ALIX Regulates Tumor-Mediated Immunosuppression by Controlling EGFR Activity and PD-L1 Presentation

1Epidermal growth factor receptor (EGFR) is either amplified or mutated in a variety of cancers and contributes significantly to tumorigenesis. Basal-like breast cancer (BLBC), a subtype of disease with the worst therapeutic outcomes, commonly exhibits elevated EGFR expression. While the subject of intense research, the complexity and plasticity of the tumor EGFR signaling network may underlie the poor response to current EGFR-targeted therapies. Thus, there is a need to seek additional molecular targets for improving EGFR targeting in basal-like disease.

2Immune checkpoint blockade has introduced exciting possibilities to the field of targeted cancer therapy to inhibit tumor growth, with exploitation of programmed death-ligand 1 (PD-L1) showing great therapeutic promise. The recent demonstration of a direct association between activating EGFR mutations and increased PD-L1 expression in non-small-cell lung cancer has shown an interdependence between cancer cell-autonomous (EGFR-dependent cell survival) and non-autonomous (PD-L1-dependent evasion of immune surveillance) mechanisms of tumor survival. Consequently, tumors harboring activating EGFR mutations benefit from two distinct but linked survival pathways, and targeting of mutant EGFR can impair the immunosuppressive phenotype of these tumors. However, targeting of wild-type receptors in human lung cancer cell lines has little effect on PD-L1 expression, suggesting that the link between EGFR and PD-L1 expression is dependent on the kinase activity of the receptor. In BLBC, PD-L1 expression is frequently observed alongside elevated wild-type EGFR expression, suggesting that the immunosuppressive characteristics of these tumors may be refractory to the effects of therapies targeting EGFR’s kinase activity. A greater understanding of mechanisms driving EGFR signaling in breast cancer is, therefore, warranted; in particular, how EGFR activity in this disease affects immunosuppressive pathways mediated by PD-L1.

3Here, we performed a bioinformatics-led RNAi screen in BLBC cells to identify regulators of EGFR activity. Among proteins identified were exosomal cargo proteins and proteins implicated in exosome biogenesis. We found that cells lacking the endosomal sorting complexes required for transport (ESCRT) component ALIX, a critical mediator of exosome biogenesis, displayed enhanced EGFR activation, suggesting unexpected parallels between mechanisms of exosome biogenesis and regulation of EGFR activity. We found that PD-L1 is secreted on exosomes in an ALIX-dependent manner, and impaired exosomal release conferred an enhanced immunosuppressive phenotype on tumor cells that was dependent upon EGFR kinase activity. Our data suggest that downregulation of ALIX provides a mechanism for enhancing both EGFR activity and PD-L1-mediated evasion of anti-tumor immunity in BLBC, driving cell-autonomous and non-cell-autonomous mechanisms of tumor survival.

4Fluorescence resonance energy transfer (FRET)-based biosensors enable in situ monitoring of signaling pathway activities in cells. The Picchu-X FRET biosensor is based on the CrkII adaptor protein, a direct target of EGFR phosphorylation. Here, we use a modified version of Picchu-X that has been optimized for fluorescence lifetime imaging (FLIM) to monitor EGFR activity in HCC1954 BLBC cells.

5Background biosensor activity in HCC1954 cells was low, indicating a minimal basal EGFR activity (Figure 1A). We observed a stimulation-dependent and PD168393-sensitive enhancement of Picchu-X phosphorylation (Figure 1A) and FRET/FLIM (Figures 1B and 1C), demonstrating the biosensor’s specificity as a readout of EGFR activity. The relationship between mean donor lifetime and mean fluorescence intensity among untreated and EGF-treated populations was incorporated into the analysis to account for lifetime variations arising from differences in biosensor expression levels (Figure 1D). The analysis confirmed that the association between EGF treatment and lifetime was unaffected by the level of biosensor expression (Figure 1E). This stringent method of analysis was applied to all lifetime data acquired from the high-content small interfering RNA (siRNA) screen.

6We determined the effects of the targeted knockdown of 533 candidate proteins on EGFR signaling in situ to identify regulators of EGFR among a network of candidate proteins extracted from a bioinformatics-led analysis of protein-interaction databases (Figure S1). Twenty hits were identified that abrogated the biosensor response to ligand (Figure 2A, gray nodes; Figure 2B, cf. difference in the slopes of linear regression lines for EGF versus non-EGF treatment cell populations within the example “non-hit” and “hit” siRNA experimental groups; Table S1). Details of the 20 hits identified in the screen are provided in Table S1, and their known interactions with members of the EGFR subnetwork are illustrated in Figure S1 (pink nodes).

7Of the 20 proteins identified in our screen, 14 were identified as exosomal cargo proteins (Table S2), suggesting that changes in exosomal sorting can influence EGFR activation. There is increasing realization that cancer cells can exploit the exosomal pathway as a mechanism for driving tumor survival and dissemination. Consequently, we sought to identify proteins providing an interface between oncogenic EGFR and exosomal components in BLBC. The expression of 12 of these 14 exosomal-related proteins has been confirmed in normal and cancerous human breast tissues (Table S3). Of these, the adaptor protein ALIX was selected for further investigation, given its association with the exosomal cargo and intraluminal vesicle (ILV) biogenesis (Table S2).

8Using HCC1954 cells stably expressing ALIX short hairpin RNA (shRNA), we confirmed that stimulation-induced EGFR phosphorylation was both enhanced and prolonged in ALIX knockdown (KD) cells, when compared with non-targeting controls (NTCs; Figure 2C). Elevated levels of downstream signaling components (endogenous CrkII phosphorylation) were also observed in these cells following EGF stimulation (Figure 2D). Results were confirmed in SKBR3 breast cancer cells (Figure S2), demonstrating that this phenotype was not unique to HCC1954 cells.

9ALIX is implicated in the multivesicular body (MVB) sorting pathway of ubiquitinated membrane receptors. This pathway is responsible for cargo incorporation into ILVs, linking ALIX directly to cargo sorting and exosome biogenesis. Therefore, in addition to altering EGFR activity, ALIX KD may also affect the quality and composition of exosomes, with downstream consequences for tumor biology. Although procedures do not yet exist for the absolute purification of exosomes, sequential ultracentrifugation provides an established technique for the enrichment of extracellular vesicle (EV) subpopulations highly enriched with exosomes. Nanoparticle tracking analysis (NTA) revealed that ultracentrifugation-enriched particles from HCC1954 culture supernatants had a mean modal diameter of 86 nm, which lies within the range of 40–120 nm associated with exosomes (Figure 3A). Examination of these particles by transmission electron microscopy (TEM) revealed cup-shaped morphologies characteristic of exosomes, while immunogold labeling confirmed that they were positive for the exosome marker CD63 (Figure 3B). Western blotting confirmed the presence of the exosomal marker proteins TSG101, CD9, and ALIX (Figure 3C). Immunoblotting analysis revealed the enrichment and loss of positive and negative exosomal markers, respectively, in these ultracentrifugations. A visual analysis of extracellular vesicle preparations from HCC1954 cells stably expressing GFP-CD63 revealed a uniform field of fluorescent particles (Figure 3C), and a mass-spectrometry-based comparison of sequential ultracentrifugation and ExoQuick methods demonstrated that the centrifugation method was associated with a lower amount of extracellular protein contamination (Figure S3). Together, these data show that ultracentrifugation provides an effective method for the enrichment of exosomes from HCC1954 culture supernatants.

10To test whether ALIX depletion alters the exosomal proteome of HCC1954 cells, exosomes from the culture supernatants of control and ALIX KD cells were examined by quantitative mass spectrometry. Liquid chromatography-tandem mass spectrometry (LC-MS/MS) identified a total of 3,700 proteins in the exosomes of parental HCC1954 cells. These included the core exosomal marker proteins CD63, CD9, and CD81 and proteins known to be associated with exosome biogenesis and endosomal trafficking, such as ALIX, syntenin, syndecan, and Rab5 (Figure S3).

11Analysis of the exosomal proteome of ALIX KD cells revealed that 12 proteins were significantly downregulated and four were significantly upregulated, when compared with controls (Figure 3D). As expected, ALIX and syntenin (a cytoplasmic adaptor protein known to interact with ALIX as part of the exosome biogenesis machinery; ) levels were significantly diminished in exosomes derived from ALIX KD cells.

12We were intrigued to find that PD-L1, a protein of high clinical relevance because of its role in mediating tumor-associated immunosuppression, was significantly depleted from the exosomes of ALIX-suppressed cells. Basal PD-L1 expression in HCC1954 cells is elevated following cell stimulation with EGF or interferon (IFN)γ, the archetypal signaling factor associated with PD-L1 expression (Figures S4A and S4B). Exosomal PD-L1 incorporation correlates with cellular expression levels (Figure 4A), and we discovered that, while IFNγ treatment resulted in the robust induction and sustained expression of PD-L1 in both NTC and ALIX KD cells (Figure S4C), PD-L1 incorporation into exosomes was significantly reduced in ALIX KD cells (Figure 4B). Flow cytometry analysis of cell-surface PD-L1 expression in unstimulated and IFNγ-stimulated cells revealed that surface levels were elevated in ALIX KD cells (Figure 5A). These data suggest that exosomal release of PD-L1 occurs at the expense of surface PD-L1 levels and that ALIX controls the balance of receptor distribution between these membranes.

13Because basal levels of PD-L1 are almost undetectable in SKBR3 cells, and are unchanged following EGF or IFNγ treatment (Figure S5A), we used mouse KPB6 tumor cells, which show a robust PD-L1 response to IFNγ (Figure S5B), as an alternative model to confirm the PD-L1 surface expression phenotype. Elevated surface PD-L1 levels were confirmed in ALIX-depleted and IFNγ-stimulated KPB6 cells (Figure S5C).

14To gain insight into the mechanism underlying the reduced exosomal packaging and enhanced cell-surface expression of PD-L1 in ALIX KD cells, we examined the subcellular localization of PD-L1. Immunocytochemistry revealed that, in addition to its cell-surface localization, PD-L1 also localized to both the limiting membrane and ILVs of CD63-positive MVBs in IFNγ-treated HCC1954 cells (Figure 5B). This was in agreement with our detection of PD-L1 in the exosomes of IFNγ-treated cells, because ILVs are the intracellular precursors of these extracellular vesicles. In ALIX KD cells, a greater proportion of PD-L1 was found at the limiting membrane of MVBs, rather than within the endosomal lumen (Figure 5B). Because MVBs are small subcellular structures, the presence of discrete intraluminal cargo is difficult to discern by epifluorescence microscopy. We used Rab5 Q71L expression to promote endosomal enlargement, facilitating the distinction between limiting membrane and endosomal lumen, which confirmed the limiting membrane accumulation and luminal exclusion of PD-L1 in ALIX KD cells (Figure S6A). Interestingly, TEM analysis of MVBs from NTC and ALIX KD cells revealed that budding profiles (luminal invaginations of the limiting membrane of MVBs that are the precursors to ILVs) were more frequently observed in ALIX KD cells, when compared with controls (Figure S6B). When taken together with the immunofluorescence data, these findings suggest that there may be a defect in intra-endosomal budding associated with ALIX depletion. These data suggest that ALIX is required for incorporation of PD-L1 from the limiting membrane into ILVs, which would account for the diminished levels of exosomal PD-L1 and enhanced surface PD-L1 in ALIX-depleted cells.

15The observation that loss of ALIX expression is associated with reduced exosomal packaging of PD-L1 and elevated cell-surface expression raises the intriguing possibility that cells with reduced ALIX expression exhibit enhanced immunosuppressive properties. To test this hypothesis, NTC and ALIX KD cells were either untreated or stimulated with IFNγ for 24 hr and co-cultured with activated primary human peripheral blood CD3+ T lymphocytes to observe the effects of co-culture on lymphocyte proliferation. Flow cytometry analysis of T cells revealed an impairment in the proliferation of T cells co-cultured with ALIX-depleted tumor cells (Figure 5C). These findings demonstrate that ALIX depletion confers enhanced immunosuppressive properties on cancer cells.

16Given that PD-L1 expression increases following EGFR or IFNγ receptor activation, and given that ALIX both regulates EGFR activity and modulates the exosomal and cell-surface distribution of PD-L1, we sought to determine whether EGFR signaling integrated with the IFNγ:PD-L1 signaling axis by examining whether the expression and cellular distribution of PD-L1 induced by IFNγ treatment in NTC and ALIX KD cells was dependent upon EGFR. Treatment of ALIX KD cells with the EGFR inhibitor PD168393 prior to IFNγ stimulation both diminished the surface accumulation of PD-L1 (Figure 5A) and reduced the immunosuppressive effect of these cells in subsequent co-culture assays (Figure 5C). These data indicate that EGFR signaling contributes to the IFNγ- and PD-L1-dependent immunosuppressive response and suggest that ALIX integrates the signaling of two important regulators of tumor-mediated immunosuppression, modulating PD-L1 surface expression and the EGFR signaling required for the associated immunosuppressive phenotype.

17Our in vitro data suggest that modulation of ALIX expression provides a mechanism for regulating surface PD-L1 expression. Furthermore, we show that differences in ALIX expression translate into differences in immunomodulatory potency. To examine the relevance of this ALIX:PD-L1 signaling axis in human disease, we performed a tissue microarray analysis of PD-L1 protein expression on 189 tumor samples across different human breast cancer subtypes and combined this with ALIX gene expression analysis using mRNA isolated from matched tissue samples. Our analysis revealed a statistically significant inverse correlation between ALIX mRNA and PD-L1 protein expression in tumor cells (Figures 6A and 6B).

18We next wanted to evaluate the impact of ALIX loss on the anti-tumor immune response in vivo. For this, we used the well-characterized and established syngeneic BALB/c mouse model with the 67NR breast cancer cell line, which demonstrates robust induction of PD-L1 in response to IFNγ (Figure 6C).

19Ex vivo analysis of the relationship between lymphocytic invasion and the ALIX:PD-L1 expression ratio in parental 67NR tumors grown in the mammary fat pad of BALB/c female mice revealed a complex intratumoral relationship in which distinct regions of low ALIX:PD-L1 and high ALIX:PD-L1 expression ratio were associated with different CD3+T cell counts (Figure 6C). Low-ALIX:PD-L1 ratio pixels were less infiltrated by CD3+ T cells than the high-ALIX:PD-L1 ratio pixels within the same tumor (Figures 6C and 6D).

20To determine the effects of ALIX depletion on tumor growth and immunosuppression, we generated 67NR cell lines stably expressing control and ALIX shRNAs and used them to induce orthotopic subcutaneous tumors in mice. These in vivo studies showed that ALIX loss resulted in significantly enhanced tumor volume at day 14 (time of animal culling), when compared with control tumor-bearing mice (Figure 7A). Furthermore, an increased frequency of CD4+ regulatory T cells (Figure 7B) and a decreased frequency of Granzyme B-expressing, tumor-infiltrating CD4+ and CD8+ T cells (Figure 7C) were observed at day 14. Together, these results suggest that reduced ALIX expression contributes to the establishment of an immunosuppressive tumor microenvironment, leading to more aggressive tumor growth.

21These observations suggest that ALIX is a regulator of both the surface expression and immunomodulatory potency of PD-L1 in breast cancer, in part, through its regulation of EGFR.

22EGFR is abnormally and heterogeneously expressed in breast cancers, and identifying factors that can delineate patient subsets for more effective therapies is of significant clinical interest. EGFR and immune checkpoint combination strategies are currently being tested in clinical trials, and while PD-L1/PD1 pathway blockade is a promising therapeutic approach in oncology, our understanding of PD-L1 regulation remains incomplete. Future success of combination therapies will depend on better understanding of growth factor:PD-L1 signal crosstalk.

23We have shown that ALIX is a negative regulator of EGFR activity, that its depletion significantly augments IFNγ-induced PD-L1 surface expression in human breast cancer cells, and that this upregulation is EGFR activity dependent. We have, therefore, extended the receptor tyrosine kinase (RTK) and immune checkpoint molecular interdependence phenomenon, previously only observed in EGFR mutated lung cancers, to BLBC, where PD-L1 expression is frequently observed and where elevated EGFR expression is associated with poor survival.

24PD-L1 has been identified in exosomes isolated from urine or plasma but has not previously been detected in the exosomes from cancer cells. Using exosomal proteome profiling, we found that HCC1954 cells package PD-L1 into exosomes and that ALIX depletion impairs this exosomal incorporation. Correlating with impaired ILV and exosomal incorporation of PD-L1, ALIX-depleted cells exhibit enhanced surface levels of PD-L1, conferring an enhanced immunosuppressive phenotype on these cells. The change in partitioning of PD-L1 between the exosomal and cell-surface compartments was associated with an altered pattern of PD-L1 localization at the endosome; namely, a loss of luminal PD-L1 and an accumulation of this protein at the limiting membrane. We suggest that defective ILV incorporation of PD-L1 results in impaired exosomal packaging, and upon MVB-PM (plasma membrane) fusion, this results in an elevation of cell-surface PD-L1 that confers an enhanced immunosuppressive phenotype. It should be noted that, given its role in viral budding, ALIX may also contribute to the formation of plasma membrane microvesicles, which could contribute to the PD-L1 cell-surface phenotype. However, our immunocytochemistry data indicate that defective trafficking (limiting membrane to ILV) of PD-L1 within MVBs is likely to be a major contributor to this phenotype.

25Analysis of tissue microarrays from diverse human breast cancer subtypes revealed PD-L1 upregulation in tissues with low ALIX mRNA levels. Previous studies observed that PD-L1 expression is heterogeneous among different breast cancers and has a prognostic and predictive value in breast cancers. Among breast cancer subtypes, BLBC cells constitutively express the highest levels of PD-L1. Furthermore, PD-L1 overexpression is more prevalent in inflammatory breast cancers (a rare and particularly aggressive form of disease) that are ER (estrogen receptor) negative, basal, and ERBB2 enriched. Our data confirm that PD-L1 expression is higher in basal-like cancers and show that PD-L1 is upregulated in tumor tissues with low ALIX mRNA levels, independently of cancer subtype.

26Ex vivo quantitative immunofluorescence analysis of 67NR mouse tumor tissue sections revealed an intratumoral association between areas of low ALIX:PD-L1 expression ratio and a reduced CD3+ lymphocytic infiltration. While this association was significant, it should be noted that it was not observed uniformly, and it is likely that analysis was confounded by the presence of non-tumor cell types expressing ALIX and PD-L1. Spectral deconvolution of more cell-type-specific fluorescent markers will be required to delineate further this phenomenon of relative T cell exclusion from low-ALIX, high-PD-L1 tumor cells.

27The generation of stable ALIX-suppressed 67NR cells enabled us to address directly the pathophysiological consequences of ALIX depletion on tumor growth and immunosuppression. 67NR cells can be used with immunocompetent BALB/c mice as a syngeneic model of breast cancer and provide an ideal tool to study tumor–immune interplay in this disease. Our in vivo studies revealed that ALIX-KD tumors were both larger than control tumors and associated with an enhanced immunosuppressive phenotype that included elevated numbers of infiltrating T-regulatory cells and decreased numbers of granzyme-B-expressing, tumor-infiltrating CD4+ and CD8+ T cells.

28Taken together, our data strengthen the link between the tumor microenvironment and suppression of the immune system in human breast cancers by shedding light on the signaling interplay between EGFR, ALIX, and PD-L1. These data will open additional avenues for therapeutic strategies and accelerate biomarker discovery programs to optimize future combinations between EGFR and immune checkpoint targeting.

29Details of antibodies used in this study are provided in Table S4.

30Human HCC1954 and SKBR3 breast cancer cells were cultured in RPMI 1640 and DMEM, respectively. Mouse 67NR breast and KPB6 lung cancer cells were cultured in DMEM and IMEM, respectively. Culture media was supplemented with 10% fetal bovine serum (FBS), 2 mM L-glutamine, 100 U/mL penicillin, and 100 μg/mL streptomycin. Cell lines stably expressing NTC and ALIX shRNAs were generated by lentiviral transduction using the pGIPZ system (GE Healthcare UK, Buckinghamshire, UK). Stable cultures of virally traduced cells were established by puromycin selection.

31The custom-designed Silencer Select Human siRNA Library, comprising siRNAs for 533 human gene targets (three siRNAs per target), was from Applied Biosystems. siRNA dilutions, transfection complex preparation, and cell seeding were performed using a Janus automated workstation (PerkinElmer). Each siRNA was evaluated in triplicate, and each experiment was performed three times. Cells were seeded in the presence of transfection complexes (0.5 pmol siRNA; RNAiMAX reagent, Invitrogen, Carlsbad, CA, USA) and, after 24 hr, were transfected with the Picchu-X biosensor using FuGENE HD reagent (Promega, Fitchburg, WI, USA). After an additional 24 hr, cells were treated with or without 100 ng/mL EGF for 30 min before fixation and FLIM analysis. Control siRNA cells were included alongside target siRNA cells to enable pairwise comparisons. siRNAs against human PDCD6IP were purchased from Horizon Discovery (Cambridge, UK). PDCD6IP siRNAs #1 and #2 refer to siGENOME IDs D-004233-01-0002 and D-004233-06-0002, respectively. siGENOME Non-Targeting siRNA ID D-001210-02-05 was used as the control. Lentiviral shRNA vectors for the stable KD of human and mouse PDCD6IP were purchased from GE Healthcare UK. For stable KD of human PDCD6IP, Clone ID V2LHS_357889 was used. For stable KD of murine Pdcd6ip, Clone IDs V2LMM_177842 and V3LMM_450035 were used.

32FLIM was used to measure FRET between the donor and acceptor fluorophores of the biosensor in cells treated with or without EGF. Because high biosensor expression increases the probability of intermolecular FRET between the donor and acceptor fluorophores of neighboring proteins through molecular crowding, such effects were considered in the analysis of intramolecular FRET (desired component). Intermolecular FRET was determined by analyzing the lifetime of the donor fluorophore at increasing concentrations of biosensor expression (determined by fluorescence intensity). Pearson correlation was used to examine the correlation between lifetime and intensity. Regression analyses were further used to quantify associations between EGF treatment, intensity, and lifetime. Tests for an interaction incorporating the product term of EGF treatment and intensity were performed to assess whether effects of EGF on lifetime differed by intensity. A lack of significant interaction indicated that the association between lifetime and EGF treatment was not affected by intensity (Figure 1E). Analyses of covariance (ANCOVA) were performed to further discern the effect of EGF treatment, with intensity and EGF treatment as predictor variables and biosensor lifetime as the dependent variable. Lifetime and intensity data were normally distributed. All statistical analyses were performed with R v3.1.2 (R Project for Statistical Computing, Vienna, Austria). Two-tailed p values < 0.05 were considered statistically significant. In all cases, data are presented as mean ± SEM.

33The Picchu-X sensor for reporting EGFR kinase activity was a kind gift from M. Matsuda, Osaka University, Osaka, Japan. The CrkII-based sensing region was excised and inserted into the equivalent sensing portion of the previously described Raichu-Rac 1011-X sensor, resulting in the generation of the EGFP-CrkII-mRFP1 biosensor with a C-terminal membrane-targeting CAAX motif.

34HCC1954 cells at approximately 80% confluency were washed with PBS and then cultured for 24 hr in FBS-free media. Culture supernatants were then collected, and exosomes were enriched by sequential centrifugation. Briefly, supernatants were centrifuged at 300 × g for 10 min to remove cell debris, at 5,000 × g for 20 min to remove large vesicles and membrane fragments, at 12,200 × g for 60 min to deplete MVs, and then at 100,000 × g for 120 min to pellet exosomes. The pellet was washed in PBS and then centrifuged at 100,000 × g for an additional 60 min before resuspension in PBS. All steps were performed at 4°C. NTA was performed using a Nanosight LM10-HS (Nanosight) as described previously, using constant flow injection. The NTA analysis software was used to obtain information regarding the particle population size distribution including the derivation of the population’s modal particle diameter. Samples for mass spectrometry were prepared and analyzed as described previously. Further details of the procedure are provided in the Supplemental Experimental Procedures.

35Western blotting was performed according to standard protocols. Dot-blotting was used to examine protein content of EVs due to its improved sensitivity compared with western blotting and involved the immunodetection of proteins spotted onto nitrocellulose membranes (1 μg total protein in 5 μL). All primary antibodies used are listed in Table S4. Densitometry analysis of protein band (western blots) and spot (dot-blotting) intensities was performed using ImageJ software. Data reported in bar graphs were obtained from ≥3 independent experiments.

36TEM was performed on a Joel 1010 electron microscope (Joel, Warwickshire, UK). Exosomes were layered onto Formvar carbon-coated copper grids. Grids were washed with H2O, and then incubated with 2% uranyl acetate (negative stain) for 30 s. For immunogold labeling, exosomes layered onto Formvar carbon-coated nickel grids were fixed with 0.5% paraformaldehyde (PFA) for 4 min and then blocked with serum at room temperature for 10 min. Exosomes were stained overnight at 4°C with primary antibody against human CD63, washed three times in PBS for 10 min, and subsequently labeled with secondary antibody (10 nm colloidal gold, BioCell) at room temperature for 90 min. The immunolabeled exosomes were washed with PBS for 5 min, fixed in 1% glutaraldehyde, and then washed twice with PBS and once with H2O. Exosomes were negatively stained as described earlier before imaging.

37HCC1954 cells were transfected with ALIX siRNA (40 nM) using RNAiMAX. After 48 hr, cells were treated with IFNγ (20 ng/mL) for 24 hr. Cells were fixed with PFA and stained for CD63 and PD-L1. Nuclei were counterstained with DAPI. Cells were imaged on a spinning disk confocal microscope (Nikon Eclipse, teamed with the CSU-X1 Andor Spinning Disk with Neo sCMOS camera), acquiring sub-saturated images as z stacks with 0.3-μm Z-spacing. At least 10 fields of view per treatment were captured, and luminal versus limiting membrane staining of endosomal PD-L1 was scored by manually scrolling through Z.

38Peripheral blood mononuclear cells (PBMCs) from healthy donor blood were collected by sucrose gradient centrifugation using Ficoll (GE Healthcare). CD3+/CD25− cells were subsequently isolated using magnetic beads (Miltenyi Biotec), labeled with eFluor 450 cell-tracker dye, and then activated with CD3, CD2, and CD28 beads at the time of co-culture with tumor cells. At 96 hr post-co-culture, cells were recovered and stained with CD3, CD4, CD8, and PD-1 antibodies and then subjected to flow cytometry. Data were analyzed using FlowJo software (FlowJo, Ashland, OR, USA). The significance of statistical differences between treatment groups was determined using ANOVA (p < 0.05 was reported as significant).

39Experiments were performed in accordance with the UK Home Office Animals Scientific Procedures Act, 1986, and the United Kingdom Co-ordinating Committee on Cancer Research (UKCCCR) guidelines. Tumors were established by subcutaneous injection of 1 × 106 cells of the syngeneic mouse breast cancer cell line 67NR into the mammary fat pad of 6- to 8-week-old BALB/c female mice (Charles River UK). Tumor growth was monitored with calipers and recorded every other day for 2 weeks; then, mice were culled and primary tumors were dissected. Tumors were minced and incubated for 1 hr at 37°C in digestion media (RPMI containing 1 mg/mL type 2 collagenase and 0.1 mg/mL bovine pancreas DNase I) and passed through a 40-μm cell strainer to form single-cell suspensions for flow cytometry.

40Cell suspensions were stained with Zombie Aqua dye, followed by membrane staining with anti-CD45-AF700, CD3-PECy7, CD4-BV785, CD8-PerCpCy5.5, and CD25-PE; fixation and permeabilization; and intracellular staining with anti-Foxp3-eFluor 450 or Granzyme B-APC. Stained cell suspensions were analyzed by flow cytometry. Data were acquired in a Fortessa II flow cytometer (BD Biosciences) and analyzed with FlowJo software (FlowJo).

41Tissues and data were collected by the KHP Cancer Biobank with approval from the East of England – Cambridge East Research Ethics Committee, reference number 12/EE/0493. Tissue microarrays (TMAs) of 218 primary breast cancers from the METABRIC study were used in this analysis. After antigen retrieval using the Ventana BenchMark system (Ventana Medical Systems), tissues were stained with anti-PD-L1 antibody. A total of 189 TMA cores were imaged using an “open” high-content microscope, and image data were subsequently processed to determine protein expression. To find the percentage of PD-L1-positive cancer cells, we adopted a previously reported algorithm.

42For the evaluation of ALIX mRNA expression, Illumina HT-12 microarray expression data were obtained from all METABRIC samples and filtered to remove arrays with outlying low intensity (mean log2 expression < 5.6). Array data were then quantile-normalized, filtered for probe detection (required p < 0.01 for >1% of King’s METABRIC samples), and COMBAT-corrected for beadchips. Normalization was performed using the “beadarray” package for R/Bioconductor. ALIX (PDCD6IP) mRNA expression was inspected using Probe identifier ILMN_1693259. Gene expression data from the METABRIC study have been reported previously. The correlation between PD-L1 positivity and ALIX mRNA tissue expression was tested using GraphPad statistical analysis software.

43Tissue samples and data from patients were obtained from The King’s Health Partners (KHP) Cancer Biobank at Guy’s Hospital (London, UK; REC no.: 07/40874/131).