Ellies, A.J. Giaccia

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T.A. Aguilera, M. Rafat, L. Castellini, M. Kariolis, L.G. Ellies, E.B. Rankin

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T.A. Aguilera, E.A. Elghonaimy, D. Jiang, M. Kariolis, A. Koong, Q.-T. Le, E.E. Graves, A.J. Giaccia

Writing, review, and/or revision of the manuscript: T.A. Aguilera, M. Rafat, A. Koong, Q.-T. Le, L.G. Ellies, E.E. Graves, A.J. Giaccia

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T.A. Aguilera, H. Shehade

Study supervision: T.A. Aguilera, E.E. Graves

We thank the Stanford Shared FACS Facility, the Stanford Human Immune Monitoring Core, the Stanford Protein and Nucleic Acid Facility, and the UT Southwestern Bioinformatics Core Facility and the fellowship program, Min Kim, and Venkat Malladi for use of their facilities and services. We thank Sharon Clarke and Damiana Chiavolini for assistance with editing the manuscript. This work was supported by the ASTRO Resident Seed Grant# RA2014-3 (to T.A. Aguilera); CPRIT first-time tenure track recruitment award #RR170051 (to T.A. Aguilera/E.E. Graves), Ruth L. Kirschstein National Research Service Award PA-14-015 grant# T32 CA 121940 (to T.A. Aguilera/M. Rafat); Katherine McCormick Advanced Postdoctoral Fellowship and NIH grant# K99CA201304 (to M. Rafat); NIH grants CA-67166 and CA-197713, the Silicon Valley Foundation, the Sydney Frank Foundation and the Kimmelman Fund (to A.J. Giaccia); NIH grant K08CA88035 (to L.G. Ellies); and the Blue Dot Fund (to A. Koong).

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.