Immune checkpoint blockade has shown unprecedented success in treating a variety of cancers [1]. Despite extensive efforts in translational research and clinical trials involving ICIs, there remain gaps in elucidating the mechanisms of response or resistance with these agents.

The objective response rates to anti-programmed death protein 1/ligand 1 (anti-PD-1/−L1) agents vary widely by tumor type, but overall average approximately 20% among tumor types with demonstrated efficacy [2]. The challenge is to identify the approximately one in five patients who are most likely to respond to single-agent ICIs or select those who may require more aggressive combination therapies. Carefully designed biomarker studies may generate clinically meaningful tests to select patients and potentially avoid unnecessary toxicities and reduce cost per life-year saved with ICIs. Predictive and mechanistic biomarker studies may also inform how to rationally combine ICIs with other modalities or agents and when to prioritize ICI use in the treatment sequence.

There are still very few validated predictive biomarkers (reviewed in [3]) and the mechanisms of sensitivity and resistance to ICIs are not completely understood. To date, the greatest focus has been on the degree of expression of PD-L1 on tumor and infiltrating immune cells, tumor infiltrating lymphocytes (TILs), tumor mutational burden (TMB), neoantigen load [4, 5], and T cell receptor (TCR) clonality (reviewed in [1, 3, 6]). While the most promising biomarker studies to date come from randomized, prospective trials, these studies are often limited in the scope of genomic and immune correlatives because they use archival specimens, and do not include on-treatment biopsies to obtain a mechanistic understanding of the dynamic antitumor immune response. The INSPIRE trial (NCT02644369) was carefully designed to leverage integrated genomic and immune parameters from freshly processed tissue biopsies and peripheral blood before, during, and after treatment with the anti-PD-1 monoclonal antibody, pembrolizumab. The primary objective of INSPIRE was to determine whether genomic and/or immune biomarkers were associated with response to pembrolizumab. In addition to informing the development of future biomarker studies and improving our mechanistic understanding of ICIs, this study provides a prioritized and optimized tissue and blood processing workflow and identifies practical issues to consider when designing in-depth correlative studies across different histologies.

For additional details related to study methods, please see Additional file 1: Supplementary methods.

From March 21, 2016 to June 1, 2017, 80 patients were accrued: eight SCCHN, 13 TNBC, 21 HGSC, 10 MM, and 28 MST patients. All patients were ECOG 0 or 1, and 79% of patients were treated with at least one prior systemic therapy. A range (0–100%) of baseline PD-L1 IHC MPS were observed, including 51% of patients with baseline tumor PD-L1 MPS of 0 (Additional file 2: Table S1).

The median (range) follow-up time was 4.1 (0–13.8) months, and the median number of cycles of pembrolizumab delivered in the SCCHN, TNBC, HGSC, MM and MST cohorts were: 4, 3, 3, 10 and 5, respectively. The objective response rates in these five cohorts were: 2/8 (25%), 1/13 (8%), 0/21 (0%), 6/10 (60%) and 4/28 (15%) with no CRs (Additional file 2: Table S3), and the median time-to-progression was: 3.4, 2.0, 2.8, not reached and 3.5 months, respectively (Additional file 2: Table S3, Additional file 4: Figure S2). Overall, pembrolizumab was well tolerated with the most common grade 1–2 adverse events (AEs) being fatigue (38%), diarrhea (21%) and nausea (16%) (Additional file 2: Table S4). Grade 3–4 adverse events were uncommon with the most frequent being fatigue (3%) and pneumonitis (3%) (Additional file 2: Table S4). Fifty of 57 (88%) patients who discontinued treatment did so due to disease progression. Four (21%) HGSC patients discontinued due to AEs, side effects, or complications (Additional file 2: Table S3).

This study provides an interim report from a phase 2 trial of pembrolizumab in a variety of solid tumors and preliminary translational data. PD-L1 status and TMB appear to enrich for response to ICIs, but application of these biomarker results and comparison against other studies in the literature are complicated by the use of different antibody clones, cell populations included or excluded during scoring, inter-observer variability, and thresholds for positivity. INSPIRE was not powered to discover and validate specific predictive biomarkers of ICIs, but rather, it was a hypothesis-generating study to comprehensively evaluate the baseline and dynamic changes in the genomic and immune landscape of several solid tumor histologies following treatment with pembrolizumab.

Some of the results generated from this study are largely technical in nature, and are provided to inform investigators who intend to perform similar translational analyses (Fig. 1). Previous studies have shown that peripheral blood and tumor-infiltrating T cells express differing levels of PD-1 [18], and here we show that this is also true for other T cell co-signaling molecules including 4-1BB and TIGIT (Fig. 2). Taken together, these studies demonstrate that fresh tumor biopsies are informative and feasible in many cases and peripheral blood immune biomarkers may not be an appropriate surrogate for ongoing immune responses in the tumor. This is an important point for future investigators to consider when designing correlative studies.

A handful of studies have identified possible biomarkers of response to ICIs in peripheral blood such as high lymphocyte or eosinophil counts, low baseline neutrophil, regulatory T cell or myeloid-derived suppressor cells (MDSC) counts, TCR gene richness and evenness at baseline [3, 19], and a high baseline frequency of classical monocytes [20]. In our study, we did not find any trends that are predictive of response to pembrolizumab based on peripheral blood samples taken at baseline; however, fold-expansion of CD4 T cells from baseline to week three was higher among responders (Fig. 5j). New promising indicators of response to ICIs, such as the post-pembrolizumab fold-change of Ki67+ PD-1+ CD8 T cells relative to tumor burden [21], CD28 expression on CD8 T cells [22] will be considered in future analyses.

TMB did not correlate with response in this study, but this was not unexpected given the heterogeneity of tumor types tested. PD-L1 MPS values by IHC and T cell PD-1 expression by flow cytometry trended toward enriching for response to pembrolizumab, but were overlapping between responders and non-responders. The addition of 4-1BB to the biomarker panel further separated responders from non-responders, better than T cell PD-1 as a single variable. 4-1BB is a co-stimulatory member of the tumor necrosis factor receptor superfamily and is upregulated on antigen-activated T cells [23]. The frequency of 4-1BB+ PD-1+ CD8 T cells in the tumor at baseline demonstrated a statistically significant separation of responders from non-responders (Fig. 5a-b), and validation of this finding using multicolor IHC is planned. Given that 4-1BB is induced by TCR signalling, it is interesting to speculate that the presence of 4-1BB+ PD1+ CD8 T cells may be indicative of an ongoing tumor-specific T cell response, which could be augmented by pembrolizumab. It has been previously shown in CD8 T cells freshly isolated from excised melanoma lesions that PD-1 expression identified tumor-reactive CD8 TILs [24]. While 4-1BB could also be used to enrich for tumor-reactive CD8 TILs, PD-1 expression more comprehensively captured the clonally expanded repertoire of tumor-reactive CD8 T cells. However, it has also been shown in ovarian cancer and melanoma that 4-1BB expression better identifies tumor-reactive CD8 TILs; 4-1BB+ CD8 TILs responded to peptide stimulation of a HLA-matched cancer cell line, while 4-1BB- CD8 T cells did not [25]. It is known that PD-1 is more broadly expressed on tumor and tumor-unrelated virus specific CD8 T cells, and other markers, such as CD39, are required to distinguish tumor-specific CD8 T cells [26]. Simoni et al compared tumor-specific CD8 T cells with tumor-unrelated CD8 T cells and found that while PD-1 expression was similar in both tumor-specific and viral-specific CD8 T cells, only the tumor-specific TILs expressed CD39, while the cancer unrelated CD8 TILs did not. This shows that PD-1 is expressed on a broader repertoire of both tumor-reactive and bystander CD8 T cells. Therefore, combining multiple parameters such as PD-1 and 4-1BB may help identify tumor-reactive CD8 TILs that have undergone recent TCR stimulation and gives rationale for using PD-1 and 4-1BB co-expression on CD8 T cells as a potential biomarker to predict response. The emerging importance of composite signatures or profiles by integrating multiple biomarkers is emphasized by recent studies; for instance, CD8 TIL infiltration and physical interaction of PD-1 and PD-L1 expressing cells were associated with clinical response to pembrolizumab in melanoma [27].

A unique feature of INSPIRE is the acquisition of on-treatment tumor biopsies which may identify pharmacodynamic biomarkers that shed insights into mechanisms of biological activity and resistance. The importance of on-treatment biopsies was demonstrated in ICI studies in MM, where CD8 TIL density after pembrolizumab [28] and fold-increase in TIL infiltration after ipilimumab, but not at baseline, was associated with objective responses [29]. Similar to previous studies [28, 29], fold-change in CD8, but not CD4 TILs from baseline positively correlated with response (Fig. 5c-d). On-treatment biopsies were performed coincidentally with the time that most patients had experienced PD-1 blockade in the tumor (Fig. 3). Given that MIH4 and EH12 clones both compete for the same epitope [30], we may consider using a secondary anti-IgG4 antibody in future studies to evaluate the level of PD-1 bound by pembrolizumab to test for occupancy of PD-1 and T cell co-signaling molecule expression kinetics. It may also be relevant to see if clinical response to pembrolizumab or other pharmacodynamic immunologic changes correlate with the degree or kinetics of intratumoral PD-1 blockade.

The current paper aims to share initial results with the community and provide an optimized sample workflow to others who may wish to conduct similar multidisciplinary studies on sequential fresh tumor biopsies. This interim report summarizes the data available from the first 80 enrolled patients, while on-treatment or end-of-treatment samples remain to be collected on these and additional patients at the time of publication. Preliminary translational data such as baseline frequency of 4-1BB+ PD-1+ T cells and intratumoral T cell fold-expansion will be validated with additional techniques and on more patients when data are available. This study has several limitations. Very few progression biopsy samples were collected and have not been analyzed, the study has a short duration of follow-up (4.1 months) and thus survival outcomes are not yet mature, and it is not powered to test for predictive genomic or immune biomarkers in multivariable models. As the number of patients with each tumor type is limited, the correlation between some biomarkers (e.g. PD-1+ 4-1BB+ CD8 T cells and TMB in MM) and the lack of correlation between others (e.g. PD-L1 and TMB) should be considered exploratory requiring validation in larger cohorts. An intriguing parameter that was omitted in this study was the microbiome, which may correlate with anti-tumor immunity and response to ICIs [31, 32]. A study such as this entails an extensive logistical and scientific undertaking that brings together clinicians and multidisciplinary laboratory researchers with the shared goal of understanding the complexities of immunotherapy.