Improved Survival Associated with Local Tumor Response Following Multisite Radiotherapy and Pembrolizumab: Secondary Analysis of a Phase I Trial

This article is featured in Highlights of This Issue, p. 6399

Tumors with sustained responses to immunotherapy demonstrate a T cell–inflamed tumor microenvironment characterized by tumor-infiltrating lymphocytes and type I/II IFN gene expression profiles (1, 2). In contrast, tumors that fail to respond to immunotherapy frequently exhibit a non–T-cell-inflamed tumor phenotype. High-dose stereotactic body radiotherapy (SBRT) has been reported to promote immunogenic cell death through activation of innate and adaptive immunity orchestrated by IFN signaling (3, 4). These observations have led to several investigations exploring the use of radiation to enhance response to immunotherapy.

We previously reported the initial results of a clinical trial investigating the safety of treating metastases in multiple body sites with SBRT followed by pembrolizumab (5). Dose-limiting toxicities (DLTs) were seen in <10% of enrolled patients with advanced solid tumors. We observed a RECIST objective response rate (ORR) of 13% in a population of predominately programmed death ligand-1 (PD-L1)-low tumor types, which was similar to the ORR of PD-L1–positive patients with cancer treated with pembrolizumab alone on KEYNOTE-28 (6). Our finding indicates that the primary effect of combination SBRT+P was localized to irradiated metastases and not to systemic, nonirradiated metastases. In addition, we observed >90% control rates of irradiated metastases at 4 months even though many metastases received the prescription dose to only a portion of the tumor volume to minimize potential toxicity, which has historically resulted in poor local control. Therefore, a secondary analysis of this trial was performed to investigate the biological effects of SBRT as they relate to pembrolizumab response, we analyzed gene expression profiles of irradiated tumors and their impact on treated metastasis control (TMC). Here, we report outcomes based on local tumor response and tumor gene expression patterns.

Patients with advanced solid tumors were enrolled between January 2016 and March 2017 (clinicaltrials.gov: NCT02608385; ref. 5). Study enrollment began after the institution internal review board approved the study, written informed consent was obtained, and pursued per Declaration of Helsinki. Data lock occurred on May 1, 2018. Patients received SBRT to at least two measurable metastases with each receiving 30–50 Gy over 3–5 fractions (5, 7). Metastases were prioritized on the basis of the largest size and/or those causing the most morbidity. The SBRT prescription dose was limited to a maximum of 65 mL such that tumors >65 mL received the prescription dose to part of the metastasis (partial-Rx). Otherwise, tumors received SBRT to the entire metastasis (complete-Rx). Pembrolizumab (200 mg i.v. every 3 weeks) began within 7 days following SBRT (SBRT+P). Tumor assessments were performed every 2–3 months using Response Evaluation for Solid Tumors 1.1 (RECIST) principles. Osseous and spinal metastases were considered controlled as part of the TMC measurement as per MD Anderson response criteria (8). Under FDA guidance, we attributed DLTs to the combination of SBRT+P therapy by anatomic organ system, and not to each specific therapy.

Thirty-five tumor biopsies were collected prior to and/or within 7 days following SBRT, but prior to pembrolizumab, from 24 patients: 19 pre-SBRT and 16 post-SBRT (11 matched samples from the same tumor; Supplementary Table S1). RNA was extracted and analyzed using Affymetrix Human Clariom D microarrays. Computational gene expression deconvolution methods were used for genome-wide expression analyses following SBRT and correlated with irradiated tumor response. We utilized Ingenuity Pathway Analysis to predict the upstream regulators of pathways involved in the gene expression changes. See Supplementary Data for details regarding baseline PD-L1 expression, microarray data preprocessing, and detection of treatment-responsive gene expression changes. A TaqMan RT-PCR assay (Thermo Fisher Scientific) was used to validate DNASE1 gene expression utilizing an exon-spanning probe (Hs00173736_m1) overlapping the Clariom D transcript cluster TC1600011345.hg. Expression values for DNASE1 were normalized to the geometric mean of β-actin and 18S rRNA.

Baseline characteristics were compared between patients who received complete-Rx (to all treated lesions) and partial-Rx (to at least one of the lesions) using t tests and Pearson χ² test for continuous and discrete variables, respectively. TMC was defined as a lack of progression for an irradiated metastasis (8–10), and duration of control was estimated using the Kaplan–Meier method. Death without irradiated tumor progression led to censoring at time of death (10). TMC and overall treatment response (i.e., CR, PR, SD, or PD) were evaluated using shared frailty Cox regression models and generalized estimating equations to account for correlation of individual metastasis responses within patients. At the first scan, patients who had a 30% or greater decrease (50% for osseous/spine) in the maximum transverse diameter of each irradiated metastasis were classified as responders, and those who had a 20% or greater growth (25% for osseous/spine) in the maximum transverse diameter of one or more irradiated metastases were nonresponders. Patients who did not fit either category were deemed to have stable disease (mixed responders). We used the Kaplan–Meier method to estimate progression-free survival (PFS) utilizing the time from SBRT to either progression or death from any cause and overall survival (OS) using the time from SBRT to death. Landmark analysis at 2 months was used to avoid a guarantee time bias when examining the association of early irradiated metastasis response with OS (11). We stratified outcomes using clinical, prognostic, and pathologic variables and compared differences using the log-rank test and Cox regression modeling. Baseline PD-L1 staining was available from 51 analyzable patients, and a cut-off of ≥1% was used to define PD-L1 positivity. P values <0.05 were considered statistically significant. Multivariable Cox regression analysis was performed for covariates with univariate P values <0.10. Data were analyzed using Stata software version 15.0 and JMP version 14 (SAS Institute Inc.).

Seventy-nine patients were enrolled and 11 patients were excluded from the analysis due to: not receiving SBRT (n = 3); refusing pembrolizumab (n = 3); or lost to follow-up (n = 5). The overall median follow-up time (n = 68) at data lock was 10.4 months (range, 2.3–24.1), and was 33.4 months (range, 15.3–42.5) for the 17 patients alive past the data lock. Sixty-eight patients with 26 different, nonradiosensitive and/or generally non-PD-1 responsive, solid tumor types who received SBRT to 139 metastases were analyzed (Supplementary Table S2). Patients had previously received a median of 5 lines of systemic therapy, and 3 had received checkpoint therapy. Table 1 summarizes baseline patient-level (n = 68) characteristics stratified on the basis of the receipt of partial-Rx to at least one metastasis. A representative example of partial-Rx versus complete-Rx treatment is shown in Supplementary Fig. S1. In 50 patients (73%), 102 metastases received the radiation prescription dose to the entire tumor volume (complete-Rx), whereas in 18 patients (26%) at least one metastasis (21 of 37 metastases) received the prescription dose to less than the entire tumor volume (partial-Rx). There were no differences in tumor histology (P = 0.19), baseline PD-L1 status (P = 0.92), or prior lines of therapy (P = 0.54) between patients stratified by who received all complete-Rx versus at least one partial-Rx. There were only six DLTs: six (all metastases complete-Rx), 0 (at least one metastasis partial-Rx), for an overall rate of 8.8%.

The 1-year Kaplan–Meier TMC rate was 89.5% (Fig. 1A) consistent with other reports following complete volumetric tumor coverage with high-dose SBRT (12–14). However, 21 of the 139 treated metastases received partial-Rx SBRT+P. The partial-Rx treated metastases (Supplementary Table S4) were larger at baseline (mean volume: 157.6 mL vs. 12.8 mL, P < 0.001) and more likely to be located in the abdomen or pelvis (P < 0.001; Supplementary Table S3). We observed no significant difference in TMC between complete-Rx versus partial-Rx SBRT+P (Fig. 2) despite differences in the metastasis target volume receiving at least 95% of prescribed radiation dose (median volume for partial-Rx vs. complete-Rx: 67.2% vs. 100%); however, the low number of events limits statistical power to draw firm conclusions. Moreover, there were no significant differences in TMC when stratified by baseline PD-L1 status (P = 0.81; Fig. 1B), or tumor volume irradiated (P = 0.86; Fig. 1C). In contrast to typical SBRT plans (15), we observed similar TMC rates independent of the metastasis size or volume receiving the complete prescription dose when delivered prior to pembrolizumab.

We analyzed treatment responses of individual metastases and at the patient level (Supplementary Fig. S2). Seven metastases (5%) had a complete response and 42 (30.2%) metastases had a partial response for an overall treated metastasis response rate of 35.2%. Eighty-one metastases (58.3%) were stable with nine (6.5%) demonstrating immediate progression following SBRT+P. No differences in SBRT+P response were observed between patients who received complete-Rx versus partial-Rx SBRT+P (P = 0.15). Partial-Rx SBRT+P did not affect local treatment response (P = 0.35; Supplementary Fig. S3A), nor did tumor volume (P = 0.44; Supplementary Fig. S3B), baseline PD-L1 status (P = 0.12; Supplementary Fig. S3C), or histology (P = 0.41; Supplementary Fig. S3D). Supplementary Table S5 summarizes all factors analyzed in this regression model.

The ORR of the nonirradiated metastases was 13.2% as reported previously. The median PFS was 3.1 months (95% CI 2.9–3.5; Supplementary Fig. S4A) and the median OS was 9.9 months (95% CI, 7.0–13.5; Supplementary Fig. S4B). No covariates were predictive of OS on univariate Cox regression (Supplementary Table S6). Also, no significant differences in OS were observed between patients stratified by partial-Rx status (Supplementary Fig. S5A) or baseline PD-L1 expression (Supplementary Fig. S5B). These findings demonstrated that TMC to SBRT+P occurred independent of baseline or treatment characteristics.

We analyzed OS based on irradiated tumor response (landmark analysis). This is supported by a direct relationship between the change in size of unirradiated lesions as a function of response of irradiated lesions (Supplementary Fig. S6). Only one patient died prior to their first scan, and of the 67 patients who survived beyond the 2-month landmark time, the median OS was 3.5 months for nonresponders (95% CI, 2.3–9.6 months), 9.0 months (95% CI, 6.5–12.9 months) for mixed responders, and 17.8 months (95% CI, 7.7–37.7 months) for responders (P = 0.0007; Fig. 3). Moreover, on post-landmark Cox regression, a larger irradiated tumor response, coded as an ordinal variable (0 = no response, 1 = mixed response, 2 = response), decreased the risk for death (HR, 0.44; 95% CI, 0.23–0.85; P = 0.015). These results demonstrate that patients whose tumors exhibited a local response to the combination of SBRT+P were more likely to experience extended survival.

We investigated radiation-induced gene expression changes associated with local response in serial biopsies acquired from irradiated metastases. First, we performed a genome-wide analysis to characterize gene expression changes following SBRT from 11 patients with both pre- and post-SBRT biopsies (matched samples). SBRT increased the expression of pathways enriched for innate and adaptive immunity, such as B-cell development, antigen presentation, and dendritic cell maturation, while concomitantly reducing the expression of pathways involved in G₂–M cell-cycle progression and DNA damage repair processes (Fig. 4A). Immune genes in the activated pathways were predicted to be regulated by type I/II IFN signaling, whereas downregulation of cell cycle and DNA repair processes were predicted to be in response to RAD21 and CDKN2A signaling pathways (Fig. 4A; ref. 16). Furthermore, the response to SBRT was associated with suppression of histone subunits and genes mediating mitosis and cell division, such as CDC20, CENPM, MCM5, and PRC1 (Supplementary Fig. S7). In contrast, SBRT led to the induction of CXCR3, which has shown to regulate leukocyte trafficking and to promote Th1 and CD8⁺ effector cell migration into inflamed tissues (17, 18). These results are consistent with preclinical models demonstrating that high-dose SBRT impacts pathways associated with cell cycle, DNA repair, and antitumor immunity.

We then examined 16 patients with post-SBRT (pre-pembrolizumab) biopsies to determine whether the same changes in gene expression were associated with irradiated tumor response. Among the 30 top-ranked genes in terms of correlation with irradiated tumor response, we found that elevated expression of immune genes, including DNASE1 (P = 0.00029) and CCR10 (P = 0.00069), were associated with increased irradiated tumor response, while elevated expression of TGFBR3L (P = 0.0014) and SIGLEC15 (P < 0.0001) were strongly associated with decreased tumor response (Fig. 4B; refs. 19–21).

DNASE1 has been shown to promote DNA degradation in the response to ionizing radiation, which may influence activation of cGAS-STING signaling in tumor cells (3, 22–24). Increased DNASE1 levels following SBRT were associated with concomitant increases in the expression of cytolytic T-cell genes (granzyme A and perforin; ref. 25; Fig. 4C) and a 2.1-fold improved local tumor response (Fig. 4D). RT-PCR analysis showed that increased DNASE1 expression following SBRT was associated with a reduction in size of irradiated metastases in the response to SBRT+P. Furthermore, the patient with the highest DNASE1 expression exhibited the greatest reduction in tumor burden (Supplementary Fig. S8). In contrast, increased TREX1 (DNase III) expression levels following SBRT were associated with decreased cytolytic gene expression (Fig. 4C) in concert with a 2.4-fold reduced local tumor response (Fig. 4D; ref. 26). These findings indicated that SBRT induces gene expression changes in immune pathways, which are associated with local tumor response in the presence of adjuvant pembrolizumab.

We initially reported the safety of multisite SBRT+P in patients with advanced solid malignancies (5). Here, we report that ORR, PFS, and OS did not differ between patients who received complete-Rx versus partial-Rx SBRT to multiple sites in the context of pembrolizumab. Also, clinical responses at the irradiated site could be induced without irradiation of an entire metastasis. Moreover, gene expression changes correlating with SBRT+P treatment response were enriched for innate and adaptive immune pathways. Finally, we found that the local response of irradiated metastases to SBRT+P was associated with OS.

Because TMC following radiotherapy in the absence of immunotherapy is a function of the overall dose, the dose per fraction, and the overall time during which the radiation is given (12–15, 27–29), we examined the relationship between irradiated tumor volume and radiation dose. Previous studies have established the minimum dose required for local control following SBRT alone is 36 Gy in 3 fractions (10, 30). While in our partial-Rx cohort, the median value for tumor minimum dose was 9 Gy in 3 fractions, which is expected to have poor rates of local control (31–33). Moreover, partial-Rx treatment increased normal tissue sparing compared with the complete-Rx group (see Supplementary Methods for biological effective dose calculations). Thus, partial-Rx treatment in combination with pembrolizumab is feasible and should be examined further.

The correlation of survival with the local response of the irradiated tumors, and not with baseline patient characteristics suggesting that multisite SBRT can limit progression of existing metastases while also augmenting antitumor immune responses to potentially improve outcomes in metastatic patients treated with pembrolizumab. Cytoreduction through radiation may be important, particularly given the clinical observation indicating that patients with lower disease burden treated on prospective immune checkpoint therapy trials have improved outcomes (34).

We also found that clinical responses could be induced, regardless of tumor PD-L1 status, without irradiation of an entire metastasis and also resulted in less toxicity. Despite the partial treatment of these metastases, their inherently larger size, and the reduced overall dose to the metastasis, we did not detect any difference in TMC when compared with tumors that received the prescribed radiation dose to the entirety of the lesion. Specifically, 15/16 (94%) metastases in our partial-Rx group responded despite receiving lower radiation doses. On the basis of prior studies using single modality SBRT prescribed to the entire metastasis volume in which 25.5 Gy was prescribed in 3 fractions, we expected few metastases receiving partial-Rx to have radiographic response (35). We propose the response seen in our study is likely multifactorial and related to both cytoreduction and local immune enhancement as previously demonstrated in preclinical models when stereotactic radiation was combined with immune checkpoint therapy (36). However, we await further prospective evidence to further understand the contribution of these two potential mechanisms. Of note, the high TMC rate seen was not associated with increased toxicities given that no DLTs occurred in the partial-Rx group. Collectively, these data suggest that a multisite SBRT approach, including partial-Rx, followed by immunotherapy may have broad applicability by enhancing response and limiting adjacent organ toxicity in patients with advanced cancer.

Consistent with preclinical evidence and the recent ORIOLE phase II randomized trial, we observed that SBRT induced both innate and adaptive immune pathways, while concomitantly decreasing genes involved in cell cycle and DNA damage repair pathways (3, 4, 37). We found overexpression of the DNase I (DNASE1) endonuclease following SBRT was associated with favorable irradiated tumor response to SBRT+P, whereas overexpression of the DNase III (TREX1) exonuclease was associated with unfavorable local tumor response. These findings suggest that differential DNA degradation in the response to SBRT might contribute to distinct treatment responses to SBRT+P. Moreover, overexpression of SIGLEC15, recently described as an important immune suppressor in the context of checkpoint blockade (38), following SBRT emerged as potential negative regulator of SBRT+P response in our analysis. Taken together, our finding that there is no difference in tumor control whether patients receive partial-Rx or complete-Rx, our work suggests that the ablative function of radiation may not be the only aspect of radiation, which should be considered. In fact, our work is novel in suggesting that the other aspects of the biological response to radiation (which are not accounted for in the α/β ratio) may be just as important.

While there is generally a lack of consensus regarding the optimal radiation dose (thought to be in the range of 8–12 Gy per fraction; refs. 10, 26, 30) and the degree of tumor coverage required to potentiate systemic immunity (39). In our study, we used higher radiation doses per treatment, and although we found an inverse relationship between treatment response and the post-SBRT expression of TREX1, we also observed that the majority of irradiated tumors responded to treatment and exhibited high levels of post-SBRT immune gene expression. We recognize that even in the context of higher fraction irradiation some portion of the tumor receives a lower dose, especially in partial-Rx tumors. However, when we investigated the percentage of the tumor that received 8–12 Gy we did not find an association with irradiated tumor or overall patient response. These findings suggest that, in humans, the optimal radiation dose and volume to synergize with immunotherapy remains undefined.

While these analyses were generated from a prospective clinical trial, there are several limitations of our investigation. For example, only 18 patients and 21 (of 37) metastases were treated with partial-Rx SBRT+P. Additional prospectively treated patients are needed to confirm the favorable local tumor response with partial-Rx SBRT+P. The single-arm nature and heterogeneous patient population within the phase I trial also have limitations and randomized evaluation within a specific disease setting is required. Regarding gene expression analysis, RNA isolation was prioritized from tumor biopsies limiting the availability of additional tissue for exploratory analyses.

Multisite SBRT, including partial-Rx SBRT to large lesions, in combination with immunotherapy may provide a benefit to the irradiated metastases, irrespective of PD-L1 status. This treatment paradigm is feasible and clinically available to patients in the standard-of-care setting. Furthermore, the association of treatment response to radiation (TMC) with OS suggests a potential clinical and radiographic biomarker that is easily evaluable. Finally, gene expression profiling changes should similarly be assessed in other studies using a range of radiation doses to identify the optimal dose and schedule for immune potentiation.

J.J. Luke reports grants from Merck (Merck Investigator Sponsored Program) during the conduct of the study; scientific advisory board relationships (no stock) with 7 Hills, Spring bank (stock) Actym, Alphamab Oncology, Arch Oncology, Kanaph, Mavu, Onc.AI, Pyxis, Tempest; consultation with compensation for Abbvie, Aligos, Array, Bayer, Bristol-Myers Squibb, Checkmate, Cstone, Eisai, EMD Serono, KSQ, Janssen, Merck, Mersana, Novartis, Partner, Pfizer, RefleXion, Regeneron, Ribon, Rubius, Silicon, Tesaro, Werewolf, Xilio, Xencor; research support from (all to institution for clinical trials unless noted) AbbVie, Agios (IIT), Array (IIT), Astellas, Bristol-Myers Squibb (IIT & industry), Corvus, EMD Serono, Immatics, Incyte, Kadmon, Macrogenics, Merck, Spring bank, Tizona, Xencor; travel compensation from Bristol-Myers Squibb, Janssen, Mersana, Pyxis; and patents from (both provisional) Serial #15/612,657 (Cancer Immunotherapy), PCT/US18/36052 (Microbiome Biomarkers for Anti-PD-1/PD-L1 Responsiveness: Diagnostic, Prognostic and Therapeutic Uses Thereof). J.W. Moroney reports other from Clovis (honoraria) outside the submitted work. J.D. Patel reports personal fees from AstraZeneca (advisor) and Takeda (advisor) and grants and personal fees from BMS (advisor; institution) outside the submitted work. P.A. Ott reports grants and personal fees from Neon Therapeutics, Merck, CytomX, Roche/Genentech, Celldex; grants from BMS, ArmoBiosciences, Pfizer, AstraZeneca/MedImmune; and personal fees from Novartis and Array outside the submitted work. G.F. Fleming reports other from Merck during the conduct of the study; personal fees from GSK (also site PI for industry sponsored study); other from corcept (supplied drug for mouse studies. Also am site PI for industry sponsored study); Abbvie (site PI for industry sponsored study), Roche (site PI for industry sponsored study), 47 inc (site PI for industry sponsored study), Iovance (site PI for industry sponsored study), Syros (site PI for industry sponsored study), Merck (site PI for industry sponsored study), Sanofi (site PI for industry sponsored study), Sermonix (site PI for industry sponsored study), Compugen (site PI for industry sponsored study), Incyte (site PI for industry sponsored study), Eisai (site PI for industry sponsored study), and plexxicon (site PI for industry sponsored study) outside the submitted work. T.F. Gajewski reports grants from Merck during the conduct of the study and personal fees from Merck (advisory board) outside the submitted work. R.R. Weichselbaum reports other from Boost Therapeutics (stock and other ownership interests), Immvira LLC (stock and other ownership interests), RefleXion Pharmaceuticals (stock and other ownership interests), Coordination Pharmaceuticals Inc. (stock and other ownership interests), Magi Therapeutics (stock and other ownership interests), and Oncosenescence (stock and other ownership interests) during the conduct of the study; grants from Varian, Regeneron, AstraZeneca (travel), Boehringer Ingelheim LTD (travel), and Merck Serono S.A. (travel) outside the submitted work; in addition, R.R. Weichselbaum has a patent for Methods and Kits for Diagnosis and Triage of Patients with Colorectal Liver Metastases pending and has served in a consulting or advisory role for Aettis Inc., Astrazeneca, Coordination Pharmaceuticals, Genus, Merck Serono S.A., Nano proteagen, NKMax America Inc, Shuttle Pharmaceuticals, Highlight Therapeutics, S.L. S.J. Chmura reports grants from Merck during the conduct of the study, as well as other from AStellas Pharma (wife employment) and grants from RefleXion (trials support) and BMS (trial support) outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

J.J. Luke: Conceptualization, methodology, writing-review and editing. B.E. Onderdonk: Software, formal analysis, investigation, writing-original draft, writing-review and editing. S.R. Bhave: Conceptualization, software, writing-original draft. T. Karrison: Data curation, software, formal analysis, supervision, writing-review and editing. J.M. Lemons: Conceptualization, writing-review and editing. P. Chang: Data curation, methodology. Y. Zha: Data curation, methodology. T. Carll: Data curation, methodology. T. Krausz: Data curation, methodology. L. Huang: Data curation, methodology. C. Martinez: Data curation, software, methodology. L.A. Janisch: Methodology, project administration. R.D. Hseu: Methodology, project administration. J.W. Moroney: Methodology, writing-original draft. J.D. Patel: Conceptualization, supervision, investigation, writing-original draft, writing-review and editing. N.N. Khodarev: Methodology. J.K. Salama: Methodology, writing-original draft. P.A. Ott: Methodology, writing-original draft. G.F. Fleming: Supervision, writing-original draft, writing-review and editing. T.F. Gajewski: Methodology, writing-original draft. R.R. Weichselbaum: Supervision, methodology, writing-original draft, writing-review and editing. S.P. Pitroda: Conceptualization, formal analysis, supervision, methodology, writing-original draft, writing-review and editing. S.J. Chmura: Conceptualization, formal analysis, investigation, methodology, writing-original draft, writing-review and editing.

We thank the patients and their families who participated in this study. We thank Bernadette Libao, Jill Stetkevych, and Lauren Wall for logistical support of the trial. We thank Merck & Co., Inc. for drug and grant support. J.J. Luke and S.J. Chmura had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

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.