Summary
PD-L1 is a ligand for immune checkpoint receptor PD1. Anti-PD-L1 antibody is an effective therapy for a variety of solid tumors, although a durable response is only achieved in a subset of patients. For unknown reasons, EGFR-mutant tumors respond poorly to checkpoint blockade. Applying quantitative cell biological methods to study PD-L1 biology in lung cancer cells, we establish that growth factors acutely regulate PD-L1 trafficking between the plasma membrane and the interior of cells. Changes in plasma membrane PD-L1 levels will impact PD1 engagement on T cells, thereby influencing PD-L1’s immune suppressive activity. To discover potential cell-intrinsic functions of PD-L1, we used APEX2 biotinylation to generate a high-resolution map of the PD-L1 proximal proteome. ESCRT pathway proteins were enriched in PD-L1’s proximal proteome, and two ESCRT-dependent functions, turnover of mutant EGFR and biogenesis of extracellular vesicles, were affected by anti-PD-L1 treatment, suggesting a link between PD-L1 and ESCRT function. Proteins that control cytoskeletal dynamics were also enriched in the PD-L1 proteome, and anti-PD-L1 treatment reduced cell migration, identifying migration as a PD-L1 associated function. PD-L1 knockout mimics the effects of the antibody treatment, suggesting anti-PD-L1 antibody effects are loss of function(s). The effects of anti-PD-L1 on the ESCRT- dependent functions and cell migration were restricted to cells harboring oncogenic EGFR mutations. Wildtype and KRAS mutant cells lines were unaffected. Our study reveals new cell-intrinsic roles for PD-L1 in EGFR mutant cells, activities that might contribute to the resistance of EGFR mutant tumors to PD-L1 checkpoint blockade.
Introduction
Therapies targeting the PD1/PD-L1 checkpoint axis are effective in many cancers, including non-small-cell lung carcinoma (NSCLC)1, 2. However, only a limited subset of patients shows durable response to anti-PD-L1 antibody (α-PD-L1) therapy. Expression of PD-L1 by cancer cells does not predict response to α-PD-L1 therapy, and some molecular subtypes, most notably EGFR mutant tumors, respond poorly to α-PD-L1 therapy, regardless of PD-L1 expression 3–5. Thus, despite the well-described role of PD-L1 as a PD1 ligand and the remarkable clinical successes of α-PD-L1 therapies, there is still much to be learned about PD-L1 biology, particularly its cell intrinsic functions and behavior, information that will provide a context for better understanding response and resistance to α-PD-L1 therapy.
PD-L1 is a single pass type I transmembrane protein widely expressed on immune cells as well as many non-lymphoid cells, including epithelial cells 6. Because PD-L1 functions to blunt immune response by binding PD1 on target T cells, regulation of the amount of PD-L1 on the plasma membrane is key for controlling its immune checkpoint activity. A number of transcriptional and post-translational regulatory mechanisms impacting PD-L1 expression have been identified. Numerous cytokines, cell stress stimuli and oncogenes stimulate PD-L1 transcription downstream of several signaling pathways 7–9. Several modes of post translational regulation have also been identified that impact PD-L1 protein turnover, including regulation of glycosylation, palmitoylation and phosphorylation 10–13, and proteins specifically involved in determining whether PD-L1 is degraded or not have also been identified (e.g. CMTM6 and HIP1R) 14–16.
Although PD-L1 as a ligand for PD1 has been the predominant focus of its role in cancers, there is mounting evidence for cell intrinsic functions of PD-L1 that might affect tumor progression independent of PD-L1 binding PD117, 18. Along these lines, PD-L1 has been implicated in the regulation of epithelial mesenchymal transition, resistance to chemotherapeutic agents as well as interferon-mediated cell toxicity17, 19–21. Some of the non-PD1 ligand activities of PD-L1 might be more reflective of PD-L1’s role outside of cancer. For example, in a non-cancer setting, PD-L1 expression by non-immune cells might have a major function in protecting healthy cells/tissue from collateral damage at sites of active immune response and these cell-intrinsic PD-L1 activities might be vital for cell survival and tissue healing.
Here we report on the interrogation of PD-L1 biology using quantitative cell biological methods. We report that growth factors, via AKT activation, acutely regulate the distribution of PD-L1 between the plasma membrane and the interior of cells. Growth factor regulation of PD-L1 trafficking is independent of previously described transcriptional upregulation by IFNγ and of the control of PD-L1 protein turnover by the CKFL-like MARVEL transmembrane protein 6 (CMTM6)14, 15. Thus, growth factor modulation of PD-L1 cycling to and from the plasma membrane provides a mechanism for rapid modulation of cell surface levels that augments previously described regulatory mechanisms that control the total amount of PD-L1. We used APEX2 biotinylation mapping to define the proteome in proximity to PD-L1. Proteins of the ESCRT pathway were enriched in the PD-L1 proximal proteome. αPD-L1 treatment of cells affected two ESCRT-dependent activities: biogenesis of extracellular vesicles (EVs) and turnover of mutant EGFR. Proteins that control cytoskeleton dynamics were also enriched in the PD-L1 proximal proteome and the cytoskeletal dependent activities of cell-migration were affected by α-PD-L1 antibody treatment. The effects of α-PD-L1 treatment were restricted to cells harboring oncogenic EGFR. PD-L1 knockout phenocopied the effects of α-PD-L1 treatment of EGFR mutant cells, establishing the effects of α-PD-L1 are due to a loss of PD-L1 function. These data reveal a heretofore undescribed interplay between constitutively active mutant EGFR and α-PD-L1 treatment. Thus, in EGFR mutant tumors, α-PD-L1 therapy not only disrupts PD-L1/PD1 interaction but potentially affects cancer cell migration as well as the broader tumor microenvironment downstream of changes in EVs that could impact response to the immune checkpoint therapy.
Results
Intracellular and plasma membrane PD-L1 are in rapid equilibrium regulated by growth factors
We used fluorescence microcopy to interrogate the subcellular distribution of PD-L1 in primary human NSCLC cells (Fig 1A). In addition to PD-L1 localized to the plasma membrane detected by staining intact cells, staining of permeabilized primary cells revealed a significant intracellular accumulation of PD-L1. Quantification of the fluorescence revealed that the majority of PD-L1 expressed in primary lung cancer cells was intracellular. Mutant KRAS or EGFR, the two most common driver oncogenes in lung cancer22, did not affect this PD-L1 distribution. (Fig 1B). Similarly, in established NSCLC cell lines and a non-transformed human lung epithelial cell line (BEAS-2B), there was a significant intracellular accumulation of PD-L1 that was predominantly concentrated in the peri-nuclear region of cells (Fig 1C). Quantification revealed a nearly equal distribution of PD-L1 between the plasma membrane and the interior of cells, independent of whether the cells were KRAS or EGF receptor (EGFR) mutants (Fig 1D).
A. Representative confocal images of primary human tumor cells immunostained for plasma membrane (PM) and total PD-L1. Red: PD-L1. Blue: Hoechst
B. Quantification of percentage of PM PD-L1 in primary human tumors. Each data point is a patient sample color coded for known oncogenes.
C. Representative confocal images of human lung cell lines with or without IFNγ treatment (40ng/mL, 24 hr) and immunostained for PM and total PD-L1. Red: PD-L1. Blue: Hoechst. Sum intensity of z-stacks is projected. Fluoresence intensity is equally scaled across panels for each cell line. Insets in -IFNγ PM are enhanced for better visualization.
D. Quantification of percentage of PM PD-L1 in human lung cell lines.
E. Total PD-L1 protein level in BEAS-2B cells treated with IFNγ (40ng/mL, 24 hr ). Each data point is an experiment. One sample t-test.
F. Quantification of PM localization of PD-L1 in BEAS-2B cells treated with IFNγ (40ng/mL, 24 hr). Each data point is an experiment. Unpaired student’s t-test.
G. Representative confocal images BEAS-2B cells stained for PM and total PD-L1 using clinical Durvalumab antibody. Red: PD-L1. Blue: Hoechst
All quantifications are calculated from average immunofluorescence staining intensity on epifluorescence images. All error bars = SEM. * ≤ 0.05, ** ≤ 0.01, *** ≤0.001
Interferon gamma (IFNγ), a potent transcriptional regulator of PD-L1, increased total PD-L1 protein levels in a non-transformed human lung epithelial line (BEAS-2B), an EGFR mutant (H1650) cell line and a KRAS mutant (A549) cell line by approximately 3 fold, without affecting either the distribution of PD-L1 between the interior and surface of cells or the punctate plasma membrane distribution of PD-L1 (Fig 1C, E & F). Immunostaining with Durvalumab, a therapeutic α-PD-L1 antibody23, 24, revealed a similar PD-L1 distribution between the plasma membrane and cell interior as the laboratory grade α-PD-L1 antibody (clone #130021, Novus Biologicals) (Fig 1G).
Plasma membrane localized PD-L1 mediates its immune suppressive function as a ligand for PD1 on target T cells, thereby raising the question of the biological function of the large intracellular accumulation of PD-L1. To explore this question, we first investigated the relationship between plasma membrane and intracellular PD-L1. We used a quantitative fluorescence microscopy assay adapted from our previous studies of membrane trafficking to interrogate the relationship between the plasma membrane and intracellular pools of PD-L125. If these pools are in equilibrium, then when live cells are incubated with a saturating concentration of an antibody that binds to the extracellular domain of PD-L1, the total amount of cell-associated α-PD-L1 antibody will increase as a function of incubation time. This is observed because the antibody-bound PD-L1 on the plasma membrane is internalized and replaced by recycling of intracellular PD-L1 that is not occupied by the antibody. The cell-associated α-PD-L1 plateaus when all the PD-L1 in equilibrium with the plasma membrane has been bound by antibody (that is, transited to the plasma membrane). Our analyses demonstrate that intracellular PD-L1 constitutively traffics to the cell surface with a halftime of approximately 10 min, establishing that PD-L1 plasma membrane levels are dynamically maintained by constitutive internalization and recycling back to the plasma membrane (Fig 2A). Furthermore, the maximum amount of cell-associated α-PD-L1 that is labeled by incubation with antibody in the medium (i.e., the plateau amount) is greater than 90% of the total amount of PD-L1 expressed in cells, demonstrating that greater than 90% of intracellular PD-L1 is in rapid equilibrium with the plasma membrane (Fig 2A). Together these data suggest one function of the intracellular PD-L1 pool is to serve as a reservoir for the plasma membrane pool of PD-L1.
A. PD-L1 steady state exocytosis in BEAS-2B cells. The line is a fit of the data to one-phase association exponential equation. Total PD-L1 stain normalized to exocytosis plateau is shown on the right. N=3
B. PD-L1 PM to total ratio with growth factors treatment or Akt inhibition in BEAS-2B cells. Cells were subjected to 4 hr of corresponding treatment prior to fixation and immunostain. Growth factors were added to serum-free media (EGF 8nM, IGF 10nM, insulin 10nM), MK2206 was added to serum complete media (MK2206 5μM). One-way ANOVA and post-hoc Tukey test.
C. Transferrin receptor (TfR) PM to total ratio in BEAS-2B cells. Cells were treated for 4 hrs. MK2206 was added to serum complete media (MK2206 5μM). One-way ANOVA.
D. PD-L1 PM to total ratio in cell lines treated with vehicle (DMSO) or Akt inhibitor for 4 hrs (MK2206 5μM). One sample t-test.
E. PD-L1 PM to total ratio in IFNγ (40 ng/mL, 24 hr) treated BEAS-2B cells incubated with vehicle (DMSO) or Akt inhibitor (MK2206 5μM) for 4 hrs. One sample t-test.
F. PD-L1 protein expression and distribution in BEAS-2B expressing doxycycline inducible CMTM6 shRNA (doxycycline 48 hr, 1 μg/mL; all cells in IFNγ 24 hr, 40 ng/mL). Left: Total PD-L1 protein. Right: PM localization of PD-L1. One sample and unpaired t-tests.
G. PD-L1 PM to total ratio in CMTM6 knocked-down (KD) BEAS-2B cells (doxycycline 48 hr, 1 μg/mL; all cells in IFNγ 24 hr, 40 ng/mL) treated with vehicle (DMSO) or Akt inhibitor (MK2206 5μM) for 4 hrs. One sample t-test.
All quantifications are calculated from average immunofluorescence staining intensity on epifluorescence images. Except panel A, each data point is an experiment. All error bars = SEM. * ≤ 0.05, ** ≤ 0.01, *** ≤0.001, # = 0.058
We next determined if the amount of PD-L1 in the plasma membrane can be acutely regulated. Incubation of cells in serum-free medium for 4 hrs resulted in a near 50% depletion of plasma membrane PD-L1 (Fig 2B). This depletion of plasma membrane PD-L1 reflected a redistribution of PD-L1 from the cell surface to intracellular compartments without a significant reduction in total PD-L1 protein (Fig S1A). Replacing the serum with different growth factors (EGF, IGF and insulin) restored plasma membrane PD-L1 levels without affecting the total amount of PD-L1, establishing that growth factors acutely control PD-L1 trafficking to and/or from the plasma membrane and consequently the distribution of PD-L1 between the plasma membrane and the interior of cells (Fig 2B). There were, however, no differences among the 3 growth factors in promoting PD-L1 accumulation on the plasma membrane.
A. Total PD-L1 protein expression in BEAS-2B cells. Cells were subjected to 4 hr of corresponding treatment prior to fixation and immunostain. Growth factors were added to serum-free media (EGF 8nM, IGF 10nM, insulin 10nM), MK2206 was added to serum complete media (MK2206 5μM). One-way ANOVA.
B. Immunoblots showing Akt T308 phosphorylation in cells treated with vehicle (DMSO) or Akt inhibitior (MK2206 5 μM) for 4 hrs. Corresponding quantification of phospho-T308/total Akt is shown in the graph.
C. CMTM6 and PD-L1 mRNA expression in doxycycline inducible CMTM6 shRNA BEAS-2B cells. (48 hr doxycycline, 1 μg/mL)
Each data point is an experiment All error bars = SEM. * ≤ 0.05, ** ≤ 0.01, *** ≤0.001
Growth factors are known to regulate membrane protein trafficking downstream of AKT activation26, 27. We used pharmacologic inhibition of AKT (MK2206) to determine its role in controlling PD-L1 trafficking. Within 4 hours of pharmacologic inhibition of AKT, plasma membrane PD-L1 was reduced by the same degree as induced by serum withdrawal, demonstrating the amount of PD-L1 on the cell surface is dynamically maintained and acutely regulated by AKT activity (Fig 2B).
The effects of serum and AKT on the expression of PD-L1 in the plasma membrane were not due to global changes in membrane protein traffic because the plasma membrane expression of transferrin receptor, a cargo protein of general endocytic recycling, was not affected by either treatment (Fig 2C). Inhibition of AKT induced downregulation of plasma membrane PD-L1 in all NSCLC cell lines examined (Fig 2D, Fig S1B). Thus, the plasma membrane level of PD-L1 is dynamically maintained and regulated by AKT, irrespective of driver oncogene.
Importantly, inhibition of AKT reduced the amount of PD-L1 in the plasma membrane of IFNγ-treated cells, demonstrating that the increased total amount of PD-L1 induced by IFNγ stimulation remained under the control of AKT regulation (Fig 2E). Consequently, the effects of IFNγ and AKT on the amount of PD-L1 in the plasma membrane are independent and additive.
Previous studies have defined a role for CMTM6 in determining the fate of internalized PD-L114, 15. In agreement with those studies, transient knockdown of CMTM6 resulted in a 30% reduction of total PD-L1 amounts (Fig 2F, Fig S1C). However, the distribution of the remaining PD-L1 between the plasma membrane and interior of cells was unchanged (Fig 2F). These data establish that the amount of PD-L1 recycled back to the plasma membrane, but not the rate at which it is recycled, is controlled by CMTM6. Furthermore, AKT regulation of PD-L1 recycling is unaffected by depletion of CMTM6, providing additional evidence that regulation of PD-L1 downstream of CMTM6 is distinct and independent from regulation downstream of AKT (Fig 2G).
PD-L1 proximal proteome
In addition to PD-L1’s function as a PD1 ligand, there is an emerging literature on cell-intrinsic functions of PD-L117, 18, 28. To provide a context for the further investigation of potential cell-intrinsic roles of PD-L1, we used APEX2 mapping to define the PD-L1 proximal proteome. APEX2 is a peroxidase that specifically oxidizes biotin-phenol to the short-lived phenoxyl radical that reacts with nearby proteins (within a radius of ∼20 nm) resulting in their biotinylation29, 30. The biotinylated proteins are isolated and identified in mass spectrometry to generate an accounting of proteins in the proximity of APEX2. We linked APEX2 enzyme to the cytoplasmic carboxyl-terminus of human PD-L1. Linkage of APEX2 to PD-L1 did not alter PD-L1’s distribution between the plasma membrane and the interior of cells or affect AKT regulation of PD-L1 expression in the plasma membrane, data that support the use of the PD-L1-APEX2 for proximity mapping studies (Fig 3A-C).
A. Representative epifluorescence image of PD-L1-APEX2 plasma membrane stain in BEAS-2B cells.
B. PM to total distribution of PD-L1-APEX2 in BEAS-2B cells. Each data point is an experiment.
C. PM to total ratio of endogenous PD-L1 and PD-L1-APEX2 construct in BEAS-2B cells treated with vehicle (DMSO) or Akt inhibitor (MK2206 5 μM) for 4 hrs. One-sample t -test. Each data point is an experiment.
D. Venn diagram of PD-L1-APEX2 proximal proteins identified in 5 biological replicate experiments.
E. Biological processes pathways enriched in PD-L1-APEX2 labeling in BEAS-2B. Enrichment fold change and FDR are shown for each pathway. N=5
F. PD-L1 co-localization with specified proteins quantified by fraction of PD-L1 intensity co-localized with each target protein. Each point is a collected from single plane images of individual cells under confocal microscopy.
All error bars = SEM. * ≤ 0.05, ** ≤ 0.01, *** ≤0.001
We generated a proximity map for PD-L1-APEX2 stably expressed in a non-transformed human lung epithelial cell line, BEAS-2B. We identified an overlapping set of 399 proteins biotinylated in 5 independent experiments (Fig 3D, Fig S2A, Table S1). PD-L1 proximal proteins reflect many biological pathways, as anticipated for a protein distributed among several cellular compartments as it constitutively cycles to and from the plasma membrane. Here we focus on 4 of the pathways as discussed below (Fig 3E).
A. Volcano plot of PD-L1-APEX2 labeled proteins. FC = fold change of (labeled/−ctrl) spectral intensity. Dashed red lines represent 4-fold change on the x-axis, and p-value corresponding to FDR=0.1 after Benjamini-Hochberg correction on the y-axis respectively.
B. Representative confocal images of PD-L1 colocalization with marked proteins. PD-L1 staining intensity co-localized with marker protein staining was calculated and shown in rightmost panel. Ratio of PD-L1 intensity in colocalized image over original PD-L1 staining was used for quantifications.
Proteins of the MHC I antigen presentation pathway were enriched among the PD-L1 proximity map proteins, including the MHC I receptor proteins HLA-A, HLA-B and HLA-C (Table S1). We confirmed the proximity of PD-L1 to HLA I by immunofluorescence (Fig 3F, Fig S2B). The specificity of the identified proteins to those in proximity to PD-L1 was supported by the fact that the transferrin receptor was neither enriched in the proximity map nor highly colocalized with PD-L1 by immunofluorescence (Table S1, Fig 3F, Fig S2B). The proximity of PD-L1 and MHC I localizes PD-L1 adjacent to the TCR-MHC I complex formed between T cells and target cells, an organization that would enhance the efficiency of PD-L1 to engage PD1 and blunt T cell activation.
There was also an enrichment of proteins that function in various aspects of membrane protein trafficking, a finding in line with constitutive trafficking of PD-L1 to and from the plasma membrane. More unexpected was the finding that 29% of the PD-L1 proximity map proteins are annotated to cytoskeleton organization, covering a broad range of both actin and microtubule biology (Fig 3E). These results infer a close association of PD-L1 or PD-L1-containing compartments with the actin cytoskeleton, which might have implications for PD-L1 function.
Most notably, APEX2 proximity mapping revealed a highly significant enrichment of proteins involved in EGF receptor signaling and trafficking, including the EGFR and several ESCRT proteins, that controls the turnover of activated EGFR. Supporting these findings EGFR and HGS (a subunit of ESCRT-0 complex) overlap with PD-L1 in immunofluorescence (Fig 3E & F, Fig S2B, Table S2).
αPD-L1 antibody induces an increase of mutant EGFR
EGFR and 12 proteins involved in EGFR trafficking and degradation were part of the PD-L1 proximity map (Table S2). As noted above, EGFR activation promotes PD-L1 accumulation on the plasma membrane. However, the other growth factor receptors (insulin receptor and IGF receptor) that can also regulate PD-L1 distribution between the plasma membrane and interior of cells, were not identified in PD-L1 proximity mapping. These findings suggest the possibility of a more direct connection between PD-L1 and EGFR. Finding EGFR and proteins associated with its signaling and trafficking was intriguing because EGFR mutant tumors respond poorly to α-PD-L1 therapy, despite expressing PD-L13–5. Perhaps the reduced response of EGFR mutant tumors is related to a physical or functional interaction between PD-L1 and activated EGFR. To further probe the relationship between EGFR and PD-L1 in the context of PD-L1 therapy, we first determined the effect of α-PD-L1 antibody treatment on EGFR (Fig 4A). Incubation of three different EGFR mutant human NSCLC cell lines for 7 hours with α-PD-L1 antibody caused a significant increase of EGFR protein, whereas similar treatment of two KRAS mutant lines and non-transformed BEAS-2B cells did not affect EGFR protein levels. Of note, the EGFR mutant lines harbor different activating mutations, demonstrating the effect is not linked to a specific EGFR mutation. Additionally, despite differences in PD-L1 expression among the lines, the effect of α-PD-L1 was restricted to cells with oncogenic EGFR. (Fig S3A).
A. PD-L1 total protein expression in cell lines. Quantification of immunofluoresncence staining is shown. Each data point is an experiment.
B. Representative immunoblot of EGFR and actin in H1650 cells treated with αTfR antibody
C. Representative immunoblot of EGFR and actin in WT, PD-L1 KO and PD-L1-reexpressing H1650 cells.
D. Representative immunoblot of EGFR and actin in H1650 cells treated with αPD-L1 antibody and αPD-L1 Fab.
E. EGF (50ng/mL) stimulated EGFR degradation in BEAS-2B cells treated with αPD-L1 antibody (10μg/mL, 7 hr). N=4. Tested for change in rate constants.
F. TGFα (50ng/mL) stimulated EGFR degradation in BEAS-2B cells treated with αPD-L1 antibody (10μg/mL, 7 hr). N=3. Tested for change in slope values.
G. EGF (0.5 ng/mL) stimulated EGFR, Akt and ERK phosphorylation in H1650 cells treated with αPD-L1 antibody (10μg/mL, 7 hr). Immunoblot quantification of EGFR protein (left) and phospho/total EGFR, Akt and ERK is shown respectively (right three panels) . N=3
H. EGF (0.5 ng/mL) stimulated gene transcription in H1650 cells treated with αPD-L1 antibody (10μg/mL,7 hr). qRT-PCR mRNA quantification of EGFR target genes is shown. N=3
I. PD-L1 colocalization with EGFR quantified by fraction of PD-L1 intensity co-localizing with EGFR via confocal microscopy. Each point is collected from single plane images of individual cells under confocal microscopy. Unpaired t-test.
All error bars = SEM. * ≤ 0.05, ** ≤ 0.01, *** ≤0.001
A. EGFR protein expression in cell lines treated with αPD-L1 antibody (10μg/mL, 7 hr). Western blot quantifications of EGFR protein normalized to actin expression are shown. One sample t-test.
B. Immunoblot of EGFR protein expression in H1650 cells treated with reagent grade αPD-L1 antibody (clone # 130021, Novus Biologicals) and clinical Durvalumab as compared to isotype IgG control antibody (all antibodies at 10μg/mL, 7 hr). Western blot quantifications of EGFR protein normalized to actin expression are shown. One-way ANOVA and post-hoc Tukey test.
C. EGFR protein in H1650 cells with αTfR antibody treatment (10μg/mL, 7 hr). One sample t-test.
D. EGFR protein in H1650 cells with αPD-L1 antibody and αPD-L1 Fab treatment (10μg/mL). One way ANOVA and post-hoc Tukey test.
E. PD-L1 protein levels and distribution in BEAS-2B cells treated with Durvalumab (10μg/mL, 7 hr). Unpaired multiple t-tests.
F. PD-L1 protein levels and distribution in H1650 cells treated with Durvalumab (10μg/mL, 7 hr). Unpaired multiple t-tests.
G. TfR protein levels and distribution in H1650 cells treated with Durvalumab (10μg/mL, 7 hr). Unpaired multiple t-tests.
H. Immunoblot of PD-L1 protein expression in WT, PD-L1 KO and PD-L1-reexpressing H1650 cells. Cells were pretreated with IFNγ (40ng/mL, 24 hr)
I. EGFR protein in WT, PD-L1 KO and PD-L1-reexpressing H1650 cells. One way ANOVA and post-hoc Tukey test.
J. EGFR mRNA in H1650 cells treated with αPD-L1 antibodies (10μg/mL, 7 hr). One sample t-test.
K. PD-L1 mRNA in H1650 cells treated with αPD-L1 antibodies (10μg/mL, 7 hr). One sample t-test.
L. Immunoblot of EGFR protein expression in H1650 cells treated with αPD-L1 antibody (10μg/mL, 7 hr) and cycloheximide (10μg/mL, 7 hr). Western blot quantifications of EGFR protein normalized to actin expression are shown. One-way ANOVA and post-hoc Tukey test.
M. Immunoblots of EGFR degradation in H1650 and BEAS-2B cells treated with cycloheximide (10 μg/mL) in serum complete media. Actin is used as loading control.
N. Quantification of EGFR degradation in H1650 cells. N=3 Tested for change in rate constants.
O. Quantification of EGFR degradation in H1650 cells. N=3 Tested for change in rate constants.
P. Ubiquitinated EGFR protein in BEAS-2B cells treated with αPD-L1 (10μg/mL, 7 hr) followed by EGF for specified time (50 ng/mL). Immunoblots of whole cell extract and EGFR IP fraction are shown. Ubiquitinated/total EGFR in IP fraction was quantified. Paired ratio t-test.
Q. Ubiquitinated EGFR protein in H1650 cells treated with antibody (10μg/mL, 7 hr) followed by EGF for specified time (50 ng/mL). Immunoblots of whole cell extract and EGFR IP fraction are shown. Ubiquitinated/total EGFR in IP fraction was quantified. Paired ratio t-test.
Except panel N &O, each data point is an experiment. All error bars = SEM. * ≤ 0.05, ** ≤ 0.01, *** ≤0.001, # = 0.051
Two α-PD-L1 antibodies, Durvalumab and reagent grade PD-L1 antibody (clone# 130021, Novus Biologicals), which bind different extracellular epitopes of PD-L1, had a similar effect on mutant EGFR levels, demonstrating the effect is not dependent on antibody binding to a specific region/epitope of PD-L1 (Fig 4B). Neither an anti-transferrin receptor antibody, which binds to the extracellular domain of the transferrin receptor, nor a control human IgG antibody, when incubated with cells for 7 hrs, affected the amount of EGFR protein, demonstrating the effect on mutant EGFR is specific for antibodies that bind PD-L1 (Fig 4B & C, Fig S3B). Finally, treatment with an α-PD-L1 Fab (monovalent) caused an increase in EGFR protein, thus proving that the effect of the antibody is not through forced dimerization of PD-L1 (Fig 4D, Fig S3C).
α-PD-L1 antibody treatment did not alter the total amount of PD-L1 nor its cellular distribution in either non-transformed or EGFR mutant human lung epithelial cell lines (Fig 4E&F). In addition, α-PD-L1 treatment did not affect the distribution of the transferrin receptor, demonstrating the treatment did not affect clathrin-mediated endocytosis or constitutive recycling back to the plasma membranes (Fig 4G).
αPD-L1 antibody reduces degradation of mutant EGFR
We next knocked-out PD-L1 in an EGFR mutant cell line using CRISPR/CAS9. We focused on EGFR mutant cells because these are the cells that upregulated EGFR in response to α-PD-L1 treatment. PD-L1 knockout resulted in an increase in mutant EGFR that was reduced to control levels by re-expression of PD-L1 (Fig 4H & I, Fig S3D). These data support the hypothesis that the effect of α-PD-L1 is not a gain of function induced by antibody binding, but rather that antibody binding disrupts a cell-intrinsic function of PD-L1.
α-PD-L1 treatment did not affect EGFR or PD-L1 mRNA levels, demonstrating a post-transcriptional mechanism is responsible for the increase in mutant EGFR protein levels (Fig 4J & K). Inhibition of translation via cycloheximide (CHX) did not alter the effect of α-PD-L1 on mutant EGFR levels, supporting the conclusion that α-PD-L1 inhibits EGFR degradation rather than increases translation of EGFR mRNA (Fig 4L). Accordingly, α-PD-L1 treatment approximately doubled the half-life of mutant EGFR without an appreciable effect on the half-life of the wildtype EGFR (Fig 4M-O). Consistent with the effect of α-PD-L1 on mutant EGFR being independent of kinase activity, α-PD-L1 did not affect the half-life of wildtype EGFR stimulated with EGF or TGF-α (Fig S3E & F). These data indicate that some feature of mutant EGFR biology or the cellular context of EGFR mutant cells is not recapitulated by acute activation of EGFR and is responsible for the effect of α-PD-L1. However, α-PD-L1 treatment did not lead to significant differences in AKT or ERK phosphorylation or an increase in transcription of immediate EGFR response genes, providing additional evidence that the α-PD-L1 effect was not a consequence of altered EGFR signaling (Fig S3G & H).
Prior studies have shown that ligand activated wildtype EGFR and constitutively active mutant EGFR have different intracellular trafficking patterns31. One prominent difference between trafficking of wildtype and mutant EGFR is that activated wildtype EGFR is sorted to luminal vesicles of multivesicular late endosomes, followed by degradation in mature lysosomes32. Mutant EGFR, despite being active, is poorly trafficked to late endosomes and rather is recycled back to the plasma membrane, thereby avoiding efficient degradation31, 33. Although unstimulated wildtype EGFR co-localized with PD-L1, this colocalization was significantly decreased upon EGF activation of the wildtype EGFR, presumably because activated wildtype EGFR is targeted for degradation. Mutant EGFR, however, remained co-localized with PD-L1 regardless of ligand stimulation (Fig S3I). Thus, the different effect of α-PD-L1 on the half-lives of wildtype and mutant EGFR may be due to differences in the trafficking of wildtype and mutant EGFR.
Ubiquitination of activated wildtype EGFR targets it for degradation by a mechanism involving the ESCRT pathway, and although mutant EGFR is ubiquitylated following binding of EGF, ubiquitination is not required for internalization of mutant EGFR32, 34–37. α-PD-L1 antibody treatment of cells reduced ubiquitination of both wildtype and mutant EGFR, suggesting differences in EGFR ubiquitination do not account for the difference in turnover of wildtype and EGFR mutants in cells treated with α-PD-L1 (Fig 4P & Q).
αPD-L1 antibody treatment effect on the whole cell proteome
ESCRT proteins were identified in the PD-L1 proximity map (Fig 3E, Table S2). Because α-PD-L1 affects mutant EGFR turnover and ESCRTs are involved in controlling the turnover of a number of proteins38, not just EGFR, we next explored the possibility that α-PD-L1 treatment has a global effect on protein turnover. To test this hypothesis, we performed whole cell proteomic analysis on wildtype, KRAS and EGFR mutant cells 48 hrs after treatment with therapeutic α-PD-L1 Durvalumab antibody (Fig 5A). Remarkably, in the EGFR mutant cell line there was an overall increase in the post-to-pre α-PD-L1 treatment ratios of individual proteins, whereas in wildtype and KRAS mutant cells lines there was a small overall decrease in post to pretreatment ratios (Fig 5A). Mutant EGFR was increased 1.3-fold in whole cell mass spec, whereas wildtype EGFR was unchanged (Table S3). These data demonstrate that α-PD-L1 treatment does not just affect mutant EGFR but that it has broader effects that are limited to cells with mutant EGFR, revealing the effect is in the context of a cell expressing constitutively active EGFR.
A. Ratio of mass spectrometry intensity of proteins in extracts from cells treated with Durvalumab (10μg/mL, 48 hr) to those from untreated cells. Frequency distribution of +/− Durvalumab ratios of whole cell proteins identified is plotted. Average value for N=2 is shown.
B. Heatmap of proteins enriched in H1650 cells with Durvalumab treatment that were identified in immune related processes. +/− Durvalumab ratio for each of the two experiments is shown.
C. Heatmap of proteins enriched in H1650 cells with Durvalumab treatment that were identified in cell polarity related processes. +/− Durvalumab ratio for each of the two experiments is shown.
D. Heatmap of extracellular vesicle marker proteins expression in whole cell and EV fraction across cell lines treated with Durvalumab (10μg/mL, 48 hr). Average +/ − Durvalumab value for N=2 is shown.
E. Ratio of mass spectrometry intensity of proteins identified in extracellular vesicles from cells treated with Durvalumab (10μg/mL, 48 hr) to those from untreated cells. Frequency distribution of +/− Durvalumab ratios of extracellular vesicle proteins is plotted. Average value for N=2 is shown.
To explore specific biology that might be most affected by α-PD-L1 treatment we queried the proteins most consistently increased by α-PD-L1 by applying thresholds of greater than 1.2-fold increase and a coefficient of variance less than 0.1 (lowest quartile of variation between the experiments). Among these proteins, there was an enrichment of proteins involved in immune responses such as cellular response to interferon-beta, regulation of innate immune responses as well as viral release from host cells, suggesting the possibility that PD-L1, in addition to its function as an immune checkpoint ligand, has cell-intrinsic immune-related function intersecting with ESCRT-dependent functions (Fig 5B, Fig S4A). We also detected an enrichment of proteins annotated to cell adhesion and cell polarity, suggesting a role for PD-L1 in regulating cytoskeletal organization (Fig 5C, Fig S4A). Notably, these proteins were unchanged in KRAS and non-transformed cells with α-PD-L1 antibody treatment (Fig 5B & C, Table S3).
A. Biological processes representing proteins that are increased in whole cell extract of H1650 cells treated with Durvalumab (10μg/mL, 48 hr) over untreated cells.
B. Lysosomal acidity in H1650 cells with PD-L1 KO or treated with αPD-L1 antibody. Fluorescein and rhodamine labeled dextran was pulsed into cell growth media at 1mg/mL for 16 hours and chased for 0-3 hr. Cells were imaged live and fluorescein/rhodamine (F/R) ratio was calculated for regions automatically determined by rhodamine threshold. Fluorescein/rhodamine ratio in H1650 following 16 hr pulse and 3 hr chase of labeled dextran is shown. αPD-L1 antibody (10μg/mL) was added during the pulse and maintained in the chase. Each data point is an experiment. One sample t-test.
C. Lysosomal maturation as shown by average fluorescein/rhodamine ratio during 3 hr dextran chase in αPD-L1 antibody treated or PD-L1 KO H1650 cells. αPD-L1 antibody (10μg/mL) was added during the 16 hr pulse and maintained in the chase. N=3
D. Lysosomal maturation as shown by fluorescein/rhodamine frequency distribution of dextran labeled compartments during 3 hr chase. Representative experiment
αPD-L1 antibody treatment affects extracellular vesicles
ESCRTs, in addition to having a key role in targeting EGFR (and other proteins) for lysosomal degradation, are also involved in the formation and secretion of extracellular vesicles (EVs)39. To determine the impact of α-PD-L1 treatment on EVs, we performed proteomic analysis on EVs isolated from the media of non-transformed, KRAS and EGFR mutant lung cell lines with and without α-PD-L1 treatment. α-PD-L1 treatment had a pronounced effect on EV biogenesis in EGFR mutant cells compared to KRAS and wildtype cells (Fig 5D & E). There was an increase in a set of 6 canonical EV proteins in EVs isolated from the medium of EGFR mutant cells treated with α-PD-L1. There was no similar increase of these proteins in EVs isolated from the media of wildtype or KRAS mutant cells incubated with α-PD-L1 (Fig 5D). The increase of these EV proteins in the extracellular space was higher than the increase in the whole cell extracts, supporting the hypothesis that α-PD-L1 treatment enhances EV biogenesis and/or release from EGFR mutant cells (Fig 5D). In addition, there was a striking overall increase of EV cargo proteins from EGFR mutant α-PD-L1 treated cells, suggesting either enhanced biogenesis of EVs or increased protein packaging into EVs (Fig 5E). Thus, our data establish that α-PD-L1 treatment affects protein turnover and EV biogenesis, two ESCRT dependent functions.
The proper maturation and acidification of late endosomes is critical for EV biogenesis as well as lysosomal activity40, 41. We determined that the acidification and maturation of late endosomes/lysosomes in EGFR mutant cells were not affected by α-PD-L1 treatment, demonstrating that the global effects of α-PD-L1 treatment on the proteome of EGFR mutant cells was not due to perturbations in lysosomal acidification (Fig S4B-D).
PD-L1 controls cell migration in EGFR mutant cells
Our APEX2 mapping (Fig 3D), in addition to revealing proximity of PD-L1 to EGFR, also identified an enrichment of proteins that function in cytoskeletal biology. We also identified proteins involved in cell polarity and cell adhesion processes among the ones increased with α-PD-L1 treatment in EGFR mutant cells. Thus, we next determined whether PD-L1 impacts cytoskeleton-dependent processes. α-PD-L1 treatment of an EGFR mutant cell line reduced migration measured in a scratch assay and reduced migration assayed in a transwell migration assay (Fig 6A-B, Fig S5A). Treatment with a control IgG antibody did not impact cell migration (Fig 6C). Furthermore, prolonged α-PD-L1 treatment did not affect total cell levels of PD-L1, demonstrating the effect of antibody is not through induced degradation and loss of PD-L1 protein (Fig S5B). α-PD-L1 treatment inhibited migration of a second EGFR mutant harboring a different EGFR mutation, whereas α-PD-L1 treatment of a KRAS mutant cell line had no effect (Fig 6D &E). In line with those results, PD-L1 knockout blunted cell migration, mimicking the effect of the α-PD-L1 antibody (Fig 6F &G, Fig S5C). Migration was restored when full length PD-L1 was re-expressed in PD-L1 knockout cells, but not when a mutant PD-L1, which lacks the cytoplasmic domain (CDΔ) was expressed. This suggests a role for PD-L1’s cytoplasmic domain in the regulation of migration (Fig 6F & G, Fig S5C). Neither α-PD-L1 treatment nor PD-L1 knockout impacted cell viability or proliferation, demonstrating the difference in migration is not confounded by changes in cell proliferation (Fig S5D & E).
A. Timelapse images of scratch assay at 0, 8 and 24 hr in H1650 cells treated with Durvalumab (10μg/mL). Leading edge of the cells are marked in white. Cells were in serum-free media for the duration of the assay.
B. PD-L1 protein expression in H1650 treated with Durvalumab (10μg/mL) for 5 days. Each data point is an experiment. One sample t-test.
C. Timelapse images of scratch assay at 0, 8 and 24 hr in H1650 KO rescue cells. Leading edge of the cells are marked in white. Cells were in serum complete media for the duration of the assay.
D. Cell proliferation as measured by MTT assay in H1650 cells treated with Durvalumab (10μg/mL). N=2 Difference in slope values is calculated.
E. Cell proliferation as measured by MTT assay in H1650 KO rescue cells. N=3 Difference in slope values is calculated.
A. Quantification of scratch assay in H1650 cells treated with Durvalumab (10μg/mL). Cells were in serum-free media for the duration of the assay. N=3 Difference in slope values is calculated.
B. Transwell migration of H1650 cells treated with Durvalumab (10μg/mL). Paired t-test
C. Transwell migration of H1650 cells treated with control IgG (10μg/mL). Paired t-test
D. Transwell migration of H1975 cells treated with Durvalumab (10μg/mL). Paired t-test.
E. Transwell migration of A549 cells treated with Durvalumab (10μg/mL). Paired t-test
F. Quantification of scratch assay in H1650 KO/rescue cells. Cells were in serum complete media for the duration of the assay. N=3-4 Difference in slope values is calculated.
G. Transwell migration in H1650 KO/rescue cells. One-way ANOVA and post-hoc Tuckey test.
For all experiments with antibody treatments, cells were pretreated with specified antibody (10μg/mL) for 5 days prior to the start of experiment. Except panels A & F, each data point is an experiment. All error bars = SEM. * ≤ 0.05, ** ≤ 0.01, *** ≤0.001
Discussion
In this study, we show that PD-L1, beyond its function as ligand for the immune checkpoint receptor PD1, is spatially linked with key protein pathways involved in EGFR/ESCRT biology and actin-dependent motility. We demonstrate that α-PD-L1 treatment, including the use of the therapeutic α-PD-L1 Durvalumab antibody, has a global effect on the whole cell proteome and leads to increased biogenesis of extracellular vesicles, specifically in EGFR mutant cells. In addition, α-PD-L1 treatment leads to decreased cell migration in EGFR mutant cells. Mechanistically, α-PD-L1 treatment results in loss of function of PD-L1, suggesting that the clinical effects of α-PD-L1 may, by affecting tumor cell-intrinsic functions of PD-L1, extend beyond the disruption of PD1 engagement.
EGFR mutant tumors are generally resistant to α-PD-L1 therapy as compared to EGFR WT tumors3–5. While the mechanism behind this therapeutic disparity is not clear, there is emerging evidence suggesting that EGFR activity and PD-L1 may be functionally linked. We have shown here that surface PD-L1 levels are acutely increased by AKT signaling downstream of EGFR without affecting total protein abundance and independently of CMTM6. We found that the majority of PD-L1 at steady state is located in a dynamic intracellular storage pool, which allows cells to rapidly modulate PD-L1 expression on the plasma membrane in response to EGF. This EGFR:PD-L1 link is further supported by our finding that PD-L1 is located in close proximity to not only EGFR, but 12 other proteins known to be involved in EGFR endocytosis, signaling and protein degradation including members of the ESCRT complexes. This suggests that PD-L1 which can be regulated by EGFR may in turn directly affect EGFR protein biology. Indeed, we show that α-PD-L1 treatment reduces EGFR degradation and results in an increase of EGFR protein in EGFR mutant cells. In our in vitro experiments we did not observe a significant difference in downstream EGFR signaling accompanying the α-PD-L1 stimulated increase of total EGFR protein; however, it is possible that the 50% increase in total EGFR might result in increased tumor growth in vivo by providing essential signals that support survival in the nutrient-deprived tumor microenvironment.
The effect of Durvalumab or other antibody binding to PD-L1 has not been extensively studied in the context of PD-L1 reverse signaling. We find that antibody binding results in a loss of function of PD-L1 as it phenocopies the effect of PD-L1 KO on EGFR expression. Although, PD-L1 antibody binding can trigger antibody-dependent cell cytotoxcity42, 43, Durvalumab lacks a functional Fc domain, and anti-transferrin receptor antibody treatment suggests the changes are not associated with cell response to antibody binding but is specific to PD-L1 function. It is possible that antibody binding causes transmembrane signaling or disrupts PD-L1 binding with partner proteins, which are responsible for signaling changes. Recent studies have identified PD-L1 signaling motifs and binding partners that result in broad transcriptional and signaling changes within the cells17, 18, 28 .
PD-L1 protein is reported to be degraded both through proteasomal and lysosomal pathways9, 12, 16, and our APEX2 mapping data are consistent with PD-L1 being in proximity to the ESCRT machinery. ESCRT proteins drive the formation of multi vesicular bodies (MVBs), which either fuse with lysosomes, resulting in degradation of contents or fuse with the plasma membrane releasing into the extracellular space the luminal vesicles as EVs38, 39, 44, 45. The global increase in total cell protein and EV biogenesis that we observe in Durvalumab treated mutant EGFR cells suggests that PD-L1 is not merely a client of the ESCRT machinery but is involved in its regulation. Finding reduced turnover of EGFR (a client of ESCRT-dependent degradation)32 and an increased biogenesis of EVs is consistent with α-PD-L1 biasing MVB maturation towards EV biogenesis. This is in line with previous reports implicating PD-L1 in the ESCRT-dependent process of autophagy46. The oncogenic EGFR-specific whole cell proteome and EV changes may explain why only cells with mutant EGFR were susceptible to antibody-induced change in migration. EVs are known to have broad roles in regulating cell migration, ranging from autocrine signaling in directed migration to recruitment and infiltration of immune cells in the microenvironment47. However, at this time we cannot distinguish between a direct effect of anti-PD-L1 on cell migration or if it is secondary to changes in ESCRT function. Additionally, EGFR signaling regulates cell migration and invasion48, 49, and, although we did not find differences in EGF-stimulated EGFR signaling, we cannot rule out the possibility of increased mutant EGFR protein directly impacting cell migration in oncogenic EGFR cells.
Although PD-L1 has been reported to affect cell migration in cell lines with non-EGFR driver oncogenes50, most of these studies rely on PD-L1 ablation and do not employ α-PD-L1 treatment. We propose that α-PD-L1 treatment does not result in a complete loss of function of PD-L1 and requires, in vitro at least, the enhanced signaling of EGFR mutant cells to display a change in phenotype. A recent study showed such a scenario where in PD-L1 induces phospholipase C-γ1 (PLC-γ1) activation only in the presence of activated EGFR18. Wildtype and mutant EGFR have been described to have largely dissimilar downstream signaling51, 52, and our data with non-transformed, KRAS and EGFR mutant cells suggests that the cellular context created by oncogenic EGFR signaling may make EGFR mutant cells more susceptible to α-PD-L1 antibody mediated signaling changes. Among the proteins that increased the most in EGFR mutant cells upon Durvalumab treatment, were those involved in immune regulatory processes, which closely relate to PD-L1’s role as an immune checkpoint protein53. Most notably, these proteins were unchanged in non-transformed and KRAS mutant cells highlighting the distinct response of EGFR mutant cells to α-PD-L1 antibody. This finding may at least partially explain why EGFR mutant tumors respond poorly to PD-L1 checkpoint blockade and suggest that PD-L1 checkpoint monotherapy should be carefully reconsidered for patients with mutant EGFR tumors.
Besides tumor cells, PD-L1 expression on immune cells has been shown to be important for regulation of migration. Specifically, PD-L1 on dendritic cells (DCs) was shown to be necessary for efficient homing of dermal DCs to the lymph nodes through modulation of chemokine receptor signaling and actin remodeling28. Our data suggests that in addition to anti-PD-L1 effect on PD-L1 blockade on DCs, shown to be crucial for tumor control54, 55, it is important to follow up on the cell intrinsic effects of anti-PD-L1 treatment on immune cell migration and proliferation.
In line with literature suggesting that PD-L1 traffics through endosomes16, 56, we find that PD-L1 is localized on the plasma membrane of cells as well as in intracellular storage pools. Unlike, previous reports of PD-L1 transiting through endosomes for degradation, we find that majority of PD-L1 at steady state is located in intracellular compartments and this pool is maintained to dynamically modulate PD-L1 expression on the plasma membrane. This change is achieved within hours, and perhaps represents the fastest mode of modulating plasma membrane PD-L1, which could be crucial to limit collateral damage and promote cell survival in an immune inflamed environment.
Through APEX proximity labeling and confocal microscopy we establish the proximity of PD-L1 with MHC I, in agreement with recent results57. This organization suggest that PD-L1 is part of the immune synapse, which would contribute to its efficiency as an immune checkpoint upon engagement of MHC-I and the T cell receptor. In accordance with PD-L1 localization to specialized domains of the plasma membrane, caveolin 1 and flotillin, both of which have roles in establishing specialized membranes within a membrane58, 59, are constituents of the PD-L1 proximal proteome (Table S1). Additional studies are required to determine how the clusters form and what role they have PD-L1 engagement of PD1 receptor60–62.
In conclusion, we describe a novel aspect of PD-L1 biology whereby α-PD-L1 treatment results in a loss of function of cell intrinsic PD-L1 function leading to profound changes in whole cell proteome and EV secretion, upregulation of mutant EGFR protein and reduced cell migration specifically in EGFR mutant lung cancer cells. The signaling mechanism behind the changes in EGFR mutant cells with α-PD-L1 treatment remain to be elucidated. Our study highlights the importance of tumor intrinsic role of PD-L1 in EGFR mutant cells and stress the need for further studies of PD-L1 antibody-based therapies beyond the immune cell regulatory context.
Materials and Methods
Cell lines and culture
A549, BEAS-2B and H23 were purchased from ATCC. H1650, HCC827, HCC4006 and PC9 were a gift from Dr. Rafaella Sordella (Cold Spring Harbor Laboratory). H1975 were a gift from Dr. Brandon Stiles (Weill Cornell Medicine). All primary human samples were obtained and used in accordance with an approved IRB protocol. Primary tumor cells were processed to obtain single cell suspensions as described previously63 and plated onto glass-bottom dishes and analyzed 3-5 days later. All cells were cultured in DMEM media supplemented with 10% FBS and 1% penicillin-streptomycin at 37°C and 5% CO2. Where indicated, cells were treated with serum-free DMEM media supplemented with 0.02M HEPES and 0.03M NAHCO3. Stable cell lines were generated via lentiviral infection using Lenti-X packaging system. Cells were either antibiotic selected via blasticidin (5 μg/mL) or puromycin (3 μg/mL), or sorted for expression of protein of interest via FACS. PD-L1 KO cells were generated via CRISPR/CAS9 with both targeting sequences (guide1: 5’-CGGGAAGCTGCGCAGAACTG-3’; guide2: 5’-TCTTTATATTCATGACCTAC-3’). Live cells were stained with PD-L1 antibody and KO cells were sorted by FACS. Bulk population of KO cells was verified for PD-L1 KO via Western Blot. CMTM6 stable inducible shRNA knockdown cells were generated with one shRNA targeting sequence (5’-TTCAAAGAAGTAGAGGCCTCCA-3’) and mRNA knock-down was confirmed via qRT-PCR.
Cell treatments
For PD-L1 PM to total measurements, cells were plated on glass bottom dishes in full growth media and assayed the next day. Cells were switched to either serum-free media, serum free media supplemented with growth factors or fresh full growth media for 4 hours. EGF was used at 8nM, IGF at 10nM and insulin at 10nM. For Akt inhibition experiments, cells were incubated in full growth media with 5 μM MK2206 or DMSO for 4 hours. Interferon gamma treatment was performed at 40 ng/mL for 24 hrs, and CMTM6 shRNA knock-down was induced by doxycycline at 1 μg/mL for 48hrs. For antibody treatment experiments, cells were treated with specified antibody or Fab for 7 hr at a final concentration of 10 μg/mL in serum free DMEM media. For protein turn-over experiments cycloheximide was used at 10 μg/mL. For stimulated EGFR degradation, cells were serum starved for 4 hr and 5 hr and treated with EGF or TGFα at 50ng/mL for indicated timepoints respectively, while antibody was used at 10 μg/mL and maintained for the duration of the experiment. For EGFR signaling experiments, cells were incubated with antibody in serum free media for 7 hr and stimulated with 0.5 ng/mL EGF for specified timepoints. For migration assays, cells were pre-incubated with Durvalumab or IgG control antibody for 5 days before being plated for experiment.
Microscopy
Cells were plated on glass-bottom dishes prior to all experiments. After corresponding experimental treatments, cells were washed and fixed in 3.7% formaldehyde for 5 minutes. Cells were stained with primary and fluorescent secondary antibodies for proteins of interest on the plasma membrane without permeabilization. Saponin (0.25 mg/mL) was used to permeabilize the cells for total protein stain. Cell nuclei were labeled with Hoechst stain. For quantifications, images were acquired on an inverted microscope with a 20× air objective or a 40x oil objective (Leica Biosystems DMI6000-B). Images were analyzed on MetaMorph image processing software and the background corrected average intensity under each cell’s area was used to quantify protein amount. The values were normalized to the average of replicate coverslip dishes in control condition in each experiment to allow for comparison across experiments. Unpaired two tailed Student’s t test or paired ratio t-test were performed on unnormalized raw data, and one sample t test was used on normalized data respectively as indicated. For comparison of three and more groups, one-way ANOVA was performed on raw data with post-hoc Tuckey test where indicated. For PD-L1 stain images, Z stack of the cells was acquired using Zeiss Confocal microscope (LSM880) with Airyscan on a 40x oil objective.
Exocytosis assay
Live cells were incubated with a saturating concentration of PD-L1 antibody (50 μg/mL) for indicated times, washed extensively with PBS, and then fixed. All cells were permeabilized and stained with a secondary antibody to visualize all antibody bound to the cells. Separate set of dishes was used to quantify total PD-L1 stain in fixed cells in each experiment. Cells were imaged and quantified as described above. All values were normalized to the plateau, which is defined as the average of values at 60 and 120min in each experiment.
Colocalization analysis
For co-localization analysis cells were fixed and stained for total PD-L1 and the protein of interest with primary antibody followed by Cy3 and Alexa488-labeled secondary antibodies. Cells were imaged on Zeiss Confocal microscope (LSM880) with Airyscan on a 40x oil objective. Images were analyzed in MetaMorph software. Intensity threshold was manually determined for each colocalization marker and kept constant for a binary mask of the marker. The binary mask was overlayed with PD-L1 staining resulting in PD-L1 stain only in pixels positive for the marker proteins. For each cell, integrated intensity of PD-L1 staining in the original images as well as the overlayed images were calculated. Ratio of overlayed intensity over original intensity was used as “fraction under mask”. Unpaired t-test was performed on raw data.
Quantitative RT-PCR
Cells were lysed after indicated treatments in each experiments and RNA was extracted using RNeasy kit, and cDNA was prepared from extracted RNA using the RNA to cDNA EcoDry Premix. Quantitative RT-PCR was performed using appropriate primer pairs indicated in Supplementary Table 4. 2(-Delta Delta C(T)) method was used to quantify gene expression.
Western blotting
Cells were lysed in laemmli buffer with protease/phosphatase inhibitors and the lysate was boiled at 95°C for 7 minutes. Lysates were run on 8% or 10% SDS gels and transferred onto nitrocellulose membranes. The membrane was blocked with 5% milk or Licor blocking buffer for 1 hr at room temperature, then incubated with primary antibody overnight at 4°C followed by secondary antibody for 1 hr at room temperature. Blots were visualized using Licor Odyssey Fc Imager for two color or Thermofisher MyECL Imager for single color assays and analyzed on ImageStudio or ImageJ software respectively. Actin or GAPDH expression was used as loading control where indicated. One sample t-test was performed on normalized data for comparison of two groups. One-way ANOVA and post-hoc paired ratio t-test was performed on raw data for three or more groups. For degradation assays, K rate constant for the non-linear regression and slopes for linear regression were used to calculate statistical difference.
EGFR Immunoprecipitation
Cells were grown to confluency and treated with anti-PD-L1 Ab (clone #130021, Novus Biologicals) for 7 hr in serum free media, followed by addition of EGF (50ng/mL) for specified times. Cells were lysed in lysis buffer (150 mM NaCl, 50 mM Tris-HCl pH 7.5, 1% Triton X-100) with protease/phosphatase inhibitors at given timepoints. Protein concentration was measured using BCA assay and lysates were incubated with anti-EGFR antibody (4 μg/mL) for 1 hr at 4°C on a rotator. Lysates were then added to pre-washed protein A/G agarose beads and rotated overnight at 4°C. Beads were washed four times with lysis buffer and boiled in 1.5x Laemmli buffer for 5 minutes at 95°C to elute the proteins. Resulting IP fraction was analyzed via Western blot along with whole cell extract.
Extracellular vesicle, whole cell extract mass spectrometry
Cells were adjusted to media supplemented with 10% exosome free FBS for 1 passage and plated at 60% confluency in exosome free media. After 48 hours of Durvalumab treatment (10 μg/mL), cell media was collected for extracellular vesicle isolation and cells were lysed in 1x RIPA buffer with protease/phosphatase inhibitors. Extracellular vesicles were isolated by sequential ultracentrifugation (12000g for 10 min to discard cell debris, 20000 g for 20 min to discard microvesicles, 100000 g for 70 min twice to isolate and wash the EVs) and resuspended in a final volume of 100 μL of RIPA buffer with protease/phosphatase inhibitors. Cell extract and EV lysates were quantified for protein concentration. A final amount of 100 μg of WCE, and equi-volume of EVs approximating 10 μg of protein in each sample was submitted for mass spectrometry analysis.
The samples were digested in solution with trypsin overnight at 37C following reduction with 5mM DTT and 14mM alkylation with iodoacetamide. The digests were vacuum centrifuged to near dryness and desalted by Sep-Pak column.
A Thermo Fisher Scientific EASY-nLC 1000 coupled on-line to a Fusion Lumos mass spectrometer (Thermo Fisher Scientific) was used. Buffer A (0.1% FA in water) and buffer B (0.1% FA in ACN) were used as mobile phases for gradient separation (1). A 75 µm x 15 cm chromatography column (ReproSil-Pur C18-AQ, 3 µm, Dr. Maisch GmbH, German) was packed in-house for peptides separation. Peptides were separated with a gradient of 3–28% buffer B over 110 min, 28%-80% B over 10 min at a flow rate of 300 nL/min. The Fusion Lumos mass spectrometer was operated in data dependent mode. Full MS scans were acquired in the Orbitrap mass analyzer over a range of 300-1500 m/z with resolution 120,000 at m/z 200. The top 15 most abundant precursors with charge states between 2 and 5 were selected with an isolation window of 1.4 Thomsons by the quadrupole and fragmented by higher-energy collisional dissociation with normalized collision energy of 35. MS/MS scans were acquired in the Orbitrap mass analyzer with resolution 15,000 at m/z 200. The automatic gain control target value was 1e6 for full scans and 5e4 for MS/MS scans respectively, and the maximum ion injection time is 60 ms for both.
The raw files were processed using the MaxQuant computational proteomics platform version 1.5.5.1 (Max Planck Institute, Munich, Germany) for protein identification. The fragmentation spectra were used to search the UniProt human protein database (downloaded on 01/16/2021). Oxidation of methionine and protein N-terminal acetylation were used as variable modifications for database searching. The precursor and fragment mass tolerances were set to 7 and 20 ppm, respectively. Both peptide and protein identifications were filtered at 1% false discovery rate based on decoy search using a database with the protein sequences reversed.
PD-L1-APEX2 labeling and mass spectrometry
APEX2 proximity assay was performed as described previously64. Briefly, BEAS-2B cells stably expressing PD-L1 APEX2 were grown until confluency. Cells were preincubated with biotin phenol in full growth media for 30 minutes and hydrogen peroxide was added for 1 minute (except in negative control). Cells were washed three times with ice cold quenching solution (10 mM sodium ascorbate, 5 mM Trolox and 10 mM sodium azide solution in PBS) and cells were lysed in 1x RIPA lysis buffer with protease/phosphatase inhibitors. Protein concentration was measured, and biotinylated proteins were pulled down using streptavidin-agarose beads overnight at 4°C. Beads were washed (twice with RIPA lysis buffer, once with 1 M KCl, once with 0.1 M Na2CO3, once with 2 M urea in 10 mM Tris-HCl,pH 8.0, and twice with RIPA lysis buffer) and pull-down proteins were submitted for mass spectrometry.
Peptides from streptavidin purified proteins were released via on-bead digestion using trypsin (Promega) and desalted on C18 STAGE Tips [PMID: 12585499]. Eluted peptides were dried and re-suspended in 5% formic acid (FA). Mass Spectrometric analysis was performed on a Thermo Orbitrap Fusion mass spectrometer equipped with an Easy nLC-1000 UHPLC. Peptides were separated with a 115 min gradient of 5–25% acetonitrile in 0.1 % FA at 350 nl/min and introduced into the mass spectrometer by electrospray ionization as they eluted off a self-packed 40 cm, 100 µm (ID) column packed with 1.8 µm, 120 Å pore size, C18 resin (Sepax Technologies, Newark, DE). The column was heated to 60°C. Peptides were detected using a data-dependent method. For each 3 sec cycle, high resolution MS1 scans performed in the orbitrap and the most abundant ions per were isolated, fragmented by CID, and scanned in the ion trap. Ions selected for MS2 analysis were excluded from reanalysis for 60 sec. Ions with +1 or unassigned charge were also excluded from analysis. MS/MS spectra were matched to peptide sequences using COMET (version 2019.01 rev. 5) (PMID: 23148064) and a composite database containing the 20,415 Uniprot reviewed canonical predicted human protein sequences (http://uniprot.org, downloaded 5/1/2019) and its reversed complement. Search parameters allowed for two missed cleavages, a mass tolerance of 25 ppm, a static modification of 57.02146 Da (carboxyamidomethylation) on cysteine, and a dynamic modification of 15.99491 Da (oxidation) on methionine. Peptide spectral matches (PSMs) were filtered to 1% FDR using the target-decoy strategy65 and then to 1% protein FDR. Label-free quantification was performed using peptide intensities from the integrated areas under each corresponding extracted-ion-chromatogram peak. Intensities for all peptides mapping to each protein were summed for each sample.
Mass spectrometry data analysis
PD-L1APEX2 data were analyzed in two different ways. In one, proteins that satisfied the following criteria were identified to be in the proximity of PD-L1: proteins with peptides in the experimental sample without peptides in the negative control (−H2O2) samples, or proteins in which peptides in the experimental sample were two-fold or greater than in the control sample. Data from 5 independent experiments were analyzed individually and the final PD-L1 proximity map was determined as those proteins satisfying the criteria for PD-L1 proximity in each of the five experiments, yielding 399 PD-L1 proximity proteins. This set of proteins was used for further analyses, including biological processes pathway analysis via geneontology.org (GO database released 7-2-2021)66 (as shown in Fig. 3D). The data were also analyzed by performing a t-test on Log10 transformed spectral intensity values between labeled and negative control samples (−H2O2) on the set of 1647 proteins common to all five experiments. Significant differences between the experimental and control were determined using Benjamini-Hochberg correction for multiple testing. Fold change was calculated as ratio of average spectral intensity in labeled samples over average in negative control samples (as shown in Fig S2A). The two different methods yielded essentially the same group of proteins.
For whole cell extract mass spectrometry analysis, sum intensity in each sample was confirmed to be similar across all samples in each experiment. For each protein, average ratio of +/- Durvalumab between two biological replicate experiments was calculated. Proteins with an average +/- Durvalumab ratio >1.2 (top 25th percentile of ratios across three cell lines) and coefficient of variance below 0.1 (bottom 25th percentile) in H1650 whole cells extract were analyzed for function across biological processes using geneontology.org (GO database released 7-2-2021)66. For EV mass spectrometry analysis, ratio of +/- Durvalumab between two biological replicate experiments was calculated for each protein and plotted as a frequency distribution.
Migration and proliferation
For scratch assays 35K cells were plated on each side of the Ibidi 2-well scratch assay chambers. Cells were allowed to attach for 24 hrs and wound healing was assayed in complete or serum-free media for KO/rescue cells and Durvalumab treated cells respectively at specified timepoints. Cell were imaged in bright-field on Leica Microscope. 9 images of each sample were taken at each timepoint, and three distances were measured in each image using Image J. For migration assay, 30k cells were plated on the top chamber of transwell-migration plates with 8 micron pore size in serum free media. Bottom chamber was filled with DMEM with 10% FBS and cells were allowed to migrate for 24 hrs. Cells were then fixed in 3.7% formaldehyde for 15 minutes, washed and then stained in Crystal Violet for 20 minutes. Unmigrated cells were scraped off with damp cotton swab and membrane was cut out and mounted on slides. Migrated cells were imaged on upright microscope (Zeiss Axioplan) with 10x objective and analyzed via ImageJ. For proliferation assay, 5k cells were plated in triplicate 96 wells (day 0) and MTT was used to quantify viable cells on days 1-5. Paired t test was performed on raw transwell migration data, and slopes of nonlinear regression lines were used to calculate significance for scratch and MTT assays.
Author contributions
AL designed and performed the experiments, analyzed the data, and wrote the manuscript.
EY designed experiments, generated the KO cells, and wrote the manuscript.
PC and ND performed and assisted with APEX2 mass spectrometry experiments and analysis
NKA conceptualized the project, contributed to experimental design and interpretations, and edited the manuscript.
TEM conceptualized the project, contributed to experimental design, performed analysis and interpretation, and wrote the manuscript.
Acknowledgements
We are grateful to Dr. Raffaella Sordella (Cold Spring Harbor Laboratories) for generously sharing EGFR mutant cell lines with us. We thank Astra Zeneca for the gift of Durvalumab antibody. We thank Dr. Alberto Benito and Dr. Kristy Brown for help with EV purification. We thank Rosemary Leahey for assistance with imaging studies. We thank the Optical Microscopy & Image Analysis Core at Weill Cornell Medicine for assistance with confocal microscopy. We thank Guoan Zhang and the Weill Cornell Medicine Proteomics and Metabolomics Core for assistance with whole cell and EV proteomics. We thank Dr. Jeremy Dittman, Dr. Geoffrey Markowitz, Dr. Lucie Yammine, Dr. Anuttoma Ray, Jennifer Wen, and Emma Johnson for helpful discussions and critical reading of the manuscript. The project was supported in part by NCI UG3 CA244697 (NKA, TEM), the Yoram Cohen family foundation (NKA), Vicky and Jay Furhman family fund (NKA) and a WCM Meyer Cancer Center pilot grant (NKA,TEM). TEM is also supported by DOD LC180227. EY is supported by a diversity supplement to CA244697.
Footnotes
Disclosures: NKA has equity in Angiocrine Bioscience. TEM receives research funding from Janssen and from Pfizer, Inc. The other authors have nothing to disclose.
References
