PD‐L1 on large extracellular vesicles is a predictive biomarker for therapy response in tissue PD‐L1‐low and ‐negative patients with non‐small cell lung cancer

Abstract

Immunotherapy has revolutionized the treatment of patients with non‐small cell lung cancer (NSCLC). High expression of tissue PD‐L1 (tPD‐L1) is currently the only approved biomarker for predicting treatment response. However, even tPD‐L1 low (1‐49%) and absent (<1%) patients might benefit from immunotherapy but, to date, there is no reliable biomarker, that can predict response in this particular patient subgroup. This study aimed to test whether tumour‐associated extracellular vesicles (EVs) could fill this gap. Using NSCLC cell lines, we identified a panel of tumour‐related antigens that were enriched on large EVs (lEVs) compared to smaller EVs. The levels of lEVs carrying these antigens were significantly elevated in plasma of NSCLC patients (n = 108) and discriminated them from controls (n = 77). Among the tested antigens, we focused on programmed cell death ligand 1 (PD‐L1), which is a well‐known direct target for immunotherapy. In plasma lEVs, PD‐L1 was mainly found on a population of CD45⁻/CD62P⁺ lEVs and thus seemed to be associated with platelet‐derived vesicles. Patients with high baseline levels of PD‐L1⁺ lEVs in blood showed a significantly better response to immunotherapy and prolonged survival. This was particularly true in the subgroup of NSCLC patients with low or absent tPD‐L1 expression, thus identifying PD‐L1‐positive lEVs in plasma as a novel predictive and prognostic marker for immunotherapy.

1. INTRODUCTION

Lung cancer is still the leading cause of cancer‐related death worldwide (Sung et al., 2021). For patients with advanced non‐small cell lung cancer (NSCLC) lacking driver mutations, immune checkpoint inhibitors (ICI) such as antibodies against programmed cell death protein 1 (PD‐1) and programmed cell death ligand 1 (PD‐L1) have revolutionized state‐of‐the‐art lung cancer treatment. The combination of immunotherapy with platinum‐based chemotherapy results in improved overall survival (OS) compared to standard platinum‐based chemotherapy alone, both in adenocarcinoma (Gandhi et al., 2018) and squamous cell carcinoma (Paz‐Ares et al., 2018), partly irrespective of tumoral PD‐L1 expression. Furthermore, NSCLC patients with high PD‐L1 expression on tumour cells (tumour proportion score, TPS) or on tumour‐infiltrating immune cells (IC) (combined positive score, CPS) benefit from single‐agent immunotherapy compared to standard platinum‐based chemotherapy (Reck et al., 2016). Moreover, dual checkpoint inhibition combined with chemotherapy has recently been shown to significantly prolong OS in patients with advanced‐stage NSCLC compared to chemotherapy alone (Johnson et al., 2022; Paz‐Ares et al., 2021). These trials underline the promising future for ICI, which is nowadays commonly included in standard‐of‐care lung cancer treatment.

To date, tissue PD‐L1 (tPD‐L1) expression as determined via immunohistochemistry is the only approved biomarker for immunotherapy. However, the predictive value of tPD‐L1 is limited (Shen & Zhao, 2018) and up to now there is no biomarker that can efficiently predict response to ICI in patients with absent or low tPD‐L1. In line, tPD‐L1 expression has been shown to differ between the primary tumour and nodal metastases (Uruga et al., 2017) which questions the reliability of tissue biopsy‐based PD‐L1 scoring. Biomarkers from liquid samples such as blood could solve this problem, as they promise an all‐embracing tissue profiling. Moreover, gathering such is easy and can be performed rapidly and repeatedly, thus allowing real‐time monitoring of therapy response prior to radiographic staging. Extracellular vesicles (EVs) are such promising liquid biopsy‐based biomarkers. These small membrane particles are released by all living cells and mediate intercellular communication by carrying proteins, lipids, and nucleic acids (e.g., DNA, RNA) from the secreting to the surrounding cells (Yáñez‐Mó et al., 2015). So far, two distinct EV populations have been described: small EVs (sEVs, formerly known as “exosomes”) with a diameter between 50–150 nm and large EVs (lEVs, formerly known as “microvesicles”) with a diameter between 100–1000 nm. sEVs are formed inside the cell by inward budding of endosomal membranes, while lEVs bud directly from the plasma membrane. Numerous studies have demonstrated that tumour‐derived sEVs are crucial in establishing a favourable tumour microenvironment and thus paving the way for metastatic spread (reviewed in (Kalluri, 2016)). However, a growing number of studies have attributed the same tumour‐supporting functions to lEVs. Despite the difference in cargo of sEVs and lEVs (Lischnig et al., 2022), tumour‐derived lEVs have been shown to increase the proliferation and invasion of neighbouring cancer cells (Arendt et al., 2014; Ciardiello et al., 2019; Menck et al., 2015). Moreover, they have been implicated in the reprogramming of stroma and immune cells (Castellana et al., 2009; Menck et al., 2013; Timaner et al., 2020), thus creating a favourable tumour microenvironment that has been associated with enhanced metastasis formation in mice (Pfeiler et al., 2019).

Due to the high amount of EVs shed by tumour cells into peripheral blood, using EVs as innovative cancer biomarkers is promising in clinical routine testing. Due to their bigger size, lEVs are easy to isolate and analyze with common diagnostic tools, such as flow cytometry (Menck et al., 2017). In this study, we aimed to investigate a panel of tumour‐related antigens for their expression on plasma‐derived lEVs and their potential as prognostic and predictive biomarkers for ICI‐based therapy in NSCLC.

2. MATERIALS AND METHODS

2.1. Patients

Samples were collected from cancer patients with confirmed lung cancer (n = 108), non‐cancer controls (n = 23) and healthy individuals (n = 54). Patient and control characteristics are summarized in Table 1+2. All human samples were obtained after informed consent as approved by the local ethics committees of the Medical Center Göttingen (approval no. 3/2/14) and of the Medical association Westfalen‐Lippe (approval no. 2020‐172‐b‐S). Samples for EV isolation were obtained prior to treatment to minimize possible contamination with apoptotic bodies. Treatment response was evaluated by computed tomography (CT) or magnetic resonance imaging (MRT) scans routinely performed between 12 to 24 weeks after baseline and analyzed according to the Response Evaluation Criteria in Solid Tumours (RECIST) version 1.1 (Eisenhauer et al., 2009). We defined responders as patients that achieved either complete remission (CR), partial remission (PR) or stable disease (SD). Non‐responders were defined as patients who showed progressive disease (PD) or succumbed to their disease. The expression of tPD‐L1 expression was collected from routine pathological reports of tumour tissue obtained via medically indicated biopsies or surgeries and was determined by immunohistochemistry using the PD‐L1 28‐8 antibody clone (Abcam) on the Ventana BenchMark staining platform. At least 100 tumour cells were analyzed for the determination of the TPS.

TABLE 1.

Characteristics of the study cohort.

Subgroups (incl. diagnoses of non‐cancer controls)		Age [years] Median [95% CI]	Male sex
NSCLC	108	64 [62‐66]	0.60
CTLh	54	33 [29‐37]	0.44
CTLnc	23	61 [53‐68]	0.57
Abscess of the left pulmonary artery	1
Atypical adenomatous hyperplasia of the lung	1
Bilateral lung transplant	1
Chronic obstructive pulmonary disease with lung emphysema	1
Diabetes mellitus	1
Idiopathic pulmonary arterial hypertension (group 1.1)	1
Interstitial lung disease • Exogenous allergic alveolitis • Chronic organizing pneumonia • Interstitial pneumonia with autoimmune features • Idiopathic pulmonary fibrosis	6 1 1 1 2
NSCLC in complete remission after R0 surgery • Adenocarcinoma, stage Ia, CR since 2019 • squamous cell carcinoma, stage IIa, CR since 2015	2 1 1
Sarcoidosis	9

TABLE 2.

Patient subgroups.

NSCLC patients
Stage	I—IIIA IIIB—IV unknown	6 101 1
Histology	Adeno SqCC Other non‐SqCC	92 10 6
Smoking history	Positive Negative Unknown	81 13 14
Molecular subtypes
KRAS	Wt Mut Unknown	51 29 28
BRAF	Wt Mut Unknown	80 3 25
EGFR	Wt Mut Unknown	81 12 15
TP53	Wt Mut Unknown	4 6 98
ROS	Neg Pos Unknown	89 0 19
ALK	Neg Pos Unknown	85 4 19
tPD‐L1	<1 1–49 ≥50 Unknown	34 30 28 16

2.2. Cells and EV isolation

We have submitted all relevant data of our EV isolation and characterization experiments to the EV‐TRACK knowledgebase (EV‐TRACK ID: EV230372) (Van Deun et al., 2017). Human NSCLC cell lines (ATCC, DSMZ) were cultured in RPMI‐1640 supplemented with 10% heat‐inactivated (56°C, 30 min) fetal calf serum (FCS) at 37°C and 5% CO₂. All cultured cells were routinely tested to exclude contamination with Mycoplasma. Platelets from healthy donors were provided by the Department of Transfusion Medicine (University Hospital Münster). To isolate EVs, cells (6xT175 flasks at a confluence of 60–80%) were washed twice with PBS and cultured for 24 h in RPMI‐1640 supplemented with 10% EV‐depleted FCS (centrifuged for 16 h at 153,700 g, 4°C and filtered through a 0.2 μm filter). The supernatants were collected and centrifuged at 500 g for 5 min and 1,500 g for 15 min to remove residual cells and debris. lEVs were pelleted at 17,000 g for 30 min and sEVs at 143,000 g for 90 min. EVs were washed once in PBS and used for downstream analyses. EV isolation from up to 15 mL of EDTA‐anticoagulated blood was performed using differential ultracentrifugation as described previously (Menck et al., 2017). All EV pellets were washed once in PBS and stored in PBS for subsequent experiments.

2.3. Density gradient centrifugation

Antigen association with EVs was analyzed on a discontinuous, top‐down iodixanol gradient which was performed based on the protocol described in (Van Deun et al., 2014) with slight modifications. Briefly, a working solution of OptiPrep™ (Sigma) in working solution buffer (0.25 M sucrose, 6 mM EDTA, 60 mM Tris‐HCl, pH 7.4) was used to prepare 5, 10, 20 and 40% iodixanol solutions in homogenization buffer (0.25 M sucrose, 1 mM EDTA, 10 mM Tris‐HCl, pH 7.4). The solutions were layered on top of each other and overlaid with the respective EVs (200 μg for cell culture‐derived EVs, or 300 μg for plasma‐derived EVs) in 1 mL PBS. The gradient was centrifuged for 18 h at 100,000 g, 4°C using a Sw32.1Ti rotor in a XPN‐80 ultracentrifuge (Beckman Coulter). Sixteen fractions of 1 mL were collected and washed once in PBS for 1 h at 100,000 g using a TLA‐55 rotor in a Max‐XP ultracentrifuge (Beckman Coulter). Pellets were resuspended in Laemmli buffer and subjected to immunoblot analysis. To measure the density of the fractions, a standard curve of the 5, 10, 20 and 40% iodixanol solutions was prepared and the absorbance of the fractions of a control gradient overlaid with PBS w/o EVs was measured in 1:4 dilutions in water at 340 nm using an Infinite M Nano microplate reader (Tecan).

2.4. Electron microscopy

EVs were pelleted as described above and fixed in 2.5% glutaraldehyde in Sörensen phosphate buffer. The specimens were post‐fixed with 1% osmium tetroxide, dehydrated and embedded in Epon. Sixty‐nanometer ultrathin sections were cut on an UltraCut E ultramicrotome (Leica) and counterstained with uranyl acetate and lead. Samples were inspected on an EM CM10 transmission electron microscope (Phillips).

2.5. Nanoparticle tracking analysis (NTA)

EV size and concentration were measured on a ZetaView PMX‐120 device equipped with a 640 nm laser and a CMOS camera (Particle Metrix). Samples were diluted in PBS to obtain a concentration of 50–400 particles/frame. For each sample, videos at 11 cell positions were recorded at 25°C with a camera sensitivity of 80–83 for sEVs and 76–79 for lEVs, respectively. Data was analyzed with the ZetaView software (v8.02.31).

2.6. Flow cytometry

lEVs (2.5 μg) were blocked in 20 μL PBS + 1% EV‐depleted FCS for 20 min and incubated with fluorescently‐labelled antibodies for CA9 (#130‐123‐340, 50 ng/sample), TROP2 (#130‐115‐097, 50 ng/sample, both from Miltenyi Biotec), CD147/EMMPRIN (#306207, 200 ng/sample), CD227/MUC‐1 (#355603, 200 ng/sample), CD274/PD‐L1 (#374511, 50 ng/sample), CD279/PD‐1 (#329905, 50 ng/sample), CD326/EpCAM (#324208, 30 ng/sample), EGFR (#352903, 100 ng/sample), ROR1 (#357803, 2.5 ng/sample), CD62P (#304910, 100 ng/sample), CD45 (#304006, 100 ng/sample, all from BioLegend), ROR2 (#FAB20641G, 12.5 ng/sample, R&D systems), or the corresponding isotype controls at the same concentration (#400321, #400132, #400113 from BioLegend, #IC003G from R&D systems or #130‐118‐347 from Miltenyi Biotec) for 20 min at RT. Signals were detected at the Attune NxT flow cytometer (Thermo Fisher) and analyzed with FlowJo (v10.7, BD). Due to the limited amount of EVs in some samples, it was not possible to measure all markers in every sample. To exclude the measurement of platelet signals in plasma‐derived lEV samples, the lEV gate was defined in relation to a control sample of whole platelets. The percentage of positive events in the lEV gate was determined in relation to the respective isotype control. The submicron bead calibration kit (#832, Bangs Laboratories) was used as size standard for flow cytometry experiments. For multi‐colour flow cytometry stainings, gates were set based on fluorescence minus one control. Flow cytometer acquisition settings were maintained for all samples, including triggering threshold, voltages, and flow rate.

2.7. Western blotting

Cells and EVs were lysed in RIPA buffer (50 mM Tris, 150 mM NaCl, 0.1% SDS, 0.5% sodium deoxycholate, 1% Triton X‐100, pH 7.2) supplemented with protease (Sigma) and phosphatase (Roche) inhibitors. Protein concentration was measured by Lowry assay (#5000112, Bio‐Rad) according to the manufacturer's instructions. Up to 17 μg of protein were separated by SDS page (8‐12% gels) and blotted onto a nitrocellulose membrane. Unspecific binding sites were blocked with TBST (137 mM NaCl, 20 mM Tris pH 7.6, 0.1% (v/v) Tween‐20) + 3% bovine serum albumin for 1 h at RT and membranes were incubated with primary antibodies against Actinin‐4 (#sc‐390205, 1:1000), Rgap1 (#sc‐271110, 1:1000), Mitofilin (#sc‐390707, 1:1000), Arf6 (#sc‐7971, 1:1000), β‐Actin (#sc‐47778, 1:2000), ApoA1 (#sc‐376818, 1:1000), ApoB (#sc‐393636, 1:1000), Albumin (#sc‐271605, 1:1000), MUC1 (#sc‐7313, 1:500), ROR2 (#sc‐98486, 1:1000), EMMPRIN (#sc‐21746, 1:250), CA9 (#sc‐365900, 1:1000, all from Santa Cruz), GM130 (#12480, 1:1000), HDAC (#2062, 1:1000), ROR1 (#16540, 1:1000, all three from cell signalling), EGFR (#18986‐1‐AP, 1:1000), PD‐L1 (#66248‐1‐Ig, 1:1000), TROP2 (#27360‐1‐AP, 1:1000, all from Proteintech), EpCAM (#ab71916, 1:1000), or Syntenin‐1 (#ab133267, 1:2000, both from Abcam) overnight at 4°C followed by incubation with suitable HRP‐coupled secondary antibodies (#7076 or #7074, 1:10,000, cell signalling) for 1 h at RT. Chemiluminescence was detected on a ChemoStar imager (Intas).

2.8. Statistics

All experiments were carried out in n ≥ 3 biological replicates unless indicated otherwise. Analyses were performed using SPSS (v 28.0.0.0) and GraphPad Prism (v9.2.0). Statistical significance was calculated with a Mann‐Whitney test for direct comparison of two groups and a Kruskal‐Wallis‐test for comparison of more than two unmatched groups, followed by uncorrected Dunn's test for multiple comparisons. Correlation was examined using the Spearman correlation. For evaluation of the prognostic potential, the receiver operating characteristic (ROC) curve was generated and the area under the curve (AUC) with 95 % confidence interval (CI) was assessed by the Wilson/Brown method. In survival analyses events were defined as death, as all can be considered as cancer‐related in our patient cohort, and data were analyzed with a Log‐rank (Mantel‐Cox) test for univariate analyses, Hazard Ratio (HR) (Mantel‐Haenszel) was examined and visualized using Kaplan–Meier plots. Cox regression was used for uni‐ and multivariate analyses for OS. In case no categories are listed, continuous values were used. P‐values < 0.05 were considered significant (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, ns: not significant). No imputation was carried out for missing data.

3. RESULTS

3.1. NSCLC cell line shed lEVs that express tumour‐related antigens

When aiming to detect specific lEVs within the multitude of EVs in the bloodstream, it is necessary to identify respective antigens that define the subgroup of interest. In order to identify potential tumour‐related antigens for lEVs secreted by NSCLC cells, we isolated EVs from five different NSCLC cell lines with distinct molecular subtypes (A549, HCC‐78, NCI‐H596, NCI‐H1975, NCI‐H2228) by differential ultracentrifugation. NTA indicated the existence of two distinct EV subgroups that differed in size, sEVs and lEVs (Figure 1a, Suppl. Figure 1A‐C). The median EV size was 186.1 nm [95% CI: 173.1‐197.9 nm] for lEVs and 155.4 nm [95% CI: 142.3‐164.8 nm] for sEVs. Transmission electron microscopy (TEM) was performed on EVs isolated from H2228 and H596 cells and confirmed the presence of vesicular structures. Although the comparison of the EV sizes measured by NTA and TEM is difficult due to vesicle shrinkage during sample preparation for TEM on the one hand (van Niel et al., 2018) and the challenges of correct NTA‐based sizing of polydisperse particle solutions on the other hand (Anderson et al., 2013; Comfort et al., 2021; Gardiner et al., 2013), both methods on their own confirmed the size difference between both EV subpopulations (Figure 1a+b, Suppl. Figure 1D for close‐up images).

FIGURE 1.

Flow cytometry identifies a panel of tumour antigens expressed on lEVs from NSCLC cell lines. (a+b) Small EVs (sEVs) and large EVs (lEVs) from H2228 and H596 cells were characterized by NTA (a) and transmission electron microscopy (b). Close‐up electron microscopy images of single EVs are provided in Suppl. Figure 1A. (c) Western blot: Expression of common positive and negative EV markers for the two distinct EV populations. The same amount of protein was loaded in every lane (10 μg for upper blot, 15 μg for lower blot). (d) Analysis of lEVs by flow cytometry. Shown is a representative FSC versus SSC plot with the lEV gate used for analysis. PBS only and size bead controls were used to confirm the specificity of the selected gate. (e) Flow cytometry: A two‐fold dilution series of H2228‐lEVs was prepared and the number of events in the lEV gate was recorded during 30 sec of measurement to exclude swarm detection (mean ± SD, n = 3). (f+g) Flow cytometry: Expression of selected tumour antigens (f) as well as PD‐1/PD‐L1 (g) on lEVs from five NSCLC cell lines. Blue: Tumour antigen, grey: isotype control. Shown is one representative histogram out of n = 2 experiments. (h) Expression of PD‐1 was confirmed on lEVs from OCI‐Ly13.2 lymphoma cells. Shown is one representative histogram out of n = 2 experiments.

While sEVs stained positive for Syntenin and CD81, the microvesicle markers Mitofilin and Rgap1 were specifically detected on lEVs (Figure 1c, Suppl. Figure 1E). Furthermore, compared to sEVs, lEVs contained higher levels of Actinin‐4 (for H596, H2228, HCC‐78, H1975) and Arf6 (for H2228, H1975), whereas the two markers were equally detectable in both EV subpopulations in the other NSCLC cell lines. Of note, A549 cells, which also secreted the lowest number of EVs (Suppl. Figure 1B), generally showed only minimal marker expression on their lEVs and lacked expression of Rgap1. We did not detect presence of the nuclear protein HDAC1 in any of the EV preparations and only slight traces of the Golgi protein GM130 in the lEVs of H596 and H1975 cells, which excludes significant contaminations of our preparations with intracellular vesicles. EV samples from all five cell lines were negative for Albumin as well as ApoA1 and there was only a very weak signal of ApoB in one lEV preparation of A549 cells (Suppl. Figure 1F). Taken together, the characterization revealed, that for all five NSCLC cell lines there was a distinctive expression of several EV markers between lEVs and sEVs indicating that these are indeed two different EV populations that can be separated with the protocol used in this study.

We then focused our analyses on lEVs, as they are faster and easier to isolate and analyze by flow cytometry, making them ideal biomarkers for clinical routine. To ensure specific detection of lEVs by flow cytometry, we used beads with pre‐defined sizes to set the gate for lEV analysis. In comparison to cell line‐derived lEVs, measurement of PBS or antibody alone as negative controls resulted in a neglectable number of events in the lEV gate (Figure 1d, Suppl. Figure 2A). Likewise, treatment of lEV samples with the detergent Triton X‐100 led to a significant loss of lEVs and fluorescent signal, thus confirming that the detected events are indeed of vesicular origin. To exclude swarm detection of aggregated lEVs, we prepared a two‐fold dilution series of lEVs from H2228 cells and recorded the number of events in the lEV gate during 30 sec of measurement (Figure 1e, Suppl. Figure 2B). As we were able to recapitulate a dilution factor of around 0.5 for all dilutions, this excluded that our flow cytometry measurements are influenced by swarm detection.

To establish a marker panel for the detection of tumour‐associated lEVs in human plasma, we screened the lEVs isolated from the five distinct NSCLC cell lines by flow cytometry for the expression of ten different membrane‐bound antigens (EMMPRIN/CD147, EGFR, MUC1/CD227, EpCAM/CD326, CA9, TROP2, PD‐L1/CD274, PD‐1/CD279, ROR1, ROR2) known to be overexpressed or involved in NSCLC (Figure 1f) (Ahmed et al., 2021; D'Incecco et al., 2015; Kim et al., 2004; Menck et al., 2021; Zamay et al., 2017; Zhang et al., 2017). While EMMPRIN was highly and equally expressed on all cell line‐derived lEVs, the expression of the other markers, such as ROR1 or MUC‐1, was restricted to certain cell lines, which indicated a subtype‐specific expression pattern. PD‐L1 was detected on lEVs from all five NSCLC cell lines, albeit its expression on lEVs from A549 cells was quite low (9 or 15.8%, respectively) (Figure 1g). PD‐1 was not found on NSCLC‐derived lEVs (Figure 1g). However, this seems to be attributed to the missing expression of PD‐1 in NSCLC cells, as it was readily detectable on lEVs isolated from the T cell lymphoma cell line OCI‐Ly13.2 (Figure 1h).

3.2. Tumour‐associated antigens are differentially expressed on lEVs and sEVs

At present, most EV biomarker studies have focused on sEVs as potential candidates. However, their use comes with several technical challenges due to their small size. To test whether lEVs might represent an alternative, we compared the expression level of our selected antigens between both EV populations, lEVs and sEVs. Characterizing the expression levels on the same amount of EV protein and cell lysate (Figure 2a), we observed that all markers were detectable on lEVs and sEVs, albeit with a cell line‐specific distribution pattern. In general, lEVs were enriched in EMMPRIN (in 5/5 cell lines) as well as MUC1 (in 3/4 positive cell lines), while ROR1 showed higher expression on sEVs (in 4/5 cell lines). No clear enrichment pattern was found for EpCAM, EGFR, TROP2, ROR2, or CA9. In comparison to ROR1, the expression level of ROR2 in NSCLC cells seemed to be rather low. MUC1 and TROP2 were not detectable in A549 cells or EVs, which is in line with the flow cytometry results (Figure 1f). While earlier studies (Chen et al., 2018) had described higher expression of PD‐L1 on melanoma cell line‐derived sEVs, we detected a distinctive enrichment of PD‐L1 on lEVs of H596, H2228, H1975 and HCC‐78 cells. In comparison, the PD‐L1 levels on the respective sEVs were rather low (Figure 2a). Only the A549 cell line, which showed a very low PD‐L1 expression per se in concordance with the flow cytometry results and earlier studies (Hinterleitner et al., 2021; Kim et al., 2019), carried more PD‐L1 on sEVs compared to lEVs.

FIGURE 2.

Tumour antigens are highly expressed on lEVs. (a) Expression of the selected nine tumour antigens was compared in cell lysates, lEVs or sEVs from five NSCLC cell lines. Actinin‐4 was included as lEV marker. The same amount of protein was loaded in every lane (20 μg for first blot, 18 μg for second blot, 25 μg for third blot, 17 μg for forth blot). (b) lEVs from the indicated NSCLC cell lines were isolated by differential ultracentrifugation according to the standard protocol and loaded on top of an OptiPrep density gradient (n = 2). The distribution of EV signals and contaminants in the 16 collected fractions was assessed by Western blot. Complete fractions, or 15 μg of protein for cell lysate and plasma sample, respectively, were loaded onto the gel. Actinin‐4 was included as marker for lEVs. Albumin was used to detect serum protein contaminations.

As EV preparations can be contaminated with protein aggregates or lipoprotein particles, we performed density gradient centrifugation to confirm the association of the selected tumour‐associated antigens with EVs. To this end, lEVs and sEVs were pelleted from the conditioned medium of H596 cells by differential ultracentrifugation and loaded without any washing step onto an OptiPrep density gradient. This setup was chosen to allow the presence of residual culture medium in the EV preparations and thus visualize successful separation of EVs from serum contaminants for method establishment. Analysis of the collected fractions by immunoblot confirmed the presence of sEVs and lEVs in fractions 8–13 (Suppl. Figure 3), which corresponds to a density of 1.10‐1.17 g/ml and fits to the values reported for EVs (Brennan et al., 2020). It also confirms that lEVs and sEVs have similar densities and cannot be directly separated from the culture medium using density gradient centrifugation. Ponceau staining indicated protein signals in fraction 1–5 with a density of 1.05‐1.07 g/ml. Since these values are in the range of low‐density lipoproteins (LDLs) (1.019−1.063 g/ml) (Brennan et al., 2020) and we additionally detected signals for the LDL component ApoB in fraction 2+3, these observations suggest that the gradient successfully separated EVs from LDLs. No signals were detectable for ApoA1, or Albumin, in any of the fractions (Suppl. Figure 3). Of note, for EGFR and Actinin‐4 weak signals were detected in the first fractions of the gradient, suggesting that these two proteins might also be associated to some extent with non‐EVs compartments. Indeed, soluble extracellular fragments of both proteins have been previously detected in other studies (Coppinger et al., 2004; Maramotti et al., 2016).

Subsequently, we analyzed the EV populations isolated with our standard protocol (including the additional PBS washing step to deplete the residual culture medium and contaminants) on the density gradients. Analysis of the lEVs from H596, HCC‐78 and H2228 cells by density gradient centrifugation confirmed the co‐localization of all tumour‐associated antigens with the EV fractions 8–13 (Figure 2b). Only ROR2 was not detectable due to its generally low expression level in the selected cell lines. For Actinin‐4 as well as MUC1, EpCAM and TROP2 weaker signals were also detectable in the first fractions of the gradient indicating, that for these markers an additional association with non‐EV particles cannot be excluded. Taken together, these results revealed the entirety of tumour‐related antigens present on lEVs, some of them even at significantly higher levels compared to sEVs.

3.3. Total lEV counts are unchanged in peripheral blood of NSCLC patients compared to controls

Using our previously established protocol to isolate EVs from plasma (Menck et al., 2017), we collected vesicles from NSCLC patients (n = 108) and healthy individuals (n = 54) as well as non‐cancer controls (n = 23) (Table 1+2). Using electron microscopy, NTA and immunoblot characterization, we confirmed the successful isolation of sEVs and lEVs from plasma samples (Figure 3a‐d). lEVs showed a more heterogeneous size distribution, were larger than sEVs and expressed the common microvesicle markers Actinin‐4 and Rgap1. To exclude a significant contamination of the lEV samples with plasma lipoproteins, we additionally stained our lEV preparations for ApoB and ApoA1, the major components of low or high density lipoparticles, respectively (Figure 3c+d). While we detected small traces of ApoB in some of the investigated lEV samples, ApoA1 expression was absent, arguing against a significant contamination of lEVs with plasma lipoproteins.

FIGURE 3.

NSCLC patients do not show differences in the number or size of lEVs in plasma compared to controls. (a) TEM of lEVs and sEVs isolated from the plasma of a NSCLC patient. The image on the left displays a wide‐field overview of the sample. The panel on the right contains a close‐up image of single vesicles. (b) Representative NTA measurements from plasma‐derived lEVs and sEVs. (c+d) Common lEV markers and contaminants were analyzed on lEVs isolated from NSCLC patients (c) or healthy controls (CTL_h) (d) by western blot. Cell lysates from two NSCLC cell lines as well as a plasma sample are shown as controls. 17 μg of protein were loaded in every lane. (e) NTA: Concentration and size of lEVs in plasma of healthy controls (CTL_h), non‐cancer controls (CTL_nc) and NSCLC patients (median ± 95%CI).

Some studies have reported an increase in the total number of EVs in the blood of cancer patients, including NSCLC (Choi et al., 2020; Fleitas et al., 2012). To state this finding in our patient cohort, we measured lEV and sEV concentration and size in NSCLC patients and controls by NTA (Figure 3e, Suppl. Figure 4). However, our results revealed no increase in total lEV or sEV levels or size in NSCLC patients compared to both control groups.

3.4. lEV‐associated tumour antigens are diagnostic biomarkers in NSCLC

To further investigate the potential of our tumour antigen signature and distinguish NSCLC patients from healthy controls, we measured the number of lEVs carrying the selected antigens in plasma samples from patients and controls by flow cytometry (Figure 4a+b, Suppl. Table 1). Since plasma EV samples have been reported to be contaminated with residual platelets (Lacroix et al., 2012), we used control samples of healthy human platelets to define the gate for lEV analysis and exclude that the measured signals originate from cells instead of EVs. Interestingly, we found a significant increase in the levels of PD‐L1⁺ and EMMPRIN⁺ lEVs in NSCLC patients compared to healthy individuals as well as non‐cancer controls, underlining their tumour‐specific role. EMMPRIN and PD‐L1 positively correlated (Suppl. Table 2). EGFR⁺, ROR2⁺, MUC1⁺ and ROR1⁺ lEVs in cancer patient samples were only significantly elevated compared to healthy, or non‐cancer controls, respectively. In particular, MUC1⁺, but also EpCAM⁺ and CA9⁺, lEVs were also prominent, albeit at statistically non‐significant levels, in the non‐cancer control group, pointing at a possible association of these three antigens with lung diseases other than cancer (Menck et al., 2017). We did not observe significant differences in the levels of CA9⁺, TROP2⁺, PD‐1⁺ or EpCAM⁺ lEVs in NSCLC patients compared to controls. To confirm the association of the six tumour antigens significantly upregulated in NSCLC compared to one of the control groups (PD‐L1, EMMPRIN, MUC1, ROR1, ROR2, EGFR) with EVs also in plasma samples, the EV preparations of two NSCLC patients were analyzed on an OptiPrep density gradient. While we observed some signals for ApoA1, Albumin, Actinin‐4 and EGFR in fraction 1–5, the expression of PD‐L1, EGFR, ROR2 and EMMPRIN co‐localized with the lEV markers Mitofilin and Actinin‐4 in fraction 12+13 for patient 1, and/or fraction 8–11 for patient 2 (Suppl. Figure 5). MUC1 and ROR1 were not detectable in the two selected patients which might be due to the fact that they are not universally expressed on the lEVs of all patients (Figure 4b).

FIGURE 4.

Tumour antigen‐loaded lEVs are elevated in the blood of NSCLC patients. (a) Representative scatter plots of plasma‐derived lEVs and platelets. Only events in the lEV gate were used for further analysis. (b) The percentage of lEVs positive for the indicated tumour antigen was measured by flow cytometry in healthy controls (CTL_h), non‐cancer controls (CTL_nc) and NSCLC patients using the gate defined in A. Boxes mark the 25−75 percentiles (line at median) and whiskers the 5–95 percentile. (c) Western blot of EMMPRIN and PD‐L1 expression in lEVs isolated from NSCLC patients and healthy controls. Lysates from H2228 and H596 cells as well as a plasma sample were included as controls. 17 μg of protein were loaded in every lane. (d+e) ROC curves were used to determine the discriminative power of significantly elevated lEV tumour antigens alone (d) or all six combined (e).

As the two antigens EMMPRIN and PD‐L1 showed the highest increase on lEVs from NSCLC patients compared to controls, we further confirmed this enrichment by immunoblotting (Figure 4c). To discriminate NSCLC patients from healthy controls, we evaluated the potential of all six elevated markers (PD‐L1, EMMPRIN, MUC1, ROR1, ROR2, EGFR) by ROC analyses (Figure 4d, Suppl. Table 3). EMMPRIN reached the best AUC of 0.75 (95% CI: 0.66‐0.84). However, the combination of all six markers was clearly superior and resulted in an AUC of 0.80 (95% CI: 0.69‐0.90) (Figure 4e, Suppl. Table 3). Taken together, these analyses indicated the potential of our lEV‐associated tumour antigen signature as diagnostic biomarker for the detection of NSCLC in patients.

3.5. PD‐L1⁺ lEVs in NSCLC patients seem to originate from platelets

PD‐L1 is known to be expressed not only on tumour cells, but also on the surface of all antigen‐presenting cells as well as on platelets (Rolfes et al., 2018; Yu et al., 2016). As was shown before, the majority of lEVs in blood are derived from platelets and leukocytes (Menck et al., 2017). Hence, we asked whether the striking increase of PD‐L1⁺ lEVs stemmed from the presence of tumour‐lEVs or from increased levels of lEVs secreted from platelets or leukocytes. By comparing the blood cell counts in the samples used for lEV isolation, we observed that the numbers of leukocytes and platelets were significantly elevated in NSCLC patients compared to healthy individuals and non‐cancer controls (Suppl. Figure 6). In contrast, there was a decrease in red blood cell counts in non‐cancer controls as well as NSCLC patients compared to healthy controls. Moreover, the level of EMMPRIN⁺ as well as PD‐L1⁺ lEVs was positively correlated with platelet, but not leukocyte, counts (Suppl. Table 4).

We therefore continued to characterize the lEVs secreted by both cell types by flow cytometry using the lymphoma cell line HDLM2 as model for leukocytes and outdated platelet concentrates as a source for platelet lEVs. The pan‐leukocyte antigen CD45 served as a marker for leukocyte‐derived lEVs and CD62P for platelet‐derived lEVs (Menck et al., 2017). The results of the characterization revealed that both cell types indeed expressed rather high levels of PD‐L1 and EMMPRIN on their vesicles, but carried none of the other NSCLC‐associated antigens used in this study (Figure 5a+b), except for a high expression of EGFR on lEVs from HDLM2 cells, which might be associated with the malignant origin of the cell line.

FIGURE 5.

PD‐L1 is enriched on a population of CD62P⁺/CD45⁻ lEVs in the plasma of NSCLC patients. (a+b) Flow cytometry: Expression of the selected tumour antigens on lEVs from platelets (a) or HDLM2 lymphoma cells (b). CD62P was included as a marker for platelet‐derived lEVs and CD45 for leukocyte‐derived lEVs. Blue: Antigen of interest, grey: isotype control. Shown is one representative histogram out of n = 3 (Platelets) or n = 2 (HDLM2) experiments. (c) The percentage of CD62P⁺ and CD45⁺ lEVs was measured by flow cytometry in healthy controls and NSCLC patients. Boxes mark the 25−75 percentiles (line at median) and whiskers the 5–95 percentile. (d+e) PD‐L1⁺ lEVs were analyzed for their expression of CD62P and CD45. Shown is one representative scatter plot of a healthy control and NSCLC patient (d). The number of PD‐L1⁺/CD62P⁻/CD45⁺ and PD‐L1⁺/CD62P⁺/CD45⁻ lEVs was counted in both groups (e). Boxes mark the 25−75 percentiles (line at median) and whiskers the 5–95 percentile.

Analysis of the number of platelet‐derived, CD62P⁺ lEVs in plasma samples of our study cohort indeed showed a significant increase in NSCLC patients compared to healthy controls, whereas the numbers of leukocyte‐derived, CD45⁺ lEVs were comparable in both groups (Figure 5c). Co‐staining of PD‐L1, CD62P and CD45 revealed that most PD‐L1⁺ lEVs were CD62P⁺, thus potentially originating from platelets. In line, PD‐L1⁺/CD62P⁺/CD45⁻ lEVs were significantly elevated in the plasma of NSCLC patients (Figure 5d+e). In contrast, the levels of PD‐L1⁺/CD62P⁻/CD45⁺ lEVs showed no difference between patients and controls. Fluorescence minus one controls used for flow cytometry gating are depicted in Suppl. Figure 7. Hence, our results suggest that a major part of the increased PD‐L1⁺ lEVs in NSCLC patients originates from patients’ platelets and not leukocytes.

3.6. PD‐L1 on lEVs is a prognostic biomarker in NSCLC

We next investigated whether the increase of PD‐L1⁺, EMMPRIN⁺, EGFR⁺, MUC1⁺, ROR1⁺ and ROR2⁺ lEVs in the plasma of NSCLC patients also influences patient survival. In contrast to our previous study, which had found a prognostic value of lEV‐bound EMMPRIN for cancer patients harbouring different solid tumours (Menck et al., 2017), we did not detect an influence of lEV‐associated EMMPRIN on patient survival in the NSCLC patient cohort. Moreover, NSCLC patients with high numbers of PD‐L1⁺ lEVs in blood even showed a significantly better OS (Figure 6a). Univariate analyses also showed a trend to confirm this effect for PD‐L1 (p = 0.066) but not for the other markers on lEVs (Suppl. Table 5). Multivariate analyses identified OS to depend on age and a trend for depending on the level of PD‐L1⁺ lEVs (p = 0.083) (Suppl. Table 6). Interestingly, we did not observe a prognostic effect for the expression of tPD‐L1 in our cohort (Suppl. Figure 8A). Of note, a subgroup analysis of PD‐L1⁺ lEV levels in patients for factors that would influence patient survival, including histology, smoking status, or molecular alterations in KRAS, EGFR, or ALK, did not indicate any statistical significant differences between mutated and wild type specimens (Suppl. Figure 9). To explain the unexpected positive effect of high PD‐L1 on lEVs for patient survival, we asked whether high levels of lEVs carrying PD‐L1 might be associated with better response to therapy.

FIGURE 6.

lEV‐associated PD‐L1 predicts therapy response in patients with absent and low tissue PD‐L1 expression. (a) Kaplan–Meier survival curves of NSCLC patients according to the number of tumour antigen‐positive lEVs in blood. (b) The level of PD‐L1⁺ lEVs in blood of treatment‐naïve patents at initial diagnosis were correlated with response to therapy at the first staging CT after 3 months (R/NR₃) or 6 months (R/NR₆). Based on the staging CT, patients were stratified as responders (R) if they showed complete or partial response or stable disease and as non‐responders (NR) if they showed signs of progressive disease. The upper panel comprises all patients independent of the choice of therapy, the lower panel displays only patients with immunotherapy in their therapeutic regimen. (c+d) Correlation of tPD‐L1 (c) and TPS (d) with the level of PD‐L1⁺ lEVs in plasma. (e) NSCLC patients were stratified according to their tissue PD‐L1 (tPD‐L1) expression. Levels of PD‐L1⁺ lEVs before chemo‐immunotherapy (CIT) alone or in combination with mono‐immunotherapy (ICI) were compared in patients classified as R or NR at the first staging CT after 3 or 6 months of treatment. (f+g) ROC analysis comparing the predictive power of either TPS, tPD‐L1 and PD‐L1⁺ lEVs (f) or of PD‐L1⁺ lEVs in patients grouped by their tPD‐L1 levels (g).

3.7. PD‐L1⁺ lEVs predict therapy response in patients with absent tissue PD‐L1 levels

Data on therapy response were available for n = 60 patients (all stage IV) in our study cohort. Patients had either received chemo‐immunotherapy (CIT, n = 37), mono‐immunotherapy (Mono‐ICI, n = 14) or targeted therapy (TT, n = 9). Response to therapy was evaluated in all 60 patients based on CT scans performed after 3 months and after 6 months according to RECIST version 1.1 (Eisenhauer et al., 2009). Patients were classified as responders if they were diagnosed with a CR, PR or SD and as non‐responders if they showed signs of PD or had succumbed to their disease. Due to a very good response to mono‐immunotherapy, two patients subsequently underwent primary tumour resection via lobectomy within the first six months in an oligometastatic setting. Response to therapy was associated with a significantly improved patient survival (Suppl. Figure 8B). When comparing the initial levels of PD‐L1⁺ lEVs in treatment‐naïve patients, we observed significantly higher levels in responders compared to non‐responders. Of note, the increase was more prominent for the early response to therapy evaluated after 12 weeks, than for the durable response evaluated after 24 weeks. The effect was seen both for the group of all patients (CIT, Mono‐ICI, TT) as well as for the subgroup of patients receiving therapy that included immunotherapy (CIT, Mono‐ICI) (Figure 6b). No significant changes were observed for the levels of lEV‐associated EMMPRIN, EGFR, MUC1, ROR1 or ROR2 in responders compared to non‐responders (Suppl. Figure 8C).

At present, tPD‐L1 is used in routine clinical diagnostics as selection criteria to predict response to immunotherapy. However, also advanced NSCLC patients with low or absent tPD‐L1 levels (<49% or <1%, respectively) can benefit from immunotherapy (Shen & Zhao, 2018) and so far no predictive biomarker exists for this patient subgroup. Interestingly, neither the tPD‐L1 levels nor the tumour proportion score (TPS), defined as percentage of PD‐L1‐positive cancer cells in relation to all tumour cells in the tissue sample, correlated with the levels of PD‐L1 on lEVs in the same patients (Figure 6c+d). In particular, many of the patients with low or absent tPD‐L1 levels showed significantly elevated levels of PD‐L1⁺ lEVs in blood. Next, we grouped patients according to their tPD‐L1 levels and compared the lEV PD‐L1 levels in responders and non‐responders to CIT or ICI. Surprisingly, we detected significantly higher levels of PD‐L1⁺ lEVs in responders versus non‐responders in the group with absent tPD‐L1 expression (Figure 6e). Similar results were observed for patients with low tPD‐L1 expression (1‐49%), although it did not reach statistical significance due to the low number of non‐responders in this subgroup. In contrast, there was no difference in patients with high (>49%) tPD‐L1 status (Figure 6e). Comparing the predictive power of lEV‐associated PD‐L1 with the so far discussed scores tPD‐L1 or TPS revealed a superior AUC value of 0.79 for lEVs compared to 0.53, or 0.57, respectively (Figure 6f). When subdividing NSCLC patients by their tPD‐L1 status, we observed that the predictive power was in particular based on the PD‐L1‐absent (AUC 0.91; p = 0.01) and low (AUC 0.90; p = 0.09) group, while the predictive power of PD‐L1⁺ lEVs in tPD‐L1‐high patients was rather low (AUC 0.57; p = 0.64) (Figure 6g). Taken together, our results have thus identified PD‐L1 on plasma lEVs as a novel biomarker that can predict response to immunotherapy in NSCLC patients with low or absent tPD‐L1 expression. Its predictive value does not require dynamic lEV sampling to assess expression changes under therapy, but is valid already at baseline.

4. DISCUSSION

Immunotherapy has revolutionized lung cancer treatment. However, the decision to allocate patients to immunotherapy is hampered by the lack of biomarkers that can reliably identify patients that benefit from this therapy. Novel diagnostic approaches using blood or tissue tumour mutational burden have recently failed to predict response to ICI across all cancer types (McGrail et al., 2021; Langer et al., 2019). Other studies have identified circulating sEVs in plasma as promising predictive biomarkers for immunotherapy (Cordonnier et al., 2020; de Miguel‐Perez et al., 2022; Del Re et al., 2018). Next to the methodological challenges associated with sEV analysis, their predictive value is based on repetitive sampling over several weeks and therefore requires a prolonged timespan for response prediction. In order to solve these problems, in this study we have focused on a specific EV subpopulation, the lEVs, in plasma and investigated ten distinct tumour‐associated antigens expressed on these vesicles for their suitability as prognostic and predictive biomarkers for immunotherapy in NSCLC. We demonstrated that a selected panel of these antigens is elevated in patients, reliably discriminates them from controls and has prognostic value for patient survival. In particular, in patients with low or absent tPD‐L1 expression, high levels of PD‐L1⁺ lEVs in plasma are associated with response to immunotherapy and therefore are a novel predictive biomarker for this patient subgroup. For prediction of therapy response, the levels of PD‐L1⁺ lEVs can be measured in a simple single‐staining approach from a liquid biopsy obtained at baseline.

Nowadays, it is well‐accepted that several EV subpopulations exist that differ in size, cellular origin and content (Yáñez‐Mó et al., 2015). Compared to the widely‐used sEVs, the analysis of the larger plasma membrane‐derived lEVs comes with the advantage of a more rapid isolation protocol applicable to standard lab equipment and the possibility to measure antigen expression by flow cytometry, a technique that is well‐established in clinical diagnostics and easy to standardize. Still it is controversially discussed, whether the total level of lEVs in blood of NSCLC patients is a solely diagnostic or even a predictive biomarker. While one study has reported an increase in lEV concentration in lung cancer patients (Fleitas et al., 2012), our own results in this cohort as well as in a previous study (Menck et al., 2017) have not found such a difference. Since total blood EV concentration in plasma is influenced by various factors such as hypoxia, inflammation, exercise, diet, and age (Alique et al., 2017; Ayers et al., 2015), these confounding factors should be carefully taken into account when considering quantitative differences in EV numbers.

Data on changes in plasma EV composition in NSCLC patients are largely lacking. In our study, we therefore aimed to define antigens that are related to tumour growth, are highly expressed on NSCLC‐derived lEVs and could serve as prognostic or predictive biomarkers in this tumour entity. Interestingly, we detected a trend for decreased levels of TROP2⁺ lEVs in NSCLC patients compared to healthy controls. While Lin and colleagues had observed a decrease of TROP2 expression in lung adenocarcinoma (Lin et al., 2012), most studies have reported an increased expression of this marker in NSCLC (Ahmed et al., 2021), which was the reason for including this marker in our tumour‐related antigen panel. However, the comparatively high levels of TROP2⁺ lEVs in the plasma of both control groups indicate, that the detected vesicles are not derived from cancerous but from benign cells. Indeed, TROP2 expression is not specific for tumour cells but has been detected in several normal cell populations, including a subset of cells in the bone marrow (Dum et al., 2022; Stepan et al., 2011). Although the quantification of TROP2 levels on normal blood cells is currently still missing, these observations suggest that the reduction of TROP2⁺ lEVs in NSCLC patients most probably arises from alterations in the composition of the non‐tumour lEV compartment in blood.

Our flow cytometric analyses identified six antigens that were significantly elevated in NSCLC patients compared to controls. The combination of all six markers proved to be a reliable diagnostic marker to identify NSCLC patients (AUC 0.80). Previously, Lobb et a prognostic sEV signature that combined four markers (MAC2BP, PSMA2, THBS1, TNC) and yielded an even higher AUC of 0.96 (Lobb et al., 2023). However, next to the requirement of an extensive and time‐consuming protocol for sEV isolation by density gradient ultracentrifugation which is not feasible in large‐scale clinical diagnostics, the patient cohorts of both studies differed substantially, as Lobb et al. tested the diagnostic potential of their sEVs on NSCLC stage I‐IIIA patients with front‐line surgery in a resectable setting and curative intent, while our work was focused on NSCLC stage IV patients with palliative intended front‐line CIT. Although the comparison of both signatures is challenging due to the different study cohorts, both studies confirmed that a single biomarker has no sufficient diagnostic power and that the definition of suitable multi‐biomarker signatures can significantly improve sensitivity of detection.

Among our six cancer‐related antigens, EMMPRIN and PD‐L1 showed the most striking increase, both compared to healthy individuals as well as diseased non‐cancer controls. A similar observation has already been made for PD‐L1 in melanoma patients (Chen et al., 2018). While EMMPRIN has previously been described as a marker for lEVs (Menck et al., 2015), the expression of PD‐L1 on this EV subclass has been neglected so far (Poggio et al., 2019; Yang et al., 2018) or was described as significantly lower compared to sEVs in the case of melanoma (Chen et al., 2018). In NSCLC, the vesicular PD‐L1 distribution pattern seems to differ as we found highly elevated levels on the lEVs of all four PD‐L1‐positive cell lines compared to sEVs. Classically, PD‐L1 is known to be present on immune cells (Han et al., 2020). However, recent reports have detected not only PD‐L1⁺ circulating tumour cells in cancer patients (Ilié et al., 2018) but also PD‐L1‐carrying platelets which presumably obtain the protein by direct cell‐cell contact with PD‐L1⁺ tumour cells (Hinterleitner et al., 2021; Rolfes et al., 2018). Surprisingly, in our study the majority of PD‐L1⁺ lEVs stained positive for the activated platelet marker CD62P pointing at an origin from this cell population. In order to exclude that the detected PD‐L1 signals arose from residual platelets in our lEV preparations, we used a very stringent gating strategy and only included EVs which were clearly distinct in size from a platelet control. Of note, previous studies on platelet‐associated PD‐L1 have reported, that high platelet PD‐L1 levels were associated with poorer patient survival, KRAS‐mutant NSCLC and smoking history (Hinterleitner et al., 2021; Rolfes et al., 2018). In contrast, we did not observe these correlations for PD‐L1⁺ lEVs in our study. Indeed, NSCLC patients with high PD‐L1⁺ lEV levels were characterized by improved overall survival, in line with their better response to therapy. Taken together, these observations support the assumption that the PD‐L1 signals detected in our study do not arise from platelets.

Cancer cells can export a major portion of their PD‐L1 onto EVs (Poggio et al., 2019), which is discussed as a possible reason for the missing correlation between the levels of tissue and EV‐associated PD‐L1 observed in many studies (Cordonnier et al., 2020; Li et al., 2019; Signorelli et al., 2020), including our own. Other explanations could be that tPD‐L1 does not adequately reflect the heterogeneous expression of PD‐L1 in the tumour tissue or in different tumour sites (Ben Dori et al., 2020). Interestingly, circulating tumour cells have also been found to be more often positive for PD‐L1 than the tumour tissue in the same patient supporting this hypothesis (Darga et al., 2021; Guibert et al., 2018).

To date, the only fully clinically implemented and FDA‐approved biomarker to predict response to immunotherapy is tPD‐L1 expression, assessed by immunohistochemistry. However, a large meta‐analysis proved that tPD‐L1 alone is insufficient to determine which patients respond to immunotherapy since patients with low (1‐49%) or even absent (<1%) tPD‐L1 expression might benefit from ICI (Shen & Zhao, 2018). This is supported by our data unravelling PD‐L1⁺ lEVs as superior in predicting response to immunotherapy (AUC 0.79) compared to tPD‐L1 (AUC 0.53) or TPS (AUC 0.57). This effect was particularly driven by the group of patients with low or absent tPD‐L1 expression, which achieved very high AUC values of 0.90 or 0.91, respectively. On the one hand, the comparably low predictive power of PD‐L1⁺ lEVs in tPD‐L1‐high patients could be concealed by the generally better response of this patient subgroup to (immuno)therapy. On the other hand, the effect could arise from differences in the shedding of PD‐L1 from tumour cells onto EVs between tPD‐L1‐high versus –low/‐negative patients, or from differences in the immune landscape of the blood vesiculome. The few studies that have so far addressed EVs as predictive biomarkers have focused on the dynamics of PD‐L1 changes on sEVs to stratify patients as responders or non‐responders to anti‐PD‐1‐therapy in melanoma (Chen et al., 2018; Del Re et al., 2018) or NSCLC (Cordonnier et al., 2020; de Miguel‐Perez et al., 2022; Del Re et al., 2018). Thus, they required repetitive sampling and a prolonged time span before the response to therapy could be predicted. In contrast, we demonstrated that the initial level of PD‐L1⁺ lEVs in treatment‐naïve patients is sufficient to predict response to immunotherapy and outperforms the current standard‐of‐care. Interestingly, melanoma patients with high levels of circulating PD‐L1⁺ sEVs showed poorer response to immunotherapy and worse clinical outcomes (Chen et al., 2018; Serratì et al., 2022), while we observed the opposite in NSCLC. In line with our results, also a study measuring PD‐L1 mRNA levels in sEVs of NSCLC patients reported higher baseline PD‐L1 copy numbers in responders compared to non‐responders to ICI (Del Re et al., 2018). A comparative study on PD‐L1 in both tumour entities has not only observed lower tumour cell PD‐L1 expression in melanoma compared to NSCLC, but also found PD‐L1 to be preferentially localized to the non‐tumour immune cell compartment in melanoma patient samples (Kluger et al., 2017). This raises the question whether the different associations of vesicular PD‐L1 levels with therapy response arise from differences in the origin of the PD‐L1⁺ EVs. In vitro data from NSCLC cell lines indicated that sEV‐associated PD‐L1 can impair T cell function (Kim et al., 2019) as it has been described for melanoma (Chen et al., 2018) suggesting that the immunosuppressive function of EVs carrying PD‐L1 is comparable in both entities. Whether the observed differences between the studies therefore originate from different biological effects of large and small EVs or from different microenvironments and mechanisms in NSCLC and melanoma is unclear at present. Obviously, prospective multicentre studies with larger patient cohorts are required to further validate the use of PD‐L1⁺ EVs as biomarkers and establish them in the clinic.

Taken together, our study has identified lEVs as novel diagnostic and prognostic biomarkers in NSCLC and demonstrated that a single measurement of PD‐L1⁺ lEVs by flow cytometry reliably predicts response to ICI in patients with absent tPD‐L1 expression.

AUTHOR CONTRIBUTIONS

Nadja Schöne: Data curation (equal); formal analysis (equal); investigation (equal); visualization (equal); writing—original draft (supporting); writing—review and editing (equal). Marcel Kemper: Data curation (equal); formal analysis (equal); resources (supporting); writing—original draft (supporting); writing—review and editing (equal). Kerstin Menck: Conceptualization (equal); funding acquisition (lead); investigation (equal); project administration (equal); supervision (equal); visualization (lead); writing—original draft (lead); writing—review and editing (equal). Georg Evers: Resources (supporting); writing—review and editing (equal). Carolin Krekeler: Resources (supporting); writing—review and editing (equal). Arik Bernard Schulze: Resources (supporting); writing—review and editing (equal). Georg Lenz: Resources (supporting); writing—review and editing (equal). Eva Wardelmann: Methodology (supporting); writing—review and editing (equal). Claudia Binder: Resources (supporting); writing—review and editing (equal). Annalen Bleckmann: Conceptualization (equal); project administration (equal); resources (lead); supervision (equal); writing—review and editing (equal).

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Supplementary Information

Schöne, N. , Kemper, M. , Menck, K. , Evers, G. , Krekeler, C. , Schulze, A. B. , Lenz, G. , Wardelmann, E. , Binder, C. , & Bleckmann, A. (2024). PD‐L1 on large extracellular vesicles is a predictive biomarker for therapy response in tissue PD‐L1‐low and ‐negative patients with non‐small cell lung cancer. Journal of Extracellular Vesicles, 13, e12418. 10.1002/jev2.12418