PD-1 Blockade Following Isolated Limb Perfusion with Vaccinia Virus Prevents Local and Distant Relapse of Soft-tissue Sarcoma

Soft-tissue sarcomas are a group of rare, heterogeneous tumors derived from mesenchymal tissues, with an annual incidence of approximately 5/100,000 persons (1). The most commonly affected sites are the extremities, accounting for approximately 40% of all cases (2). Limb-conserving surgery and adjuvant radiotherapy is highly effective in securing long-term control of the primary tumor (3). Although such treatment secures local control in the limb in more than 85%–90% of patients, rates of systemic relapse are high, approaching 30% at 5 years for all patients with sarcoma (4, 5). The evidence base that adjuvant chemotherapy can lessen the risk of metastatic sarcoma is not strong (6, 7). Once metastatic disease is established, systemic cytotoxic chemotherapies are relatively ineffective, with patients having a median survival of only 12 months (8). Therefore, the development of new strategies for the prevention and/or treatment of metastatic sarcoma is an area of great clinical importance.

Immunotherapies, in particular immune checkpoint inhibitors, have markedly improved the prognosis of patients with historically poor-prognosis cancers, such as metastatic melanoma (9,–11). However, evidence from early clinical studies suggests that these agents have limited efficacy in either soft-tissue or bone sarcomas (12, 13). Determinants of response to immune checkpoint blockade in other pathologies have been defined, with such factors as tumor mutational burden, CD8⁺ T-cell infiltration, upregulation of IFNγ-related genes and tumor cell expression of immune checkpoint ligands found to be associated with greater activity (14,–18). By combining immune checkpoint inhibitors with other immune-activating cancer therapies, it is thought that it may be possible to cause tumors to display some of these characteristics, increasing their sensitivity to immune checkpoint blockade. Oncolytic viruses represent attractive candidates for such combinations, because they are increasingly recognized as immunotherapies in their own right and have highly favorable toxicity profiles (19).

To date, the only route of viral delivery shown to be clinically effective is intratumoral injection, with the modified herpes simplex type I virus talimogene laherparepvec (Imlygic, Amgen) now licensed for patients with stage IIIB–IV melanoma (20). However, the optimal route for viral delivery to soft-tissue sarcoma remains to be defined. Intratumoral delivery may be less efficacious than in melanoma due to the size and depth of these tumors. Systemic viral delivery would be ideal, allowing multiple tumor sites to be treated simultaneously, but this approach is yet to demonstrate clinical efficacy. This may be due to eradication or sequestration of circulating virus by antiviral defenses, preventing the delivery of sufficiently high viral titers to the tumor. Nevertheless, intravenous delivery of several strains of virus has demonstrated tropism for tumors, with mounting evidence that virus delivered in this fashion can promote a more inflamed tumor microenvironment, which may render the tumor more susceptible to subsequent immune checkpoint blockade (21 –24).

We have previously shown that locoregional delivery of virus by isolated limb perfusion (ILP) to be an effective alternative to systemic administration (25, 26). ILP is the locoregional delivery of chemotherapy, usually melphalan and TNFα, to a tumor-bearing limb that has been surgically isolated from the systemic circulation. Not only does ILP allow the delivery of melphalan at doses many times greater than could be tolerated systemically, it also allows the safe delivery of intravascular TNFα, whose vasoactive effects enhance the permeability of the tumor's microvasculature, thereby increasing the penetrance of melphalan within the tumor (27). ILP has a well-established role in the management of locally advanced extremity sarcomas, either as a stand-alone treatment to avoid amputation or a neoadjuvant therapy to facilitate function-preserving resections (28, 29).

Locoregional delivery of virus by ILP has theoretical benefits over intratumoral injection or systemic delivery. As well as targeting the delivery of high viral titers to the tumor-bearing limb, isolation of virus within the limb circulation prevents first-pass metabolism in the liver and viral sequestration within the reticuloendothelial system. Furthermore, the vasoactive effects of TNFα may increase viral penetrance of the tumor. Previous work by our team, using an animal model of extremity sarcoma and a modified vaccinia virus (GLV-1h68), found that the addition of GLV-1h68 to standard-of-care ILP delayed tumor growth and prolonged survival (25). However, when this combination was used neoadjuvantly prior to surgery and radiotherapy, viral ILP was insufficient to prevent the development of metastatic disease even though local control could be achieved (26).

Here we demonstrate that the delivery of GLV-1h68 delivered by ILP substantially improves response to subsequent PD-1 blockade in a rodent model of high-grade extremity sarcoma. The combination of a viral ILP with anti-PD-1 antibodies led to dramatic alterations in the sarcoma tumor microenvironment (TME), with the expansion of both effector T cells and dendritic cells. When combined with surgical resection and radiotherapy, this combination immunotherapy achieved durable cure of both local and distant disease in a previously treatment-refractory model.

Using an immune-competent, orthotopic, and metastatic animal model of ESTS, we characterized the efficacy of anti-PD-1 antibody monotherapy. We assessed the therapeutic benefit of combining anti-PD-1 blockade with the locoregional delivery by ILP of a modified vaccinia virus alongside standard biochemotherapy. The ability of this combination therapy to control local and systemic disease relapse was evaluated in both palliative and neoadjuvant protocols. To elucidate the mechanism responsible for enhanced therapy, we characterized effects of treatment on the tumor microenvironment in our in vivo model using flow cytometry, IHC, and RNA sequencing (RNAseq). We also assessed the production of recognized markers of immunogenic cell death in rodent and human sarcoma cell lines in vitro. A single investigator performed all the surgical procedures and adjuvant irradiations. Sample sizes were determined using a power calculation based on detecting a 20% difference between control and treated cohorts.

BN175, HT1080, and SW872 cell lines were passaged in DMEM, supplemented with 5% heat-inactivated FBS, 2.5% l-glutamine, and 1% penicillin/streptomycin and maintained at 37°C and 10% CO₂.

GLV-1h68 was produced and provided by Genelux Corporation. GLV-1h68 is an attenuated vaccinia virus and was constructed as described previously (30). Melphalan (Alkeran) was purchased from Laboratoires Genopharm. Recombinant human TNFα was purchased from First Link Ltd. Anti-PD-1 antibody (clone J43) and the relevant isotype (Armenian Hamster IgG) were purchased from 2BScientific.

Recombinant rat PD-1 protein was purchased from Sino Biological. After reconstitution, the recombinant protein was conjugated to flow cytometry beads using the Functional Bead Conjugation kit from Becton Dickinson. Samples were stained with the J43 anti-PD-1 antibody and the relevant isotype control with an Alexa Fluor 488 fluorescent secondary antibody (Thermo Fisher Scientific) and assessed using flow cytometry.

Inbred, male Brown Norway rats weighing between 225 and 275 g were obtained from Envigo. Tumors were established by injecting 1 × 10⁷ BN175 cells subcutaneously into the left biceps femoris muscle. For anti-PD-1 monotherapy, treatment was started 6 days after tumor implantation, with 3 doses of 200 μg of anti-PD-1 or isotype antibody delivered by intraperitoneal injection at 48-hour intervals. ILP was performed as described previously and combined with anti-PD-1 antibodies/isotype controls in palliative (Fig. 2A) and neoadjuvant (Fig. 3A) protocols (25). An Institutional Animal Care and Use Committee (IACUC) approved the in vivo experiments.

For ATP release, cells were plated at 5 × 10⁴ cells/well, incubated overnight, and treated with the indicated therapeutics. At the relevant time points, 200 μL of medium was collected and centrifuged to remove cellular debris. Fifty microliters of CellTiter Glo (Promega) was then added and luminescence immediately quantified with a SpectraMax 384 plate reader (Molecular Devices). For HMGB1 secretion, 1 × 10⁶ cells/well were incubated overnight and treated with the indicated therapeutics. Culture supernatants were collected after 72 hours and secreted HMGB1 quantified by ELISA (IBL International). For cell surface calreticulin (CRT) expression, cells were plated at 3 × 10⁵ cells/well, incubated overnight, and treated with the indicated therapeutics. After 24 hours, cells were collected and stained with a primary anti-CRT antibody (Thermo Fisher Scientific) and Fixable viability dye-eFluor 780 (Thermo Fisher Scientific). CRT expression was then analyzed with flow cytometry, with dead cells excluded from analysis.

For in vitro analysis of PD-L1 expression, cells were stained with a PD-L1 primary antibody (ProteinTech) and an Alexa Fluor 488 secondary (Thermo Fisher Scientific). For in vivo analysis of tumors and tumor-draining lymph nodes, tissues were harvested from tumor-bearing Brown Norway rats. Both tissues were stained with two separate panels. Panel 1 comprised CD3-APC, CD4-PE, Granzyme B-Pacific Blue (BioLegend), CD8-PE Cy7, Foxp3-Alexa Fluor 700, and Fixable viability dye-eFluor 780 (Thermo Fisher Scientific). Panel 2 comprised CD3-APC, CD161-PE, Granzyme B-Pacific Blue (BioLegend) and Fixable viability dye-eFluor 780 (Thermo Fisher Scientific).

Analyses were performed on formalin-fixed paraffin-embedded (FFPE) sections. Primary antibodies specific for rat PD-L1 were sourced from ProteinTech and for rat CD3 and CD8 were sourced from Thermo Fisher Scientific. Digital images of the stained slides were obtained using the Nanozoomer-XR platform (Hamamatsu Photonics). For the comparison of immune infiltrates in tumors at the humane endpoint, 10 fields of view at 20× magnification were randomly selected from 6 samples in each cohort. For the analysis of CD8 topography, 5 fields of view at 20× magnification were randomly selected from both the invasive margin and the tumor parenchyma, respectively, from 6 samples in each cohort. Staining was quantified using Cell Profiler software (31).

Fresh frozen samples were collected from 3 animals in each treatment cohort 12 days after tumor engraftment. Tumor samples were snap frozen in RNAlater (Thermo Fisher Scientific) and stored at −80°C prior to RNA extraction. Samples were processed by homogenization using a Precellys24 homogenizer (Bertin Technologies) and RNA was extracted using an RNeasy Mini Kit (Qiagen Ltd). Genomic DNA was removed from 400 ng of total RNA samples using genomic DNA eliminator column from RNeasy Plus Micro Kit (Qiagen Ltd), polyA RNA was selected using the NEBNext mRNA magnetic Isolation Module (New England Biolabs) following manufacturer's instructions. From the resulting mRNA, strand-specific libraries were created using NEBNext Ultra II Directional RNA Library Prep Kit for Illumina (New England Biolabs). Final libraries were quantified using qPCR and clustered at a molarity of 13.5 pmol/L; sequencing was performed on an Illumina HiSeq 2500 using SE x50 cycles RAPID v2 SBS chemistry. The 18 samples were run at 9 samples per lane to achieve coverage of 30–40 million reads per sample. For the bioinformatics, raw count data were normalized to correct for library size and RNA composition bias. Each therapeutic cohort was compared with the untreated controls, which were used as a base to calculate differential gene expression. Read data were filtered prior to analysis by removing (i) any genes with 0 reads across all samples and (ii) noncoding protein genes. Differential expression analysis was performed in R using the Bioconductor package DESeq2 (version 1.18.1). Pathway analyses were performed using the WebGestaltR package (version 0.1.1).

All statistical analyses, with the exception of RNAseq data, were conducted using Graphpad Prism, version 7.0. Grouped data are presented as means ± SEM. Differences between multiple groups were assessed using one- or two-way ANOVA with post hoc analysis using Bonferroni correction. Differences between two groups were assessed using a two-tailed unpaired t test. Survival outcomes were compared using the Kaplan–Meier method and the log-rank test.

Materials can be obtained from The Institute of Cancer Research (ICR) by a material transfer agreement.

To determine the potential efficacy of PD-1 blockade in BN175 sarcomas, the basal immune landscape of these tumors was characterized. On IHC, BN175 sarcomas contained CD3⁺ tumor-infiltrating lymphocytes (TIL) and expressed PD-L1, the ligand of PD-1 (Fig. 1A–C). Expression of PD-L1 was inversely related to the density of TILs, with the majority of TILs located in the tumor periphery where PD-L1 expression was absent (Fig. 1D–F). Flow cytometry was used to further characterize the composition of TILs in BN175 sarcomas, revealing both effector and regulatory T-cell populations (Fig. 1G). BN175 cells were also found to express PD-L1 at baseline in vitro, with no further upregulation noted following treatment with IFNγ suggestive of cell-autonomous induction of PD-L1 (Supplementary Fig. S1A). As no rat-specific anti-PD-1 antibodies are commercially available, the cross-reactivity of mouse-reactive anti-PD-1 antibody (clone J43) was tested. Recombinant rat PD-1 protein was conjugated to flow cytometry beads and stained with the candidate therapeutic antibody or isotype control. Fluorescence was only noted with the therapeutic antibody, indicating that the J43 mouse-reactive anti-PD-1 antibody cross-reacts with rat PD-1 (Supplementary Fig. S1B). This antibody was then used to determine therapeutic efficacy of PD-1 blockade in vivo. PD-1 blockade exerted limited efficacy as monotherapy, causing a significant delay in tumor growth but only a modest improvement in survival (median 10 vs. 13 days, P = 0.006; Fig. 1H and I). Dose escalation yielded no further therapeutic benefit, implying that the characteristics of the TME rather than the dosage of anti-PD-1 antibody limited efficacy in this model (Supplementary Fig. S1C and S1D).

Having demonstrated that anti-PD-1 monotherapy has limited efficacy, we investigated whether pretreatment with our established regimen of viral ILP (GLV-1h68, melphalan, TNFα) could increase this effect (25, 26). Tumor-bearing Brown Norway rats underwent a treatment protocol that mimicked the clinical use of ILP as palliative treatment (Fig. 2A). Combination therapy with viral ILP and PD-1 blockade enhanced therapy compared with either modality alone (Fig. 2B). However, recurrent disease rapidly developed within the limb (Fig. 2C) and, as such, survival was not improved compared with viral ILP alone (Fig. 2D). In some cases, local recurrence occurred while PD-1 therapy was ongoing, suggesting that compensatory intratumoral immunosuppressive mechanisms may be responsible. Furthermore, tumor immune infiltration by both CD3⁺ and CD8⁺ T cells was significantly reduced once the humane endpoint was reached following treatment with viral ILP ± PD-1 blockade compared with untreated controls, suggesting that treatment resistance may involve enhanced immune evasion (Fig. 2E–G). The metastatic potential of BN175 sarcoma has previously been described, with lung metastases developing in animals surviving beyond 30 days postimplantation (26). Lung metastases were noted in all animals that survived beyond this point, always in association with locally recurrent disease, indicating the ability of resistant tumors to evade systemic immune surveillance (Fig. 2H and I).

Although viral ILP and PD-1 blockade improved initial therapy, control of local and distant disease was not achieved in this aggressive model. However, we saw that the enhanced initial response to therapy offered a window of opportunity for further intervention. ILP can be used clinically as a neoadjuvant preoperative treatment. We hypothesized that surgical resection of regressing tumors after viral ILP could prevent the evolution of resistant disease and secure durable control of local and distant disease. To investigate this hypothesis, a modified protocol of neoadjuvant viral ILP was developed (Fig. 3A). Following ILP, a compartmentectomy was performed to remove the tumor en-bloc within the biceps femoris muscle, with the aim of achieving negative circumferential surgical margins (Fig. 3B). At the time of resection, tumor volumes were significantly smaller in rats treated with viral ILP combined with PD-1 blockade (Fig. 3C). Whereas local recurrence occurred in two-thirds of animals treated with viral ILP alone, no local relapses were noted in the combination therapy cohort (Fig. 3D and E). The combination of viral ILP with PD-1 blockade significantly improved survival compared with viral ILP alone (P = 0.018; Fig. 3F). Metastatic disease developed in a third of animals treated with viral ILP alone, with one animal euthanized for symptomatic pulmonary lesions, whereas no evidence of metastatic disease was found in the combination therapy cohort (Fig. 3G). As such, all animals in the combination cohort were cured of local and distant disease.

Having demonstrated that viral ILP and PD-1 blockade was able to secure local and systemic disease control, we sought to determine the importance of GLV-1h68 in increasing sensitivity to PD-1 inhibition. In order to do so, tumour-bearing Brown Norway rats underwent the same neoadjuvant ILP protocol as described above, but without the addition of GLV-1h68. At the time of tumour resection, no difference in tumour volume was noted with the addition of PD-1 blockade to standard ILP, although a marked reduction in tumour volume was noted in 2 out of 9 subjects receiving the PD-1 inhibitor (Supplementary Fig. S3A). Animals receiving standard ILP and anti-PD-1 antibodies had a lower incidence of local recurrence compared with standard ILP alone (Supplementary Fig. S3B and S3C). However, symptomatic metastases developed in 2 of these animals. As a result, the addition of anti-PD-1 antibodies to standard ILP did not improve survival when compared with standard ILP alone (Supplementary Fig. S3D).

We then sought to determine the possible mechanism by which melphalan and GLV-1h68 improve the efficacy of PD-1 blockade. The manner of cell death is known to influence the generation of antitumor immune responses. Immunogenic cell death is characterized by the production of several damage-associated molecular patterns (DAMP), with release of ATP, cell surface calreticulin (CRT), and secreted HMGB1 thought to be the most vital to establishing antitumor immune responses (32).

We have previously shown that melphalan and GLV-1h68 are cytotoxic in both rat and human sarcoma cell lines, with at least an additive effect with combined therapy in vitro (25). Although treatment with melphalan induced marked secretion of ATP from BN175 cells, only modest secretion was noted in one human sarcoma cell line (Supplementary Fig. S4A–S4C). In contrast, GLV-1h68 induced substantial secretion of ATP in all lines tested (Supplementary Fig. S4D–S4F). Similarly, melphalan significantly increased cell surface CRT expression in the BN175 and SW872 cell lines, but not in HT1080 cells (Supplementary Fig. S4G–S4I). However, GLV-1h68 led to increase CRT expression in all cell lines, to consistently higher levels than noted with melphalan (Supplementary Fig. S4J–S4L). Significant increases in secreted HMGB1 were seen in all cell lines treated with either melphalan or GLV-1h68, with the highest levels noted after viral infection (Supplementary Fig. S4M–S4O).

To determine the effects of therapy on the TME in vivo, tumors and tumor-draining lymph nodes were collected from 3 animals in each cohort at the time of surgery and analyzed by flow cytometry (representative gating strategies in Supplementary Fig. S5). Combination treatment with viral ILP and PD-1 blockade significantly increased the number of tumor-infiltrating CD3⁺ T cells (Fig. 4A). This comprised a significant increase in both CD8⁺ cytotoxic T cells and CD4⁺ Foxp3⁻ effector T cells (Fig. 4B and C). No significant change in CD4⁺ Foxp3⁺ regulatory T cells was noted (Supplementary Fig. S6A), although a trend toward an increase was seen with all ILP-based therapies. Viral ILP, with or without PD-1 blockade, also significantly increased the proportion of CD8⁺ T cells expressing the activation marker Granzyme B compared with untreated controls (Fig. 4D). No significant difference in tumor infiltration by CD3⁻ CD161⁺ NK cells was noted (Supplementary Fig. S6B), although again a trend toward an increase was seen in the viral ILP cohorts. Both viral ILP alone and its combination with PD-1 blockade caused an increase in overall immune infiltration of the tumor when compared with other therapies (Supplementary Fig. S6C). Although no significant differences were noted within individual immune subsets, viral ILP and PD-1 blockade also increased immune infiltration within tumor-draining lymph nodes, harvested from above the level of the tourniquet and, therefore, outside of the perfusion field (Supplementary Fig. S6D). As such, viral ILP with PD-1 blockade not only resulted in increased immune infiltration of the tumor but also promoted immune infiltration in lymph nodes located outside of the perfusion field, indicative of at least a locoregional and, perhaps, a systemic immune-priming effect.

Having found that viral ILP and PD-1 blockade most notably increased infiltration by CD8⁺ cytotoxic T cells, we sought to determine whether treatment altered the topography of these cells within BN175 sarcomas. Using tumors collected on day 12 following treatment with viral ILP ± anti-PD-1 antibodies, the density of CD8⁺ cytotoxic T cells at both the invasive margin and within the tumor parenchyma was quantified. In untreated tumors, although CD8⁺ cytotoxic T cells were present, the majority were limited to the invasive margin with very few found within the tumor parenchyma equating to an immune-excluded phenotype (33). The density of CD8⁺ cytotoxic T cells at both the invasion margin (Fig. 4E; Supplementary Fig. S7A–S7F) and within the parenchyma (Fig. 4F; Supplementary Fig. S7G–S7L) was significantly increased with all ILP-based treatments, with the most marked increase seen with viral ILP and PD-1 blockade. Indeed, viral ILP and PD-1 blockade completely reversed the topography of CD8⁺ cytotoxic T cells infiltration, significantly increased infiltration of the tumor parenchyma compared with the invasive margin (Fig. 4G–I).

Although evidence of increased effector T-cell infiltration was noted in all ILP-based therapies, only the combination of viral ILP and PD-1 blockade resulted in long-term disease control at local and distant sites. Cognizant of the potential contribution of other intratumoral immune populations and various chemo/cytokines to cure, we sought to further interrogate alterations in the tumor microenvironment using RNAseq, utilizing RNA extracted from the same tumor samples analyzed by flow cytometry.

Pathway analyses demonstrated evidence of the immune-priming effects of viral ILP. The top 10 upregulated pathways following viral ILP included antigen processing and presentation (rno04612), Th1 and Th2 cell differentiation (rno04658), and Th17 cell differentiation (rno04569). Furthermore, all of the remaining top upregulated pathways were immune-based being involved in infectious or autoimmune responses (Supplementary Table S1). In contrast, none of these pathways was among the top 10 upregulated following ILP alone.

The relative proportions of intratumoral immune cell populations were calculated using CIBERSORT (Fig. 4J; ref. 34). The most prominent immune subset present at baseline was macrophages, expressing a predominantly M2-like phenotype. Treatment with ILP alone promoted a more immunosuppressive environment, with an expansion in the proportion of regulatory T cells and a reduction in CD8⁺ T cells. The addition of anti-PD-1 to ILP lead to an expansion of CD8⁺ T cells without reducing the regulatory T-cell population, a potential reason for the lack of benefit noted with this combination therapy. In contrast, viral ILP promoted a more favorable TME, with a dramatic reduction in the proportion of regulatory T cells with an increase in CD8⁺ T cells when compared with ILP alone. However, the most marked difference between therapies was noted with viral ILP and PD-1 blockade, which in addition to increasing the proportions of effector T cells also led to a substantial reduction in M2-like macrophages and the accumulation of intratumoral dendritic cells (DC). CIBERSORT also allows an estimation of the relative abundance of immune populations (Supplementary Fig. S8A). Viral ILP and PD-1 blockade led to an almost 3-fold increase in total infiltration when compared with controls, demonstrating increases in absolute terms of both CD8⁺ T cells and DCs.

Having demonstrated such an increase in DCs with our curative regime, we sought to determine the activation status of these cells. Significant upregulation of a number of genes known to be associated with an activated state in DCs was noted following viral ILP and PD-1 blockade, in stark contrast to the other cohorts (Supplementary Fig. S8B; ref. 35). This is suggestive that not only did this combination therapy increase the number of intratumoral DCs but also augmented their function.

Alterations in the expression of various chemo/cytokines were noted in all therapeutic cohorts (Supplementary Fig. S8C–S8F). However, the majority of alterations were seen following treatment with viral ILP and PD-1 blockade. Furthermore, the expression of a number of chemo/cytokines was exclusively altered in this cohort. Within the CCL family, upregulation of the CCR1 receptor was only noted in this cohort, alongside increased expression of its ligands CCL3 and CCL5. Other exclusive alterations in the CCL family included upregulation of CCL4 and downregulation CCL27. Within the CXCL family, the only exclusive alteration with viral ILP and PD-1 blockade was the upregulation of CXCL2, with increased expression of its receptor CXCR2 also noted. Multiple exclusive alterations in expression were noted in the interleukins with upregulation of IL6, IL15, and IL17b and downregulation of IL24 with concordant alterations in the expression of their relevant receptors (IL6r, IL15ra, IL17ra/rd, IL20rb).

Further evidence of a favorable modulation of the TME following VV-ILP and PD-1 blockade was noted in the expression of genes recognized be adversely or favorably prognostic across a range of cancers (36). VV-ILP and PD-1 blockade led to significant downregulation of all of the top 10 adversely prognostic pan-cancer genes, in contrast to all other therapeutic cohorts (Supplementary Fig. S8G). In addition, this treatment led to the upregulation of the majority of the top 10 favorably prognostic pan-cancer genes (Itm2b, Cbx7, Cd2, Satb1, Saraf, Fuca1; Supplementary Fig. S8H).

There is a great need for novel therapies that can both prevent the development of and effectively treat metastatic sarcoma. The successes of immunotherapy in nonsarcomatous pathologies have not yet been replicated in sarcoma, which is generally viewed as a nonimmunogenic malignancy. Hence there is a great interest in identifying novel therapeutic combinations that may augment antitumor immune responses in sarcoma. Here we demonstrate the ability of an oncolytic virus delivered by ILP to dramatically alter the TME in soft-tissue sarcoma. In doing so, viral ILP markedly increased the efficacy of PD-1 inhibition leading to durable cures.

The potential for various viruses to synergize with immune checkpoint inhibitors in nonsarcomatous pathologies has now been demonstrated in both preclinical and clinical studies (37,–40). The optimal route of viral delivery remains the topic of much debate. In all except one of the studies above, virus was delivered by intratumoral injection (37, 38, 40). While this is an effective strategy in both mice and men with subcutaneous disease, it may be a less practical approach in sarcoma due to the size and depth of these lesions. Systemic delivery would be preferable, but is yet to demonstrate clinical benefit in solid tumors (21, 23). Locoregional delivery of virus via ILP allows many of the barriers met with systemic administration to be overcome and we have previously shown this route of delivery to be more effective than systemic delivery (25, 26).

The combination of viral ILP with PD-1 blockade dramatically altered the TME in BN175 sarcoma. Not only did this treatment increase the number of CD4⁺ and CD8⁺ effectors cells within the tumor, but it also increased their expression of activation markers (Fig. 4A–D). Increased infiltration by effector cells was not accompanied by expansion of intratumoral regulatory T cells tipping the balance in favor of antitumor immune responses. Furthermore, this treatment radically modified the topography of immune infiltration within the tumor. Untreated BN175 sarcoma demonstrated an immune-excluded phenotype, with CD8⁺ infiltration limited to the invasive margin and sparse infiltration of the tumor parenchyma. Although infiltration at both sites increased following viral ILP and PD-1 blockade, the preferential expansion of parenchymal CD8⁺ cells led to a complete reversal of the topography of infiltration (Fig. 4E–I).

Location rather than presence of TILs is thought to be increasingly important in both response and resistance to immune checkpoint blockade. While the presence of CD8⁺ TILs at the invasive margin has been found to be predictive of response to PD-1 blockade in patients with melanoma, responders were characterized by expansion of these cells within the tumor (18). Again in melanoma, patients who relapsed after PD-1 therapy were found to have fewer CD8⁺ TILs that were limited to the margin of the tumor when compared with their original lesion (41). Similar features were seen with relapse following viral ILP and PD-1 blockade in our model (Fig. 2E–G). Our data attach a similar importance to the location of TILs in this model of sarcoma, demonstrating that oncolytic viruses can induce favorable topographical changes in the sarcoma TME.

In trying to predict how responses to immune checkpoint blockade may be maximized, and how oncolytic viruses may facilitate such augmentation, much of the focus has been on T cells, in particular cytotoxic CD8⁺ T cells. Although significant alterations in both CD8⁺ T-cell infiltration and distribution within the tumor were noted in our curative regimen, similar changes, albeit to a lesser extent, were noted in other therapeutic cohorts that were compromised by both local and distant relapse. This suggests that the engagement of other immune cell populations is critical to engendering antitumor immunity. Our data suggest that the critical population in this process is DCs, with a marked expansion of this population only noted with our curative regimen. By processing and presenting antigen, DCs play a vital role in activating the adaptive immune system and so it follows that the engagement of this cell population is critical in generating systemic antitumor immunity following a local therapy (42). DCs are known to express both PD-L1 and PD-L2 and blockade of the PD-1 axis has been shown to promote the ability of these cells to stimulate effector T-cell functions (43). Hence, while viral ILP may promote a more inflamed TME characterized by increased infiltration by T-cell effectors, its efficacy may be limited by tumor mechanisms of immunosuppression. The additional blockade of PD-1 may then disinhibit DCs to stimulate a more effective, durable antitumor immune response. Given the complexity of the curative regimen utilized in this model, it is possible that the other therapeutics may also be implicated in this process, most notably TNFα, which has been shown to enhance DC maturation (44). However, given that no substantial alteration in DCs was noted following standard ILP with or without subsequent PD-1 blockade, it does not appear that TNFα is critical to the expansion of this cell population in this model.

Alterations in the chemokine profile of BN175 sarcoma support the importance of DCs to cure. CCL4 was exclusively upregulated in the curative regimen and has been previously shown to play a critical role in the recruitment of DCs in both clinical and preclinical models of melanoma (45). A lack of intratumoral DCs was associated with limited response to checkpoint blockade, a phenotype that was reversed if these cells were then delivered by intratumoral injection. CCL3 and CCL5 have also been shown to play a role in both T cell and DC recruitment in models of melanoma and colorectal cancer (46, 47). While these chemokines were also upregulated in other therapeutic cohorts in our model, their receptor CCR1 was only upregulated in the curative regimen, suggesting greater activity of this axis. Furthermore, only viral ILP and PD-1 blockade led to increased expression of IL6, which has been shown to be critical for the effective induction of Th17 cells by DCs (48).

Our data also highlight the importance of combining immunotherapy with effective local therapies to maximize its curative potential. The evolution of resistant disease is well recognized in patients receiving immune checkpoint inhibitors for melanoma, although the responsible mechanisms are complex and still under investigation (41, 49). Such lesions are typically unresponsive to further therapy, portending poor outcomes. In our palliative model, in which the tumor remains in situ, an initial improvement in therapy was seen when combining viral ILP with PD-1 blockade, but resistant disease rapidly developed leading to both local and distant treatment failure (Fig. 2). In contrast, effective local treatment with early surgical resection and adjuvant radiotherapy of regressing tumors led to durable cures (Fig. 3). The combination of surgery, GLV-1h68, and PD-1 blockade were crucial to cure, as the removal of any part of this protocol was associated with relapse. Similar results have been demonstrated with the neoadjuvant administration of Maraba virus prior to surgery in a model of triple-negative breast cancer (50). It may be that such neoadjuvant strategies allow the generation of an effective antitumor immune response, which eliminated micrometastatic disease, while simultaneously preventing the evolution of resistance by effectively treating the primary tumor.

The authors recognize certain limitations of this study. As the in vivo experiments were conducted in a single tumor model, the results on the TME at baseline and its transformation following therapy should be interpreted with caution in the context of the diverse histologic subtypes seen in human soft-tissue sarcoma. We have previously attempted to classify the BN175 tumor with standard IHC stains (S100, desmin, smooth muscle actin, caldesmon, and Factor VIII). The only positive marker was Factor VIII, typically associated with angiosarcoma. However, morphologically, this tumor is not typical of a liposarcoma or angiosarcoma and was felt to be more in keeping with an undifferentiated pleomorphic sarcoma. Several studies have sought to delineate the TME in soft-tissue sarcoma, although they are limited by both diversity in the histologic subtypes included and the methods employed. In translational analyses of baseline samples from patients enrolled in a phase II study combining pembrolizumab with cyclophosphamide, CD8 T-cell infiltration was low, as seen in the BN175 tumors (13). However, PD-L1 expression was also low, positive in only 6 of 50 patients, and colocalization rather than exclusion of CD8 T cells was noted. In contrast, while studies of historical pathologic samples have demonstrated similar results in terms of CD8/PD-L1 colocalization, markedly higher levels of both PD-L1 expression and T-cell infiltration were found (51 –53). As the vaccinia virus has no known entry receptor, one would hypothesize that viral infection would not limit the applicability of this approach to diverse histological subtypes. However, given the variability in the biological behaviors noted between these subtypes, the ability of infection to lead to a transformation of the TME and subsequent sensitivity to PD-1 blockade should not be assumed and requires confirmation in the clinical setting.

The combination of viral and PD-1 blockade prior to surgery has the potential for direct clinical translation. ILP is often used in the neoadjuvant setting prior to surgery for soft-tissue sarcomas (28). A function-preserving resection is typically performed between 6 and 8 weeks later, followed by adjuvant radiotherapy if viable tumor is found within the specimen. Although typically reserved for patients with locally advanced tumors, our data provide a rationale for extending the use of a modified neoadjuvant protocol to those patients at greatest risk of systemic relapse. Such patients are readily identifiable at diagnosis by such factors as tumor size, grade, and histologic subtype (54). As an alternative to standard practice, these high-risk patients could undergo a viral ILP followed by administration of anti-PD-1 antibodies for 6–8 weeks prior to surgery, which could continue thereafter. A phase I study to determine the tolerability and safety of combining OV and ILP, and to identify predictive biomarkers of response, is currently recruiting (NCT0355502). While it is recognized that trials combining such novel treatments are typically reserved for patients with established metastatic disease, our data add to an increasing body of evidence in various pathologies supporting the use of combination immunotherapies in early-stage disease.

K.J. Harrington reports receiving speakers bureau honoraria from AstraZeneca, Bristol-Myers Squibb, Amgen, Merck-Serono, and MSD, is a consultant/advisory board member for AstraZeneca, Amgen, Bristol-Myers Squibb, Merck-Serono, MSD, Pfizer, and Replimune, and reports receiving commercial research grants from AstraZeneca, MSD and Replimune. No potential conflicts of interest were disclosed by the other authors.

Conception and design: H.G. Smith, D. Mansfield, K.J. Harrington, A.J. Hayes

Development of methodology: H.G. Smith, D. Mansfield, V. Roulstone, J.N. Kyula-Currie, K.F. Bergerhoff, J.T. Paget

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H.G. Smith, V. Roulstone, J.N. Kyula-Currie, J.T. Paget

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H.G. Smith, M. McLaughlin, A. Khan, K. Thway, K.J. Harrington

Writing, review, and/or revision of the manuscript: H.G. Smith, M. McLaughlin, M.T. Dillon, A. Khan, A. Melcher, K. Thway, K.J. Harrington, A.J. Hayes

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H.G. Smith, V. Roulstone, J.N. Kyula-Currie, R.R. Patel, K.F. Bergerhoff

Study supervision: D. Mansfield, K.J. Harrington, A. Hayes

RNAseq and bioinformatics were performed by Kerry Fenwick, Alistair Rust, and Nik Matthews of the Tumour Profiling Unit (TPU) at The Institute of Cancer Research (ICR, London, United Kingdom). The authors are indebted to Prof. A. Eggermont for the donation of the BN175 cell line. H.G. Smith received academic grants from The Royal College of Surgeons of England (160729), Sarcoma UK (SUK203.2016/SUK203.2017), The Meirion Thomas Cancer Research Fund, and The McAlpine Foundation. V. Roulstone and J.N. Kyula-Currie were funded by Oracle Cancer Trust/The Mark Donegan Foundation. J.T. Paget and A. Khan were funded by the Wellcome Trust (WT098937MF; 200175/Z/15). K.J. Harrington acknowledges funding from the Rosetrees Trust (A1292). A. Melcher, K. Thway, K.J. Harrington, and A. Hayes acknowledge support from The Royal Marsden/The Institute of Cancer Research, National Institute for Health Research Biomedical Research Centre (NIHR BRC).

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.