Checkpoint musical chairs
Anti–PD-1 or PD-L1 antibodies can reinvigorate antitumor immunity in a subset of patients with cancer. To better understand the mechanisms behind successful therapy, Mayoux et al. characterized various ligands on the surface of dendritic cells (DCs). PD-L1 on DCs can bind B7.1 on the same cell, potentially preventing PD-1 ligation on T cells or B7.1 ligation of its partner CD28. They saw that PD-L1 was expressed in excess of B7.1, likely preventing T cell stimulation through these two pathways. Patients with a high DC signature before treatment were more likely to respond to PD-L1 blockade. These results reveal that in cis interactions on DCs have immunological and likely clinical consequences for checkpoint blockade therapy.
Abstract
PD-L1/PD-1 blocking antibodies have demonstrated therapeutic efficacy across a range of human cancers. Extending this benefit to a greater number of patients, however, will require a better understanding of how these therapies instigate anticancer immunity. Although the PD-L1/PD-1 axis is typically associated with T cell function, we demonstrate here that dendritic cells (DCs) are an important target of PD-L1 blocking antibody. PD-L1 binds two receptors, PD-1 and B7.1 (CD80). PD-L1 is expressed much more abundantly than B7.1 on peripheral and tumor-associated DCs in patients with cancer. Blocking PD-L1 on DCs relieves B7.1 sequestration in cis by PD-L1, which allows the B7.1/CD28 interaction to enhance T cell priming. In line with this, in patients with renal cell carcinoma or non–small cell lung cancer treated with atezolizumab (PD-L1 blockade), a DC gene signature is strongly associated with improved overall survival. These data suggest that PD-L1 blockade reinvigorates DC function to generate potent anticancer T cell immunity.
INTRODUCTION
PD-L1/PD-1 checkpoint inhibition as cancer immunotherapy induces durable clinical responses in several cancer types, highlighting the fundamental importance of PD-L1/PD-1 in restraining preexisting anticancer T cell immunity (1–6). It is currently believed that blocking the interaction between PD-L1 [on tumor cells (TCs) and immune cells (ICs)] and PD-1 (on T cells) reinvigorates dysfunctional tumor-infiltrating effector T cells to overcome adaptive immune resistance (7–9). Recent evidence suggests that PD-L1 expressed by ICs, rather than by TCs, is a better biomarker to predict clinical response (4, 10, 11). Whereas PD-1 has two ligands PD-L1 and PD-L2, PD-L1 also binds B7.1 (CD80) (12), a key costimulatory molecule expressed by antigen-presenting cells such as dendritic cells (DCs). PD-L1 can bind B7.1 in cis on the same cell (13–15); if this interaction occurs on DCs, B7.1 might trap PD-L1, preventing PD-1 ligation and inhibition of T cell function (16, 17). PD-1 signaling also appears to directly affect CD28 signaling axis (18, 19), suggesting that B7-directed costimulation from infiltrating ICs contributes to T cell reinvigoration in response to PD-L1/PD-1 blockade. Tumor-associated DCs display an immature state and are frequently defective in their functional activity and contribute to immune suppression. They seem to either lack requisite stimulatory signals or are actively kept in an inert state by inhibitory signal (20–22). The precise role of DCs in educating T cells in the tumor to respond to PD-1/PD-L1 blockade is largely unknown. We hypothesized that PD-1/CD28 on T cells and PD-L1/B7.1 on DCs participate in a complex network and thereby regulate T cell priming in the tumor microenvironment.
RESULTS
PD-L1 is more abundantly expressed than B7.1 on DCs in patients with cancer
We first evaluated the expression of PD-L1 and its two receptors PD-1 and B7.1 on tumor and peripheral DC subsets from patients with lung cancer. The conventional DCs (cDCs) are classically divided into two subsets: CD141+ and CD1c+. Both peripheral DCs and tumor-associated DCs express high PD-L1 under steady-state conditions (Fig. 1A). The CD141+ and CD1c+ DC subsets share similar expression patterns. Tumor-associated cDCs express B7.1, whereas peripheral cDCs do not. Marginal expression of PD-1 was observed on cDC CD141+ subsets. This expression pattern on DCs from patients with cancer was similar to the profile on peripheral DCs from healthy individuals (figs. S1 and S2). In contrast, plasmacytoid DCs (pDCs) only express PD-L1 (Fig. 1A). We observed PD-L1+B7.1+Clec9a+ cDC cells (fig. S3), a DC subset that is capable of cross-presenting antigen to CD8+ T cells (23), in fresh renal cell carcinoma (RCC) samples. Thus, we hypothesized that potential cis interaction of PD-L1/B7.1 and PD-L1/PD-1 on these tumor-associated DCs might be critically relevant in the regulation of immune function. It was recently suggested that the cis PD-L1/B7.1 interaction on DCs may limit PD-1 function when DCs express a substantial amount of B7.1 (16). However, when quantifying the receptor copy numbers on these primary patient DCs (CD141+ cDCs and CD1c+ cDCs), we found that the abundance of PD-L1 [as determined by specific antibody (Ab) binding capacity (SABC)] is, on average, ~20-fold higher than B7.1 (Fig. 1B and table S1). B7.1 expression on tumor-associated cDCs is heterogeneous (Fig. 1C). Thus, we next gated on PD-L1+B7.1+ cDC and confirmed that PD-L1 is overexpressed relative to B7.1 on both CD141+ and CD1c+ cDC subsets (Fig. 1D). Therefore, under these conditions, the highly abundant PD-L1 may sequester B7.1 in cis on DCs, but given the proportionally higher expression of PD-L1, free PD-L1 should be available for PD-1 binding.
B7.1, PD-L1, and PD-1 surface expression on peripheral (from PBMCs) and tumor DCs from three patients with lung cancer determined by flow cytometry. Subsets were defined as conventional DC (cDC; CD45+HLA-DR+Lin1−CD123−CD11c+), CD1c+ CD141− (named CD1c+ cDC) or CD141+ CD1c− (named CD141+ cDC) and plasmacytoid DC (pDC; CD45+HLA-DR+Lin1−CD123+CD11c−). (A) Histogram from one donor representative of three donors; specific signal (color) compared to signal from isotype control (black). (B) Quantification of marker expression on cDC subset CD141+ (red) or CD1c+ (black) by measurement of the SABC. Each dot represents one donor (n = 3). (C) Representative staining of B7.1 and PD-L1 on peripheral and tumor cDC from the same patients. Representative of three patients. (D) Quantification of B7.1 and PD-L1 on cDC subset CD141+ (red) or CD1c+ (black) PD-L1/B7.1 double-positive population with the Ab binding capacity. Each dot represents one donor (n = 2).
Blocking PD-L1 on DCs dissociates B7.1/PD-L1 in cis interaction to enhance CD28 costimulation on T cells
To characterize the functional consequence of PD-L1/B7.1 cis interaction, we turned to an in vitro system using monocyte-derived DCs (mono-DCs) from healthy donors. Immature mono-DCs (iDCs) express low PD-1, PD-L1, and B7.1. Upon maturation by Toll-like receptor 4 (TLR4) engagement [lipopolysaccharide (LPS)], PD-1 expression is quickly down-regulated, whereas both PD-L1 and B7.1 are up-regulated and colocalized on the cell surface (Fig. 2A and fig. S4). Quantification of PD-L1 and B7.1 expression on both immature and mature mono-DCs confirmed the same pattern as observed on the tumor-associated DCs: PD-L1 is overexpressed relative to B7.1 (Fig. 2B). Given that the PD-L1/B7.1 interaction has a threefold higher affinity than the B7.1/CD28 interaction (12), we wondered whether B7.1 is sequestered by the more abundantly expressed PD-L1 and therefore less capable of costimulating CD28 on the T cells (24).
(A) Expression and colocalization of B7.1, PD-L1, and PD-1 on monocyte-derived iDCs and mDCs derived from healthy peripheral blood mononuclear cells (PBMCs) by confocal imaging analysis. Histogram of intensities was calculated from the raw fluorescence intensities along the diagonal line. Scale bars, 5 μm. Images shown are representative of more than 10 images taken from two independent experiments. (B) Quantification of PD-L1 and B7.1 on iDC and mDC, with the Ab binding capacity determined by flow cytometry on eight unrelated donors. Each dot represents one donor (n = 8). (C and D) Left: Expression of (C) PD-1/PD-L1/CD28 PD-1 or (D) B7.1/PD-L1/CD28 by confocal imaging analysis in immunological synapse at the intersection of mDC with unmatched T cells after 6 hours of coculture. Images shown are representative of more than 10 images taken from two independent experiments. Right: Pearson’s correlation coefficients of (C) PD-1 on T cell and PD-L1 on DC or CD28 on T cell (n = 13) or (D) B7.1 on DC with PD-L1 on DC or CD28 on T cell (n = 18) colocalization for multiple images. Each dot represents one DC:T synapse. ****P < 0.0001, unpaired Mann-Whitney U test.
During the formation of the DC–T cell immunological synapse, CD28 and PD-1 are both polarized to the interface of mature DCs (mDCs) and T cells, as shown by an overlapping intensity profile and positive Pearson correlation coefficient. However, PD-L1/PD-1 colocalization is less frequent than colocalization of PD-1/CD28 (Fig. 2C and fig. S5). This suggests that PD-1 and CD28 are participating in T cell receptor (TCR) signaling within the immunological synapse, which is consistent with the recent finding that PD-1 signaling leads to dephosphorylation of CD28 on T cells (18, 19). However, unexpectedly, B7.1 on mDCs had little to no interaction with CD28 in the synapse on T cells, in sharp contrast to a much higher degree of cis interaction of PD-L1/B7.1 on DCs (Fig. 2D and fig. S6). This is likely due to the higher binding affinity of PD-L1/B7.1 versus that of B7.1/CD28 (12).
We next tested whether disrupting PD-L1/B7.1 cis interaction by PD-L1 blocking monoclonal Ab (mAb) could promote the release of B7.1, making it available to ligate CD28. To model physical interactions of PD-L1/B7.1 and B7.1/CD28, we created a Tag-Lite assay system of ligands and receptors, an alternative method to fluorescence resonance energy transfer (FRET). We first confirmed that PD-L1–expressing cells are bound by the anti–PD-L1 mAb (clone 6E11) (fig. S7) with a ~700-fold higher affinity [dissociation constant (KD) = 0.38 nM] than B7.1-Fc (KD = 54.74 nM) (Fig. 3A), suggesting that anti–PD-L1 mAb should easily displace the binding of B7.1 to cell surface PD-L1. Anti–PD-L1 mAb blocks B7.1 binding to PD-L1 more efficiently than B7.1-Fc (Fig. 3B). We next engineered cells to express both B7.1 and PD-L1, mimicking the DC surface pattern with PD-L1 about fourfold more abundant than B7.1 (fig. S8) and added CD28-Fc to mimic T cells in this setting. We then tested our hypothesis that there would be increased B7.1/CD28 interaction upon anti–PD-L1 mAb binding to PD-L1. In a dose-dependent manner, CD28-Fc showed strong interaction with B7.1 on the cells, and as we hypothesized, treatment with anti–PD-L1 mAb enhanced the binding signal between B7.1 and CD28-Fc (Fig. 3C). We confirmed this mechanism in a primary cell-cell system, where we found that preincubating mDCs with the anti–PD-L1 mAb enabled significantly stronger interaction of B7.1 on DCs and CD28 on T cells (P = 0.0003, Fig. 3D; fig. S9 and movies S1 and S2). Together, these data indicate that PD-L1 blocking Ab can disrupt PD-L1/B7.1 cis interaction to release B7.1 from sequestration by PD-L1, thereby enhancing B7.1/CD28 interactions.
(A) Binding of B7.1-Fc-A647 or anti–PD-L1-A647 (clone 6E11) to HEK cells expressing PD-L1–Tb by Tag-Lite assay. (B) HEK cells expressing PD-L1–Tb were co-incubated with B7.1-Fc-A647 in the presence of an increasing concentration of a competitor: B7.1-Fc, anti–PD-L1 6E11, or a nonbinding Ab (control). (C) HEK cells coexpressing PD-L1 and B7.1-Tb in cis were co-incubated with increasing concentration of CD28-Fc-A647. Anti–PD-L1 Ab (100 nM, clone 6E11) or a control Ab was added to block the interaction of B7.1/PD-L1, therefore freeing B7.1 to bind to CD28. (A to C) Representative results from three independent experiments. (D) mDCs preincubated with or without anti–PD-L1 mAb for 1 hour at 37°C, washed, and then cocultured for 6 hours with allogeneic T cells to allow immunological synapse formation. Expression of B7.1 (on DCs) and CD28 (on T cells) is shown. Histogram of intensities was calculated from the raw fluorescence intensities along the diagonal line. Scale bars, 5 μm. Right: Pearson’s correlation coefficients of B7.1 and CD28 colocalization for multiple images (n = 19 for control and n = 20 for anti–PD-L1 group). Each dot represents one DC:T synapse. Images shown are representative of more than 15 images taken from two independent experiments. ***P = 0.0003, unpaired Mann-Whitney U test.
To measure the functional consequences of PD-L1/B7.1 disruption with an anti–PD-L1 mAb, we evaluated CD28 downstream signaling in Jurkat reporter cells driven by the nuclear factor of activated T cells (NFAT) promoter. In an attempt to gauge the relative contribution of PD-L1/B7.1, in addition to PD-L1/PD-1 on CD28 signaling, we first incubated Jurkat cells with PD-1 blocking Ab to abrogate the PD-L1/PD-1 trans binding, and followed by specifically disrupting PD-L1/B7.1 binding by incubating mDC with a PD-L1 blocking mAb (Fig. 4A). Blocking PD-1/PD-L1 trans-interaction using a PD-1 blocking mAb increased CD28 signaling by 32% as compared to control; blocking PD-L1 Ab on mDCs further enhanced CD28 signaling from 32 to 52% (Fig. 4B and table S2). Thus, by disrupting the PD-L1/B7.1 cis interaction with an anti–PD-L1 mAb, CD28 signaling is enhanced significantly, over and above the magnitude induced by a PD-1 blocking mAb. To further investigate the exclusive contribution of B7.1 to CD28 costimulation, we included anti–PD-L1 mAb (clone MIH3), which does not substantially block PD-L1/B7.1 interaction (25). CD28 signal induction is significantly lower in MIH3 mAb-treated DCs as compared to the 6E11 clone (Fig. 4C). However, it is important to note that, in this setting, the MIH3 clone does block PD-L1/B7.1 interaction but at a much lower extent than the 6E11 clone (Fig. 4D). These data provide evidence that anti–PD-L1 mAb disrupts PD-L1/B7.1 cis interaction on DCs and allows enhanced CD28 downstream signaling on T cells upon priming.
(A) Drawing of the different hypothetical effects on CD28 signaling when Jurkat CD28-CD3z incubated or not with anti–PD-1 are cocultured with mDC incubated or not with anti–PD-L1 clone 6E11. (B) NFAT signal measured from the experiment described in (A) after 6 hours of incubation. Single point represents average signal, with SD derived from technical duplicates or triplicates from every DC donor (n = 6). (C) NFAT signal measured when incubating Jurkat CD28-CD3z with mDC incubated or not with anti–PD-L1 clone 6E11 or MIH3. Single point represents average signal, with SD derived from technical duplicates or triplicates from every DC donor (n = 8). (D) HEK cells expressing PD-L1–Tb were co-incubated with B7.1-Fc-A647 in the presence of an increasing concentration of a competitor: B7.1-Fc, anti-PD-L1 clone 6E11 or MIH3, or a nonbinding Ab (Control). Single point represents average signal derived from technical duplicates. Matched-pairs t test performed on log-transformed luminescence values for all pairs of donors. Not adjusted for multiple testing: *P < 0.05; **P < 0.01; ***P < 0.001.
PD-L1 blockade–treated DCs induce T cell priming and proliferation
We next sought to study the functional consequences of PD-L1 blockade on DCs. Pretreatment with an anti–PD-L1 mAb directly activated human iDCs to stimulate proliferation of allogeneic T cells (Fig. 5A). The next question is whether PD-L1 inhibition would arm DCs with an enhanced capacity to prime naïve T cells and generate a de novo immune response. For these studies, we used mouse splenic CD11c+ DCs cocultured with ovalbumin antigen (DQ-OVA) to cross-present the SIINFEKL peptide to naïve CD8+ T cells from OT-I mice, which are specific for SIINFEKL in the context of H-2Kb. Pretreatment of DCs with an anti–PD-L1 mAbs enhanced their capacity to present processed DQ-OVA, resulting in enhanced proliferation and production of granzyme B by CD8 T cells (Fig. 5, B and C). Furthermore, DCs treated with anti–PD-L1 Ab induced higher T cell effector potential evidenced by increased T-bet and Eomes (fig. S10). Similarly, an increase of T cell priming after PD-L1 blockade was observed with SIINFEKL peptide-pulsed DCs (fig. S11).
(A) T cell proliferation induced by allogeneic DCs preincubated with or without anti–PD-L1 mAb. Single points represent average proliferation derived from technical duplicates or triplicates from every donor (n = 5). **P < 0.01, matched-pairs t test. (B) T cell priming by mouse splenic DCs pretreated with or without anti–PD-L1 mAb in the presence of DQ-OVA to cross-present to naïve OT-I cells. Granzyme B (GrzB) production per proliferation peak on T cells was shown in (C) for DC:T ratios of 1:3 and 1:10. Data are representative of three independent experiments, performed in duplicate or triplicate culture. A two-way ANOVA analysis was performed for the statistical significance. **P < 0.01.
DCs are critical targets of PD-L1 blockade treatment for patient with cancer
Having established the biological basis of PD-L1 blockade on DCs that involves PD-L1/B7.1 cis interaction and a downstream CD28 costimulation, we next evaluated the contribution of tumor-associated DCs in response to PD-L1 blockade in the clinical setting. For this purpose, we sought for a DC gene signature in pretreatment tumor specimens from patients with RCC enrolled into a phase 1 study (NCT01375842) of atezolizumab (anti–PD-L1) (n = 56) (4) and patients with non–small cell lung cancer (NSCLC) enrolled into the randomized phase 2 study POPLAR (NCT01903993) of atezolizumab versus docetaxel (n = 188) (2) based on RNA-sequencing (RNA-seq) gene expression data. To weigh the abundance of tumor-associated DCs, we developed an algorithm of cumulative expression score based on genes that define major DC subsets (XCR1, BATF3, IRF8, and FLT3). Patients with RCC with a high DC signature at baseline showed improved overall survival (OS) in response to atezolizumab [median OS not reached; hazard ratio (HR), 0.38; 95% confidence interval (CI), 0.16 to 0.91] compared to patients with low DC signature (median OS, 16.6 months) (Fig. 6A and table S3). To evaluate the specificity of PD-L1 blockade and this beneficial DC signature, we analyzed a cohort of patients with NSCLC treated with either atezolizumab or docetaxel chemotherapy. Similar observations were made in atezolizumab-treated patients with NSCLC: The group of patients with a high DC signature showed an approximate 8-month OS benefit compared to the low DC signature group (HR, 0.54; 95% CI, 0.31 to 0.96). In contrast, there was no significant survival difference between high and low DC signature groups in patients treated with docetaxel (HR, 0.82; 95% CI, 0.50 to 1.34). Furthermore, in high DC signature patients, atezolizumab demonstrated a survival benefit versus docetaxel (HR, 0.56; P = 0.05), highlighting the importance of DC signature to support decision-making on the treatment option (Fig. 6B and table S4). A differential benefit was also observed in patients expressing PD-L1 protein on ≥5% of TCs or tumor-infiltrating ICs [i.e., patients with PD-L1 immunohistochemistry (IC or TC) status 2 or 3]: The patients with high DC signature have not reached the median OS (HR, 0.25; 95% CI, 0.07 to 0.88), whereas low DC signature patients have a median OS of 8.4 months (fig. S12 and table S5). Although PD-L1 gene expression is not a good predictive marker for response to atezolizumab (fig. S13), the clinical benefit of atezolizumab in high DC signature patients was only found in those whose ICs are positive for PD-L1 (IC 1/2/3; HR, 0.43; P = 0.028) (Fig. 6D), but not in PD-L1− patients (IC 0; HR, 0.91; P = 0.85) (Fig. 6C). Together, these clinical data indicate that the presence of tumor-associated PD-L1+ DCs is associated with improved clinical benefit from PD-L1 blockade with atezolizumab. This may, in part, explain why PD-L1− patients benefit from PD-L1 blockade (26), as DC presence does not correlate with PD-L1 levels (fig. S14).
RNA-seq was performed on pretreatment tumor tissue from RCC and NSCLC patients with cancer before treatment in studies NCT0137584 and NCT01903993, respectively. A cumulative score for DC signature (composed of genes including XCR1, BATF3, IRF8, and FLT3) was used to divide the patients into two groups (low DC signature versus high DC signature). (A) Kaplan-Meier estimate of OS of patients with RCC treated with atezolizumab (n = 56). (B) Kaplan-Meier estimate of OS of patients with NSCLC treated with atezolizumab (n = 92) or docetaxel (n = 96). (A and B) HR refers to the high DC signature versus low DC signature group. Kaplan-Meier estimates in DC gene signature subgroups in IC PD-L1− (C) and IC PD-L1+ patients (IC1/2/3) (D). HR refers to the high DC signature in atezolizumab- versus docetaxel-treated patients.
DISCUSSION
We report the importance of tumor-associated DCs in explaining clinical response to immunotherapy blocking the PD-L1/PD-1 pathway. In this study, we introduce a mechanism for how PD-L1 checkpoint expression on DCs impairs their costimulatory function with respect to T cell priming and/or restimulation, or both. The described mechanism suggests that DCs are primary cellular targets of anti–PD-L1 Abs through disruption of the PD-L1/B7.1 cis interaction. This allows CD28 costimulation by increased B7.1/CD28 interaction and subsequent de novo antitumor T cell immunity. Our study provides mechanistic understanding of the role of DCs in response to a checkpoint inhibitor such as anti–PD-L1 Ab. It also explains why in animal studies cross-presenting DCs are required for antitumor efficacy of anti–PD-L1/anti–PD-1 Abs (27–29). It is possible that PD-L1/B7.1 cis interaction on DCs might prevent PD-L1 ligation to PD-1 on T cells (16), depending on the relative protein abundance of PD-L1 versus B7.1. We observed in patient samples that PD-L1 is much more abundantly expressed relative to B7.1 on iDCs and mDCs in both peripheral and tumor tissues, and that functionally B7.1 is sequestered by PD-L1 in cis, but not vice versa. The correlation of a DC gene signature and clinical response in two distinct clinical cohorts further supports the contribution of tumor-associated DCs to anti–PD-L1 therapies in patients with cancer. We propose that anti–PD-L1 Abs may mediate effects that are mechanistically different from anti–PD-1 Abs by having a dual action: (i) disrupting PD-L1/PD-1 interaction in trans between T cell and tumor or IC, thus removing PD-1–mediated inhibitory signaling on T cells, and (ii) disrupting the PD-L1/B7.1 cis interaction on DCs, thereby freeing up B7.1 that would be sequestered by PD-L1 to bind to CD28 (fig. S15).
In our study, we do not exclude the possibility that freeing up B7.1 might also lead to enhanced competition from CTLA-4. A recent study suggested that PD-L1 blockade disrupts PD-L1/B7.1 cis interaction and releases B7.1 for stronger binding to CTLA-4 on T cells (17). Advancing our understanding of the complex network between B7 molecules and their competing partners CTLA-4, PD-L1, and PD-1 will further help us revisit the modes of action of current immunotherapies. We feel that the importance of this additional activation via PD-L1/B7.1 disruption in the treatment of human cancer may be both tumor/immune context dependent and treatment time dependent. We did not investigate the differential effect of anti–PD-L1 Ab versus anti–PD-1 Ab. A recent study indicates that targeted PD-1 depletion on myeloid cells leads to enhanced antitumor immunity (22), suggesting that anti–PD-1 Ab may directly affect functionality of myeloid cells, including DCs. It remains to be further investigated the differential role of PD-L1 versus PD-1 in patients with little to no evidence of preexisting immunity but a prominent DC presence within their tumors. The correlation of tumor-infiltrating DC with the clinical response from PD-L1 blockade may prove to have significant implications for future clinical development of anti–PD-L1/PD-1 therapies and combination immunotherapy.
MATERIALS AND METHODS
Study design
This study was designed to characterize the role of DCs in response to PD-L1 blockade. We hypothesized that PD-L1, being coexpressed with its two receptors PD-1 and B7.1 on DCs, might regulate their functions. In this study, we used an in vitro system to evaluate how an anti–PD-L1 Ab disrupts the PD-L1/B7.1 cis interaction on DCs and their downstream T cell activation signal. To translate the in vitro findings into the human setting, we integrated clinical and transcriptomic data in patients with cancer treated with atezolizumab.
The sample size was determined by previous experience and preliminary experiments; no statistical method was used. Sample sizes are denoted in figures or figure legends and refer to number of animals or human donors unless stated otherwise. In vitro assays using human primary cells were repeated independently using cells derived from at least five unrelated donors. All studies were performed in technical duplicates or triplicates as indicated. All samples that met proper experimental conditions such as appropriate mono-DC phenotype and proper positive and negative delta controls were included in the analysis. Results shown are either representative of or mean of at least three independent experiments (otherwise annotated in the legends). For confocal imaging, the image acquisition and analysis was performed blindly. Primary data are reported in data file S1.
Lung cancer samples
Lung cancer samples and corresponding patient blood were derived from the private hospital Hirslanden, Zürich, Switzerland. All human biological samples were collected after ethical approval of the studies by the local ethics committee (Kantonale Ethikkommission Zürich) and after written informed consent of the patients was obtained and in line with Good Clinical Practice (GCP) guidelines and the Declaration of Helsinki.
Lung cancer tumor samples were mechanically dissociated and digested using accutase (PAA Laboratories or Sigma-Aldrich), collagenase IV (Worthington), hyaluronidase (Sigma-Aldrich), and deoxyribonuclease (DNase) type IV (Sigma-Aldrich) within several hours after excision. After 20 to 30 min of digestion at 37°C on a rotator, tumor-derived cells were carefully mashed through a 70-μm cell strainer and washed two times with cold phosphate-buffered saline (PBS; centrifugation, 250g, 10 min, 4°C). Red blood cell lysis was performed before cells were slow-frozen and stored in vapor phase until thawed for subsequent fluorescence-activated cell sorting (FACS) analysis.
PBMCs were isolated from patient-derived blood using density gradient centrifugation (450g, 30 min at room temperature without brake) over Histopaque-1077 (Sigma-Aldrich). Human PBMCs were isolated from the interphase and washed several times with Dulbecco’s PBS (DPBS). After red blood cell lysis, PBMCs were stored in vapor phase [90% fetal bovine serum (FBS) supplied with 10% dimethyl sulfoxide (DMSO)] until thawed for subsequent FACS analysis.
Human in vitro DCs
For in vitro generation of DCs, PBMCs were isolated using Ficoll gradient (GE Healthcare) from buffy coats processed by Blutspende, and red blood cells were lysed using BD Pharm Lyse (BD Bioscience). Monocytes were enriched using the Human Monocyte Enrichment Kit (Stem Cell Technologies) and cultured in RPMI 1640 GlutaMAX medium (Gibco) supplemented with 10% heat-inactivated FBS (Gibco) and 1% penicillin-streptomycin 100× (Gibco) for 5 days in the presence of granulocyte-macrophage colony-stimulating factor (GM-CSF) (10 μg/ml) and interleukin-4 (IL-4) (R&D Systems). The cytokines were added every 2 days. To induce maturation, harvested DCs were cultured in the presence of LPS from Escherichia coli O55:B5 (Sigma-Aldrich) at 10 ng/ml for 24 hours.
Flow cytometry and Abs
TCs and PBMCs from patients with lung cancer were thawed in FBS and incubated for 10 min with DNase type IV (Sigma-Aldrich, #D5025) before being counted (Cellometer Auto 2000, Nexcelom Bioscience). Cells were first stained with Live blue dye (Life Technologies) diluted at 1:800 in PBS or Zombie Aqua Fixable Viability Dye (BioLegend) diluted at 1:400 in PBS (30 min at 4°C). To reduce nonspecific binding, human cells were next blocked with normal mouse immunoglobulin G (IgG; Life Technologies; 60 μg/ml) in Flow Cytometry Staining Buffer Solution (eBioscience), and human monocyte–derived DCs were additionally incubated with Human TruStain FcX and True-Stain Monocyte Blocker at the recommended concentration (BioLegend) for 10 min at 4°C. Cells were subsequently incubated with various combinations of Abs against DC or T cell markers (30 min at 4°C). For intracellular staining, cells were permeabilized with FACS permeabilization solution (BD Biosciences or eBioscience) and incubated with Abs against various intracellular molecules for 30 min at 4°C. Flow cytometry acquisition was performed on a custom-designed BD Biosciences Fortessa and analyzed using FlowJo software (Tree Star). Lung cancer PBMCs and TCs were fixed using Cytofix (BD Biosciences) before acquisition.
The following Abs were used: (i) anti-human Abs: CD1a (clone HI149, BioLegend), HLA-DR (clone TU36, Invitrogen or clone G45-6, BD Biosciences), CD14 (clone HCD14, BioLegend), CD209 (clone DCN46, BD Biosciences), PD-1 (clone EH12.2H7, BioLegend), PD-L1 (clone 29E.2A3, BioLegend), B7.1 (clone L307.4, BD Biosciences), CD3 (clone HIT3a, BioLegend), CD8 (clone SK1, BioLegend) and CD4 (clone OKT4, BioLegend), CD11c (clone B-ly6, BD Biosciences), CD141 (clone 1A4, BD Biosciences), Lin1 (targeting CD3, CD14, CD16, CD19, CD20, and CD56, BD Biosciences), CD1c (clone F10/21A3, BD Biosciences), CD123 (clone 9F5, BD Biosciences), and CD45 (clone 2D1, BioLegend); (ii) anti-mouse Abs: CD11b (clone M1/70, BioLegend), CD11c (clone N418, BioLegend), PD-L1 (clone 10F.9G2, BioLegend), F4/80 (clone BM8, BioLegend), PD-1 (clone 29F.1A12, BioLegend), CD19 (clone 6D5, BioLegend), TCRb (clone H57-597, BioLegend), CD45 (clone 30F11, BioLegend), MHCII (clone M5/114.15.2, BioLegend), CD8b.2 (clone 53-5.8, BioLegend), granzyme B (clone NGZB, eBioscience), T-bet (clone 4B10, BioLegend), and Eomes (clone Dan11mag, eBioscience).
Quantification of B7.1, PD-L1, and PD-1 on tumor and peripheral DCs from lung cancer samples was performed using the Quantum Simply Cellular anti-mouse IgG technology (Bang Laboratories). Directly labeled B7.1, PD-L1, and PD-1 Abs or their respective isotype controls were used at their estimated saturated concentration. The abovementioned CD45, Lin1, HLA-DR, CD11c, CD123, CD1c, and CD141 human Abs were used to gate on the DC subsets of interest (fig. S1). Quantification of B7.1, PD-L1, and PD-1 on mono-DC was performed separately using the QIFIKIT technology (Dako) with unlabeled B7.1, PD-L1, and PD-1 Abs (clones mentioned above) or their isotype control and following the manufacturer’s instructions.
Confocal imaging analysis
Mono-DCs were first labeled or not for 15 min at 37°C with CellTracker Blue CMAC Dye (Thermo Fisher Scientific) and washed, and 7.5 × 103 cells were seeded per well in eight-well glass bottom slides (Ibidi) previously treated with T100A RetroNectin Recombinant Human Fibronectin Fragment (5 μg/cm2; Takara Bio) and blocked with PBS supplemented with 2% of heat-inactivated FBS (Gibco). Alternatively, they were seeded in 24-well plates (TPP) with sterile round glass slide on the bottom incubated with polylysine (Sigma-Aldrich) for 20 min at 37°C and washed with medium. Next, the cells were stimulated or not with LPS (10 ng/ml; Sigma-Aldrich) for 24 or 48 hours to generate mDC. mDCs were incubated or not with anti–PD-L1 blocking mAb (clone 6E11, murine IgG1, Roche; 10 μg/ml) in medium for 1 hour at 37°C and washed. Allogeneic CD3+ T cells isolated from PBMCs using the Human Pan T Cell Isolation Kit (Miltenyi Biotec) were added to the DCs at a DC:T cell ratio of 1:5. After 6 hours of incubation, to allow formation of DC:T cell conjugates, the cells were first blocked by normal IgG mouse (120 μg/ml; Invitrogen), normal IgG rat (60 μg/ml; Invitrogen), and Fc blocker (10 μg/ml; BD Biosciences) for 10 min at 4°C in flow cytometry staining buffer (eBioscience). Cells were next stained for surface marker PD-1 (clone EH12.2H7, BioLegend), PD-L1 (clone 29E.2A3, BioLegend), CD28 (clone CD28.2, BioLegend), or B7.1 (clone 2D10, BioLegend or clone L307.4, BD Biosciences) diluted at 1:20 in blocking solution, washed, and fixed using BD Cytofix fixation buffer (BD Biosciences). The staining was acquired with confocal laser scanning microscope (LSM 700, Carl Zeiss). Digital images (1024 × 1024 pixels) were captured with 405-, 488-, 555-, and 639-nm laser excitation. Images were acquired with a 40×/1.3× oil-immersion objective. Images from all groups were captured with identical photomultiplier gain settings and processed with image analysis software IMARIS 8.4.1 (Bitplane). Histogram plots were produced using Fiji software (Imaging Processing and Analysis in Java). Ab blockade and staining were performed by M.M., and image acquisition and analysis was performed blindly by S.Chen as samples’ associated specific ID did not contain experimental information.
Tag-Lite assay
The interaction between PD-L1, B7.1, and CD28 molecules was studied using a time-resolved FRET assay (referred to as Tag-Lite). Human embryonic kidney (HEK) Epstein-Barr virus nuclear antigen (EBNA) cells [American Type Culture Collection (ATCC)] were transfected with complementary DNA (cDNA) encoding for human PD-L1, CD28, and B7.1 fused to a SNAP (soluble N-ethylmaleimide–sensitive factor attachment protein) tag using Lipofectamine 3000 (Invitrogen). Briefly, cells were transfected in T75 flasks using 5 μg of DNA for single transfections and 5 μg of each for cotransfections. After 16-hour incubation, cells were washed with 5 ml of DPBS and labeled with SNAP-Lumi4-Tb (Terbium) (SSNPTBG, Cisbio). Labeling efficiency was determined by fluorescence measurement at 615 nm (excitation, 343 nm; Victor3, Perkin Elmer) of 10,000 cells per well in a 384-well plate (Greiner). Cells were then frozen in culture medium supplemented with 10% DMSO (Sigma) and stored at −80°C.
Anti–PD-L1 (clone 6E11, huIgG1, Roche), CD28-Fc, and B7.1-Fc (R&D Systems) were labeled with Alexa Fluor 647 NHS Ester (A647, Invitrogen) according to the manufacturer’s instruction.
After thawing, cells were washed once in PBS before being resuspended at 1 Mio/ml of Tag-Lite labeling medium (LabMED; Cisbio). The incubation of 10,000 cells with a gradient or fixed dose of Ab-A647 and with or without anti–PD-L1 blocking Ab (clone 6E11, huIgG1, Roche or clone MIH3, BioLegend), B7.1-Fc (R&D Systems), or a control Ab [anti–endothelial growth factor receptor (EGFR) Ab, huIgG1, Roche] at different concentrations was performed in a 384-well plate in 20-μl final volume of Tag-Lite labeling medium. Measurement of the emission at 620 and 665 nm after excitation at 340 nm was done (Infinite M1000Pro, Tecan) after 0-, 1-, 2-, and 4-hour incubation at room temperature and after overnight incubation at 4°C. For each well, the ratio of the measurement 665/620 nm * 10,000 (R) was calculated. The delta ratio (ΔR) is obtained by subtracting the background ratio (cells only). For KD determination, the results were analyzed in GraphPad Prism6. The assay was performed in duplicate, and the mean and SD are displayed.
Jurkat CD28-CD3z assays
Jurkat reporter cells expressing a chimeric receptor fusion of CD28 and CD3z (see below for details regarding the coding sequence) were generated by transducing Jurkat cells harboring an integrated NFAT-luciferase response element (“JNL,” Signosis) with the respective lentiviral virus-like particles (VLPs), coding for the chimera. VLPs were produced following Lipofectamine LTX-based transfection of HEK293T cells (ATCC CRL3216) grown to 60 to 70% confluency with a mixture of transfer plasmids and pCMV-VSV-G:pRSV-REV:pCgpV transfer plasmids at a 3:1:1:1 molar ratio. Sixty-six hours after transfection, supernatants were collected, centrifuged for 5 min at 250g to remove cell debris, and filtered through a 0.45- or 0.22-μm polyethersulfon filter. VLP-containing supernatants were used directly to transduce JNL cells via spinfection for 90 min (32°C, 1000g). Transduced cells were enriched as pools under puromycin selection (1 μg/ml), applying selection pressure on the third day after spinfection.
Fasta-format of the amino acid sequence of the CD28_CD3z chimeric receptor fusion used for the transduction of Jurkat cells: >CD28_CD3z
MGWSCIILFLVATATGVHSNKILVKQSPMLVAYDNAVNLSCKYSYNLFSREFRASLHKGLDSAVEVCVVYGNYSQQLQVYSKTGFNCDGKLGNESVTFYLQNLYVNQTDIYFCKIEVMYPPPYLDNEKSNGTIIHVKGGGGSFWVLVVVGGVLACYSLLVTVAFIIFWVRSKRSRLLHSDYMNMTPRRPGPTRKHYQPYAPPRDFAAYRSRVKFSRSADAPAYQQGQNQLYNELNLGRREEYDVLDKRRGRDPEMGGKPRRKNPQEGLYNELQKDKMAEAYSEIGMKGERRRGKGHDGLYQGLSTATKDTYDALHMQALPPR.
Fifty thousand Jurkat CD28-CD3z cells, incubated or not for 1 hour at 37°C with anti–PD-1 (clone 314, human IgG1, Roche) and washed, were seeded with mDC, incubated or not for 1 hour at 37°C with anti–PD-L1 (6E11 Roche Ab or MIH3 BioLegend AB) and washed, at a Jurkat:mDC ratio of 1:5 in a 96-well plate U bottom (TPP) and incubated at 37°C for 6 hours in 200-μl final volume. One hundred microliters of supernatant was removed and replaced by 100 μl of ONE-Glo reagent (Promega) and transferred after 5-min incubation at room temperature in a white 96-well plate for readout of luminescence (Perkin Elmer).
Human T cell proliferation
Human mono-DCs were cultured with or without anti–PD-L1 blocking mAb (10 μg/ml; clone 6E11, murine IgG1, Roche) for 24 hours, followed by intensive wash before coculture at different ratios with 150 × 103 allogeneic CD4+ T cells isolated from PBMCs of an unrelated donor using a human CD4+ T cell isolation kit (Miltenyi Biotec). CD4 T cells were prelabeled with carboxyfluorescein succinimidyl ester (CFSE) proliferation dye (eBioscience) according to the manufacturer’s instructions (0.5 μM at 5 × 106 cells/ml for 7 min at 37°C). After 5 days of coculture, phenotype and proliferation (dilution of CFSE) were evaluated by flow cytometry.
Cross-priming of T cells
Spleens from C57BL/6 mice (Charles River) were digested in RPMI medium supplemented with 10% FBS, 1% penicillin/streptomycin (all from Gibco–Thermo Fisher Scientific), DNase (0.05 mg/ml), and collagenase D (1 mg/ml; both from Sigma-Aldrich) for 30 min at 37°C. After incubation, EDTA (Sigma-Aldrich; 0.5 M) was added and the cells were washed with PBS (Gibco–Thermo Fisher Scientific). Red blood cells were lysed with red blood cell lysis buffer (BD Biosciences), and CD11c+ DCs were isolated using the CD11c MicroBeads UltraPure Isolation Kit (Miltenyi Biotec) according to the manufacturer’s instructions. After isolation, DCs were either incubated with DQ-OVA (Thermo Fisher Scientific; 100 μg/ml) or pulsed with OVA peptide 257–264 SIINFEKL (GenScript; 0.5 μg/ml) for 30 min at 37°C and washed with PBS. Next, DCs were incubated with PD-L1 Ab (generated in-house; 5 μg/ml) for 30 min at 4°C and used for priming of naïve CD8 T cells isolated from OT-I mice using the Naïve CD8a Isolation Kit (Miltenyi Biotec) and labeled with CTV (cell trace violet) proliferation dye (Thermo Fisher Scientific) according to the manufacturer’s instructions. After 72 hours, proliferation (dilution of CTV proliferation dye) and activation were analyzed by flow cytometry.
Patients and clinical studies
The clinical studies with atezolizumab were sponsored by Genentech Inc. In total, 56 patients with RCC (NCT01375842) were treated with atezolizumab (anti–PD-L1; MPDL3280A; F. Hoffmann–La Roche/Genentech). In the NSCLC trial (NCT01903993), there were a total of 287 patients treated. Among those, we analyzed all 188 patients who had RNA-seq on baseline tumor biopsies: 92 patients were treated with atezolizumab and 96 patients were treated with docetaxel. The study was done in full accordance with the guidelines for GCP and the Declaration of Helsinki. Protocol (and modification) approval was obtained from an independent ethics committee for each site.
The PD-L1 expression in patients with NSCLC was scored on TCs and tumor-infiltrating ICs with the VENTANA SP142 PD-L1 immunohistochemistry assay (Ventana Medical Systems, Tucson, AZ, USA), as previously described (6). PD-L1+ TCs were scored as a percentage of total TCs according to the following scale: TC3 ≥ 50%, TC2 ≥ 5% and TC2 < 50%, TC1 ≥ 1% and TC1 < 5%, and TC0 < 1%. Tumor-infiltrating ICs were scored into categories as percentage of tumor area according to the following scale: IC3 ≥ 10%, IC2 ≥ 5% and IC2 < 10%, IC1 ≥ 1% and IC1 < 5%, and IC0 < 1%. In this analysis, we marked IC 2/3 or TC 2/3 as the PD-L1+ NSCLC. We marked IC0 IC PD-L1− and IC 1/2/3 IC PD-L1+. Archival tumor biopsies from those patients at baseline were obtained for genome-wide expression profiling by RNA-seq (6). We investigated the impact of multiple genes involved in human DC development and function (XCR1, BATF3, FLT3, and IRF8) by defining a cumulative DC gene score (DC score), reflecting the cumulative expression of these marker genes. Each gene’s expression (on log scale) was first standardized by a z score
Statistical analysis
Data were analyzed with SAS JMP software version 13.0.0 (SAS Institute Inc.) or R statistical programming language version 3.3.0. All statistical estimates and tests were performed after proper log transformation of data, using a two-tailed matched-pairs t test (Figs. 4, B and C, and 5A). A two-way analysis of variance (ANOVA) was used to compare two different groups treated with different conditions (Fig. 5C). A nonparametric unpaired Mann-Whitney test was performed for the receptor colocalization analysis within DC:T cell synapses (Figs. 2, C and D, and 3D). Differences with P values (not adjusted for multiple testing) smaller than 0.05 were considered statistically significant.
For the clinical data analysis, we used the Kaplan-Meier method to estimate median OS and to draw survival curves for predefined subpopulations. We used stratified Cox regression model to estimate HRs and 95% CIs in different subpopulation (adjusting for same variables in all strata). P values lower than 0.05 were considered statistically significant. The RCC cohort was adjusted for the sex of patient and the stage at initial diagnosis. The NSCLC cohort was adjusted for the smoking status, Eastern Cooperative Oncology Group performance status, and sex of patient.
SUPPLEMENTARY MATERIALS
stm.sciencemag.org/cgi/content/full/12/534/eaav7431/DC1
Materials and Methods
Fig. S1. Gating strategy for ex vivo DC study from tumor and PBMC from patients with lung cancer and PBMC from healthy donors.
Fig. S2. PD-1, PD-L1, and B7.1 surface expression on peripheral DC subsets from healthy donors.
Fig. S3. B7.1 and PD-L1 are coexpressed on Clec9a+ DC from patients with RCC.
Fig. S4. Expression of PD-L1 and B7.1 by mono-DCs.
Fig. S5. PD-1 preferably colocalizes with CD28 in cis on T cells than with PD-L1 in trans within the immune synapses.
Fig. S6. B7.1 colocalizes preferably with PD-L1 in cis on DCs than with CD28 in trans within the synapses.
Fig. S7. Schematic diagram of anti–PD-L1 Ab clone 6E11.
Fig. S8. Quantification of PD-L1 and B7.1 on HEK cells.
Fig. S9. Anti–PD-L1–treated DCs enhances B7.1/CD28 colocalization within the DC:T cell synapses.
Fig. S10. Blocking PD-L1 on mouse DCs enhances their T cell priming capacity.
Fig. S11. PD-L1 blocking on DC enhances their T cell priming capacity.
Fig. S12. DC signature is associated with the improved survival benefit to atezolizumab in patients with NSCLC in a PD-L1–dependent manner.
Fig. S13. PD-L1 gene expression is not predictive for clinical benefit to atezolizumab.
Fig. S14. PD-L1 gene signature is not correlated with DC signature.
Fig. S15. Schematic drawing illustrating the modes of action of anti–PD-L1 mAb on DCs.
Table S1. Summary of surface expressions (specific antibody binding capacity) of B7.1, PD-1, and PD-L1 on tumor and peripheral DC subsets from patients with lung cancer.
Table S2. Relative contribution of PD-L1/B7.1 in cis interaction and PD-1/PD-L1 in trans interaction on CD28 signaling.
Table S3. Clinical response to atezolizumab in patients with RCC based on their DC gene signature.
Table S4. Clinical response to atezolizumab in patients with NSCLC based on their DC gene signature.
Table S5. Clinical response to atezolizumab in PD-L1+ (TC2/3 or IC2/3) patients with NSCLC based on their DC gene signature.
Movies S1 and S2. Three-dimensional movie construction of DC:T synapse of the confocal images from Fig. 3C.
Data file S1. Primary data.
This is an article distributed under the terms of the Science Journals Default License.
REFERENCES AND NOTES
- Copyright © 2020 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original U.S. Government Works
