Last week we were treated to a barrage of good news regarding PD-1/PD-L1 therapeutics and the ability to select responders. The centerpiece was a trio of papers in Nature.
Powles et al. presented data on the use of MPDL3280A, an anti-PD-L1 IgG1 antibody that has been engineered to lack all ADCC function (link 1). The antibody blocks the interaction of PD-L1 with PD-1 and with CD80, two receptors found primarily on lymphocytes. The paper focused on the application of ‘3280’ therapy in chemotherapy-resistant metastatic urothelial bladder cancer (UBC). Nearly all patients (93%) had failed platinum-based chemotherapy; 72% had failed 2 or more lines of prior therapy. 75% had visceral metastases, most had poor renal function and the majority (59%) had a performance score of 1 (very poor). In a word, these patients were incurable. Preliminary Phase 1 data demonstrating efficacy in UBC was presented at ASCO and led to breakthrough designation for ‘3280 for the treatment of UBC in June 2014.
The original Phase 1 trial had enrolled UBC patients whose resection or biopsy tissue demonstrated the presence of tumor-infiltrating lymphocytes (TIL) with dark staining (score 2 or 3) for PD-L1. The expansion cohort allowed for the enrollment of patients whose tissue specimens contained TIL which were PD-L1 dim (score = 1) or negative. 205 patient tissues were analyzed (see table 1 in the paper). 67 patients were enrolled and evaluable with PD-L1 staining results as follows:
A total of 17 patients responded and 16/17 responses were ongoing (i.e. durable) at the time of data cutoff. The longest duration of response was a remarkable 30 weeks in the cohort with the brightest PD-L1 TIL staining, although the range was broad (from 1 week to 30). Median duration of treatment was 9 weeks, so this is really an early snapshot. Regardless, the ability to invoke an anti-tumor response in a cohort of patients that are this ill, and deemed incurable, is remarkable.
With reference to the staining pattern of PD-L1 and the relevance of PD-L1 expression to successful response, the authors came to the following conclusions:
1) therapy triggered expansion of the circulation CD8+ T cell population, and transient elevation of IL-18 and IFNgamma was observed; these systemic changes reflect the proposed mechanism of action of ‘3280 but did not correlate with response.
2) expression of PD-L1 on TIL, but not on tumor cells, was predictive of response to therapy. On note, this was true whether the available tissue sample was new acquired or archival (up to 10 years old). This suggests that there is an ongoing and futile immune response in these PD-L1+/TIL+ tumors. The lack of association with tumor PD-L1+ status is discussed more extensively in the companion paper (see below).
3) the efficacy of PD-L1-directed therapy in UBC and also NSCLC and melanoma, all tumors with very high mutational burdens, suggests that antigen diversity or antigen “burden” may be important for successful induction of an anti-tumor immune response in ‘3280-treated patients.
The UBC cohort was part of a much larger clinical trial that included diverse solid tumors. A companion paper by Herbst et al. investigates the utility of PD-L1 TIL expression in other cohorts (link 2). The focus of this work is on the biomarker application, particularly with respect to PD-L1+ TIL staining, as defined in the prior paper. Patients (n=277) with advanced incurable cancers were enrolled in a ‘3280 dose ranging study, given drug iv every 3 weeks. Across tumor types high PD-L1 expression on TIL, but not tumor cells, was associated with response and increased PFS. Note here that the PFS gain, while encouraging, does not suggest that we will see a high percentage of truly durable (“long tail”) responses in this particular patient population, even in those patients with PD-L1 bright (score of 3) staining:
There were some interesting additional analyses. In NSCLC patients who had been smokers, 43% responded to therapy, while only 10% of non-smokers responded. Such data have been reported before, and are often taken to mean that the higher mutational burden seen in smokers with NSCLC biases their tumor toward immune recognition (this echoes the mutational diversity/mutational burden argument made in the Powles UBC paper). Sticking with NSCLC, 83% of patients with a PDL1+ TIL staining score of 3 (lots of cells and therefore dense/dark staining) responded versus 38% of patients with a PDL1+ TIL staining score of 2 (diffuse staining, fewer cells). Response was positively correlated with CTLA4+ staining on TIL, and negatively correlated with fractalkine expression. In melanoma (but not NSCLC or RCC) response was associated with elevated IFNgamma and IDO1 and CXCL9 that are induced by IFN gamma. Strikingly, positive anti-tumor responses were not associated with a measureable change in FoxP3 expression, suggesting the T regulatory T cells were not playing a role in the setting of ‘3280 therapy.
What about the non-responders, as these make up the majority of the patients across indications? Progressing tumors were characterized into three classes:
1) few or no TIL present – “immune ignorance”
2) TIL present but little or no PD-L1 expression – “non-functional immune response”
3) TIL present and PD-L1+ but located on the edge of the tumor – “excluded infiltrate”
Missing here I think is an analysis of tumors with PD-L1+ TIL with high staining scores (2 or 3) that progressed, i.e. did not respond to therapy. It seem to me unlikely that these all fell into category “3” above, so this analysis may be coming in a follow-up paper.
The authors make a very interesting point about this data, which is that they seem to refute the consensus model of “immune resistance” in which it is postulated that CD8+ T cells infiltrating tumors secrete IFNgamma and other cytokines that induce PD-L1 expression on the tumor cells themselves, and these tumor cells in turn produce factors that create an immunosuppressive environment that includes potently immunosuppressive, PD-L1 bright T regulatory cells. The “immune resistance” model further postulates that the expression of PD-L1 on tumor cells and T regulatory cells is responsible for shutting down CD8+ T cells by binding to PD-1.
There are several key messages in this paper – first, responses in these incurable patients are measureable and remarkable, if they respond (most do not). Second, CD8+/PD-L1+ TIL are highlighted as a potential prognostic indicator of the potential for response the ‘3280 therapy. Finally, it is clear that other signals will have to be disabled or enhanced in order to induce a productive and durable immune response in more patients and/or move PD-1/PD-L1-directed therapies to front line.
Now, the final paper in this triad turns things upside down. Tumeh et al. analyzed tumor tissue samples from 46 metastatic melanoma patients treated with pembrolizumab, an anti-PD-1 antibody (link 3). The analytic methods used are elegant and overlap but also extend the analyses used in the prior 2 papers: quantitative immunohistochemistry, quantitative multiplex immunofluorescence, and TCR deep sequencing (NGS).
This paper is strictly about melanoma. The ORR in this small study was 48% (22/46). The authors focused on expression of PD-L1 on tumor cells and of PD-1 on CD8+ T cells. Doing so they come to strikingly different conclusions than the papers discussed above. Responders in this study had PD-1+ CD8 T cells massed on the tumor margin, adjacent to PD-1+ tumor cells. Response was associated with infiltration of the tumor by those CD8+ T cells, which also increased in density (proliferated). Therefore the paper specifically supports the “immune resistance” model in which tumor-expressed PD-L1 suppresses PD-1+ CD8 T cells. CD8 T cell proliferation was associated with expression of granzyme B within the tumor and phosphorylated STAT1 at the tumor margin where CD8+ T cells were infiltrating (phospho-STAT1 in induced by IFNgamma receptor signaling). Finally, response was associated with T cell (TCR) clonality, i.e. the fewer tumor antigens, and thus the lower the antigen burden that is invoking a response, the better. This is a different take than we got from the prior papers.
So, perhaps melanoma is distinctly different.
Aside from that, these papers provide critical take-home messages and perhaps even more critical questions to be addressed:
1) CD8 T cells are good. That’s pretty clear, whatever they are expressing. We can argue more about their geography, but if they are not present, you will not respond.
2) IFNgamma is good. We see this especially in the melanoma setting as detailed in two of the papers.
Neither of these conclusions is novel nor surprising.
3) Biomarker development beyond CD8+ T cell staining remains complex.
4) Regardless of their biomarker status most patients still do not respond and we do not know why. As we consider combination therapy, will other markers be used to further sort patients into rational combination buckets, or will this simply too complex to be useful?
5) Finally, what about those T regulatory cells we’ve been obsessed with for the last decade? These are hardly mentioned in the context of PD-1/PD-L1 therapeutics in the three studies.
>>> back to those tumor antigens? New papers, preclinical and clinical, shed some light… and
>>> those T regulatory cells may be important in some settings, but were betting on the tumor microenvironment to yield interesting new targets for therapy