Platelet PD-L1 reflects collective intratumoral PD-L1 expression and predicts immunotherapy response in non-small cell lung cancer

Immune-checkpoint inhibitors (ICI) have transformed oncological therapy. Up to 20% of all non-small cell lung cancers (NSCLCs) show durable responses upon treatment with ICI, however, robust markers to predict therapy response are missing. Here we show that blood platelets interact with lung cancer cells and that PD-L1 protein is transferred from tumor cells to platelets in a fibronectin 1, integrin α5β1 and GPIbα-dependent manner. Platelets from NSCLC patients are found to express PD-L1 and platelet PD-L1 possess the ability to inhibit CD4 and CD8 T-cells. An algorithm is developed to calculate the activation independent adjusted PD-L1 payload of platelets (pPD-L1Adj.), which is found to be superior in predicting the response towards ICI as compared to standard histological PD-L1 quantification on tumor biopsies. Our data suggest that platelet PD-L1 reflects the collective tumor PD-L1 expression, plays important roles in tumor immune evasion and overcomes limitations of histological quantification of often heterogeneous intratumoral PD-L1 expression.

I mmune-checkpoint receptors like CTLA4 and PD-1 are crucial for preventing excessive immune responses and autoimmunity 1,2 . Seminal discoveries made by Allison and Honjo provided preclinical proof of concept data that blockage of CTLA4 and PD1 signaling unleashes marked anti-tumor immune responses 3,4 . Clinical evaluation revealed remarkable therapeutic potential of immune-checkpoint inhibition in human cancer patients and for the first time allowed for long term survival of patients with advanced metastasized solid tumors [5][6][7][8][9] . Besides melanoma patients, especially patients suffering from non-small cell lung cancer (NSCLC) benefit from treatment with antibodies inhibiting the PD1 and CTLA4 immune checkpoints [5][6][7] . Nevertheless, simple and robust biomarkers to predict therapy responses towards ICI are still missing.
With 1.8 million deaths per year, lung cancer represents one of the most frequent and lethal cancers worldwide 10 . In the US 254,170 new lung cancer cases are expected to be diagnosed in 2021 11 . Given the high frequency of lung cancer and the cost of checkpoint inhibitory therapies, the lack of robust biomarkers to select patients who best possibly benefit from ICI represents a major burden for our health systems. Histological quantification of intratumoral PD-L1 expression is routinely performed in an attempt to predict therapy responses towards ICI, however, only an insufficient correlation between detection of PD-L1 expression in tumor biopsies and the overall response rate (ORR) was found 12 . In lung cancer, evaluation of smoking history, tumor mutational burden (TMB), microsatellite instability (MSI), high expression of CTLA4, low expression of CX3CL1 and infiltration of CD8+ T cells within the tumor microenvironment (TME) seems to be superior in predicting therapy responses towards anti-PD-1/PD-L1 directed ICI [13][14][15] when compared to histopathological PD-L1 quantification, however these markers so far could not be translated into a robust and clinically easy to use biomarker signature.
Here, we show that platelets, during their frequent interaction with tumor cells, ingest PD-L1 and present it on their surface, a process which is dependent on fibronectin, α5β1 and GPIbα. PD-L1 expressing platelets are detected in the TME and peripheral blood of NSCLC patients. The functionality of platelet PD-L1 (pPD-L1) is confirmed by inhibition of CD4+ and CD8+ activity. pPD-L1 correlates with tumor stage/grade and the occurrence of metastases. We develop an algorithm allowing to calculate the total PD-L1 payload of platelets (pPD-L1 Adj. ) without the need of artifact prone in vitro stimulation procedures. Strikingly, in our study pPD-L1 Adj. is shown to be superior in predicting response to ICI when compared to immunohistochemistry-based quantification of PD-L1 on tumor biopsies.

Results
Tumor cells transfer PD-L1 to platelets. To address whether the immune regulatory protein PD-L1 can be transferred from tumor cells to platelets, we co-incubated platelets obtained from healthy donors with four different NSCLC tumor cell lines harboring varying expression levels of PD-L1 (NCI-H23, A549: PD-L1 low/ negative, NCI-H226, NCI-H460: PD-L1 positive) (Fig. 1a, b). PD-L1 positivity was determined by flow cytometry and defined as PD-L1 expression in ≥ 5% of all tumor cells. PD-L1 expression on platelets (pPD-L1) was observed after co-incubation with the PD-L1 expressing NCI-H226 and NCI-H460 cells but not after coincubation with the PD-L1 low/negative cell lines NCI-H23 and A549 (Fig. 1c-e). Results were validated using a flow cytometrybased approach (Fig. 1f). Co-incubation of platelets with all tumor cell lines resulted in platelet activation, as indicated by P-selectin (CD62P) induction (Fig. 1g), however only coincubation with PD-L1 positive NCI-H226 and NCI-H460 cells resulted in an increased PD-L1 expression on the platelet surface (Fig. 1h). To ensure that platelets from healthy donors, used in this assay, do not harbor relevant amounts of endogenous PD-L1 we conducted western blot analyses on platelet whole-cell lysates. Indeed, Western Blot data confirmed that platelets from healthy donors do not express relevant PD-L1 levels ( Supplementary  Fig. 5b).
Of note, conditioned medium from tumor cells induced platelet activation but did not result in increased levels of PD-L1 protein on the platelet surface ( Supplementary Fig. 3c, d), suggesting that PD-L1 transfer from tumor cells to platelets is dependent on a direct cell-cell contact between both cell types. Of note, frequent interaction with platelets was not restricted to adherent tumor cells but could for example also be observed for non-adherent A549 lung cancer cells ( Supplementary  Fig. 3e, f).
To gain deeper insights into the interaction of platelets and lung cancer cells, we took advantage of a live-cell imaging platform, where platelets are added to the medium and circulate through an imaging chamber that contains human NSCLC cells. Real time video microscopy revealed distinct interactions of tumor cells and platelets (Fig. 1i, j and Supplementary Movie 1). Strikingly, platelets remained fully agile and re-entered the circulation after contacting the tumor cell membrane (Fig. 1k-j and Supplementary Movies 1,2). These data suggest that platelets can re-circulate after tumor cell attachment and activation and are in line with studies by Cloutier and Michaelson et al. 16,17 .
While platelets are anuclear, protein translation from RNA can nevertheless occur within platelets [16][17][18] . We therefore set out to investigate whether PD-L1 expression in platelets depends on a transfer of PD-L1 protein from tumor cells to platelets or whether a transfer of PD-L1 mRNA with subsequent protein synthesis within the platelet is involved. Transfection of vectors encoding for PD-L1-GFP and FLAG-GFP fusion proteins into PD-L1 negative A549 cells (Fig. 1m-o) resulted in high numbers of GFP positive platelets upon co-incubation ( Fig. 1p-s). As inhibition of protein translation in platelets by cycloheximide did not result in a reduction of PD-L1-GFP expression in platelets (Fig. 1t), our data suggest that PD-L1 protein transfer and not mRNA transfer is underlying the observed pPD-L1 expression after interaction of platelets and tumor cells 19,20 .
While the transfer of PD-L1-GFP or FLAG-GFP was robustly observed across various NSCLC cell lines, we nevertheless noted differences in protein transfer efficacies. For example, platelets showed low levels of FLAG-GFP and PD-L1-GFP after coincubation with NCI-H322, NCI-H522 and NCI-H23 cells, while HOP-62 and HOP-92 cells displayed significantly higher protein transfer rates (Fig. 2a-d). Given that our data indicated that a platelet-tumor-cell contact is necessary for a sufficient transfer of PD-L1 from tumor cells to platelets, we hypothesized that expression levels of adhesion molecules might determine the efficacy of protein transfer from tumor cells to platelets. Along these lines we found that PD-L1 transfer rates positively correlated with fibronectin (FN1) mRNA expression levels, while no significant correlation was found for fibrinogen alpha chain (FGA) or tissue factor (F3) mRNA expression ( Fig. 2e-g). Of note, fibronectin expression also correlated with platelet-tumor cell interaction in vitro ( Fig. 2h-j). Immunofluorescence staining as well as analysis of protein-protein interaction via proximity ligation assay (PLA) revealed close proximity of fibronectin and PD-L1 ( Fig. 2k-n) at the cell surface. In line with these observations, we found that siRNA mediated knockdown of fibronectin resulted in a significant reduction of PD-L1 transfer, thus functionally validating fibronectin as a key regulator of protein transfer from tumor cells to platelets (Fig. 2o-q).
Platelet adhesion to fibronectin is known to be mediated via several molecules including GPIbα, integrin α 5 ß 1 or GPIIbIIIa 21,22 . We therefore set out to address whether inhibition of these adhesion molecules on platelets reduces adhesion to fibronectin and PD-L1 uptake from tumor cells.
To address the significance of our findings for human cancers, we next quantified PD-L1 expression on platelets in healthy lung tissue or NSCLC tumor tissue. While platelets were detected in high abundance in healthy lung tissue and PD-L1 negative NSCLC, we could not observe any relevant PD-L1 expression on these platelets (Fig. 3a, b, d, e). In contrast PD-L1 positive platelets were observed in high abundance in tissue sections from patients suffering from PD-L1 positive NSCLC (Fig. 3c-e). To quantify the number of PD-L1 positive platelets outside the tumor, we next isolated platelets from the peripheral blood of a cohort of 64 healthy donors and 128 NSCLC patients. Fluorescence-Activated Cell Sorting (FACS) revealed threefold higher numbers of PD-L1 positive platelets in NSCLC patients as compared to healthy donors (median pPD-L1 expression in healthy donors 0.29 (95%CI: 0.21 -0.44), median pPD-L1 expression in NSCLC patients 0.89 (95%CI: 0.61-1.21) (Fig. 3 f). The detected differences were even higher, when total pPD-L1 levels were determined using a quantitative enzyme-linked immunosorbent assay (ELISA). While platelet rich plasma (PRP) from NSCLC patients in average contained 108.3 pg/mL PD-L1, PRP from healthy volunteers only contained 1.8 pg/mL (Fig. 3g, h). Differences in pPD-L1 expression in NSCLC patients versus healthy volunteers were also confirmed using western blot analysis ( Supplementary Fig. 5b-d). Interestingly, PD-L1 expression was highest in patients with advanced (UICC stage IV) tumors ( Supplementary Fig. 5e). Of note, immunofluorescence (Fig. 3i) and immunoelectron microscopy ( Fig. 3j) revealed frequent PD-L1 clusters in platelets obtained from peripheral blood of a PD-L1 positive NSCLC patient, further underlining functionality of pPD-L1, as immune ligand clustering has been described to be a prerequisite for proper binding to its receptor 23 . Prompted by these results, we next explored whether pPD-L1 exerts immune-inhibitory functions. We stimulated human T cells from healthy donors with EBV/CMV-derived peptides in the presence or absence of PD-L1 positive platelets obtained from NSCLC patients. T cell activation was evaluated using an enzyme-linked-immuno-Spot (ELISpot) assays determining the effector cytokines IFNy and TNFα. In line with published data we observed that platelets dampen T cell activity independent of their PD-L1 expression status (Fig. 4a-c and Supplementary  Fig. 6a, b) 24,25 . However, when PD-L1 expressing platelets were pre-treated with the anti-PD-L1 mAb Atezolizumab their T cell inhibitory effect was abolished (Fig. 4a-c). Next, we expanded our work towards tumor-associated antigens. New York esophageal squamous cell carcinoma 1 antigen (NY-ESO-1) belongs to the family of cancer-testis antigens, but is also aberrantly expressed in many tumor entities including NSCLC 26 . Stimulation of T cells from healthy donors with NY-ESO-1 peptides predominantly resulted in a clonal expansion of NY-ESO-1 specific CD4 + T cells (CD62L−/CD45RO + and CD27−/CD28+) (Fig. 4d), which were further specified as CD4+ effector memory T cells (T EM , CD62L−/CD45RO+ and CD27−/CD28+) (Fig. 4d, e). Remarkably, T EM activity, as determined by IFNγ and TNFα release, decreased significantly upon co-incubation with PD-L1 positive platelets. However, T cell activity could be restored when pPD-L1 positive platelets were pre-treated with anti-PD-L1 ( Fig. 4f-k).
To investigate a potential impact of PD-L1 positive platelets on other immune cells, we also characterized changes in the overall immune cell composition (peripheral blood) in 10 NSCLC patients and five healthy controls ( Supplementary Fig. 7a). In NSCLC patients pPD-L1 tended to be correlated negatively with the total number of NK (p = 0.1), CD4+ T cells (p = 0.09) and CD8+ T cells (p = 0.02) ( Supplementary Fig. 7b). Moreover, in NSCLC patients more PD-1 and PD-L1 was expressed on dendritic cells (DCs), natural killer (NK) cells and CD4+ and CD8+ T cells ( Supplementary Fig. 7c, d). In our analyses we did not observe a correlation of PD-1 or PD-L1 expression and pPD-L1 in DCs, NK cells or CD4+ T cells ( Supplementary Fig. 7e, f). However, we detected a positive correlation of pPD-L1 and PD-1 on CD8+ T cells (p = 0.02). We also quantified T cells and T cell infiltration in the TME in eleven NSCLC patients with different levels of pPD-L1 using the MACSima ultradeep tissue profiling platform. Noteworthy, in patients with high pPD-L1 we observed lower numbers of T cells in the TME and less infiltrating T cells ( Fig. 5a-d). In contrast to our findings in the peripheral blood, we observed an inverse correlation of PD-1 on T cells and pPD-L1 (Fig. 5e, f).
Regulation of pPD-L1 during platelet activation. As it is well established that expression levels of platelet surface proteins Fig. 1 Direct platelet-tumor cell interactions result in PD-L1 expression on the platelet surface. a Representative immunofluorescence staining of PD-L1 (red)and the platelet marker CD41 (green) in four different NSCLC tumor cell lines (A549, NCI-H23, NCI-H460, NCI-H226) co-incubated with human platelets (n = 3 biological replicates). Scale bars 200 μm. b Immunofluorescence microscopy of NCI-H460 cells interacting with human platelets derived from a healthy donor (PD-L1: red, CD41: green) (n = 3). Scale bar left 20 µm, right 10 µm. c Quantitative analysis of the PD-L1+ platelets per field of view (FoV), analyzed by immunofluorescence microscopy (n = 9 FoV (small symbols) were analyzed out of a total of n = 3 independent experiments (large symbols)). Horizontal lines represent means. d Percentage of PD-L1+ tumor cells per FoV (n = 3). Data are mean ± SEM. e Correlation of % PD-L1+ platelets/FoV vs. % PD-L1 tumor cells/FoV (n = 3). Correlation was determined by simple linear regression analysis. f Flow cytometry gating strategy for the quantification of PD-L1+ tumor cells (upper) and platelets (lower) after co-incubation. g, h Surface expression of PD-L1 and CD62P on control platelets (PLT) and platelets after co-incubation with A549, NCI-H23, NCI-H226, NCI-H460 cells (n = 4). Data are mean ± SEM. Statistical significance was calculated by two-tailed Student's t test. i Phase-contrast image of A549 tumor cells after 41 min coculture with platelets (ratio 1:1000). Overlaid migration tracks were color-coded based on their mean velocity. j Image sequence depicting tumor cell interaction and protrusion of a single platelet followed by detachment derived from zoom-in area indicated in (i). k Percentage of stable platelet-tumor cell contacts lasting from contact initiation until the end of the observation period (total observation time: 41 min). l Contact duration of platelet-tumor cell interactions. Data derived from the analysis of n = 75 platelets out of one independent experiment. Boxes represent median and 25th to 75th percentiles, whiskers are minimum to maximum. m Scheme of vectors expressing PD-L1-GFP and FLAG-GFP used for transfection. n Immunofluorescence stainings of GFP of A549 cells transfected with FLAG-GFP or PD-L1-GFP (n = 3). Scale bar 50 µm. o Western blot analysis for PD-L1 in untreated and transfected A549 cells (n = 2). Vinculin was used as loading control. Presentation of full scan blots are provided in the Source data file. p, q Representative immunofluorescence microscopy of untreated, FLAG-GFP and PD-L1-GFP-transfected A549 cells interacting with platelets (n = 3). Tumor cells and platelets were stained for GFP (upper) and PD-L1 (lower). Scale bar 50 µm. r, s Flow-cytometry-based quantification of GFP (l) and PD-L1 (m) on platelet surfaces after co-incubation with untreated and transfected A549 cells (n = 3). t Expression of PD-L1 on platelets pre-treated with 100 µM cycloheximide after co-incubation with PD-L1-GFP-transfected A549 cells (n = 3). r-t Data are mean ± SEM. c, d, r-t Statistical significance was calculated by one-way ANOVA and Tukey's multiple comparisons test. Source data are provided as a Source Data file. ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-27303-7 correlate with the platelet activation status, we reasoned that different degrees of platelet activation might underlie varying levels of pPD-L1 expression on the platelet surface. Indeed, when we analyzed the platelet activation marker CD62P, we observed varying CD62P expression levels which showed a strong positive correlation with pPD-L1 expression (Fig. 6a). Of note, while PD-L1 expression in general was lower in unstimulated platelets, we were able to robustly detect pPD-L1 on the platelet surface of resting (CD62P negative) platelets ( Supplementary Fig. 5a). In line with this, we also detected pPD-L1 in α-granules (Fig. 6b).
As even highly standardized blood collection procedures can result in varying levels of shear-stress mediated platelet activation and therefore complicates standardization 27,28 , we hypothesized that different levels of platelet pre-activation might complicate the interpretability and comparability of pPD-L1 levels on freshly collected platelets from different patients. We therefore reasoned that a controlled in vitro activation of platelets with subsequent maximization of pPD-L1 expression might most adequately uncover the total payload of platelet PD-L1 and best possibly allow a comparison between different patients. Indeed, we found that pPD-L1 expression was maximized upon controlled platelet  (Fig. 6g, h) and thus might allow for a better comparability of pPD-L1 levels between different patients. However, controlled platelet activation and subsequent measurement of CD62P and pPD-L1 is technically demanding and might prevent the use of pPD-L1 as a biomarker in clinical routine. We therefore set out to explore whether a normalized PD-L1 level on the platelet surface could be calculated without in vitro manipulation of platelets. To do so we developed an adjustment model based on the calculation of ΔPD-L1 (ratio of PD-L1 before and after stimulation) as a function of pPD-L1 expression in unstimulated platelets and the degree of pre-activation (CD62P expression) (Fig. 6i, j). Specifically, we devised a matrix which allows us to calculate PD-L1 Adj. for subgroups of patients harboring different levels of platelet pre-activation (CD62P expression) and pPD-L1 expression ( Fig. 6j and Supplementary  Fig. 9). Taking advantage of our matrix, corrected PD-L1 levels, designated pPD-L1 Adj. , were determined for all patients (Fig. 6k).
Adjusted platelet-derived PD-L1 serves as a prognostic and predictive marker in NSCLC. We first used the calculated pPD-L1 Adj. levels and performed a receiver-operating characteristic (ROC) analysis for overall survival (OS). We found that pPD-L1 Adj. levels in the subgroup of maximal platelet activation (CD62P 80-100%) showed highest accuracy in predicting OS and were superior in predicting OS compared to pPD-L1 (Fig. 7a). Since no cut-off value for pPD-L1 Adj had been established so far, we analyzed OS in pPD-L1 Adj. quartile groups (Q1-4) using the Kaplan-Meier method (Fig. 7b, c). Details regarding characteristics of our patient population are provided in Supplementary  Table 1. The median observation time for monitoring OS in our study was 23.5 months (95%CI: 3.4-67.55 months). At data cutoff for overall survival, 42 of 128 patients (32.8%) were still alive. Strikingly, patients with high pPD-L1 Adj. levels showed a significantly shortened OS (Fig. 7b). The median survival in Q1 (low pPD-L1 Adj. levels) was 43 months compared to only 24 months in Q3 (high pPD-L1 Adj. levels) (hazard ratio (HR) for death Q1 vs. Q3: 2 (95%CI: 1-3.9)) and 14 months in Q4 (very high pPD-L1 Adj. levels) (hazard ratio (HR) for death Q1 vs. Q4: 3.64 (95% CI: 1.97-6.72)). Importantly, the observed differences in OS were not restricted to the time since initial diagnosis but were still significant when analyzing the time period since platelet analysis (Fig. 7c).
It has been reported that mutations in key oncogenic drivers do not only fuel proliferation via cell intrinsic cues but also impact tumor biology via modulation of the tumor microenvironment [29][30][31] . Along these lines, we found increased pPD-L1 Adj. levels in patients suffering from KRAS mutated NSCLC as compared to those with KRAS wildtype status (Fig. 7d). In contrast, mutations in EGFR, ALK fusions and ROS1 fusions or mutations showed no association with pPD-L1 Adj. levels, respectively (Fig. 7d, e).
We also explored a potential correlation of pPD-L1 Adj. with other clinical parameters. For example, we found that patients with higher tumor stages (T, p = 0.03), higher degrees of lymph node invasion (N, p = 0.04) and a higher tumor grading (G, p = 0.002) expressed more PD-L1 on the platelet surface ( Fig. 7f-h). No association was found between pPD-L1 Adj. and the region of tumor origin (central vs. peripheral, Fig. 7i). However, pPD-L1 Adj. strongly correlated with the occurrence of metastases (p < 0.001), especially liver (p = 0.005) and bone metastasis (p = 0.001) (Fig. 7j-l). In line with previous studies 32 , we also found pPD-L1 Adj. to be positively correlated with smoking history and the amount of pack years ( Supplementary  Fig. 10c, d). Moreover, pPD-L1 Adj. was positively correlated with platelet count, LDH and CRP ( Supplementary Fig. 10j-l).
To further elaborate on the potential of pPD-L1 Adj. as a predictive biomarker in NSCLC, we conducted sequential measurements of pPD-L1 Adj. in 12 patients undergoing conventional chemotherapy or ICI ( Supplementary Fig. 11a, b). Details on therapeutic regimens are provided in Supplementary Fig. 4. In these patients baseline pPD-L1 Adj. levels were determined prior to the first cycle of the respective 1st line treatment. The second measurement was conducted in parallel to the first CT scan. In patients treated with a platinum-based chemotherapy the tumor responses were quantified according to the Response Evaluation Criteria in Solid Tumors (RECIST) guidelines 33 . For patients receiving ICI iRECIST guidelines 34 were used. In both groups . r Representative images of platelet adhesion to fibronectin-coated surface in the presence or absence of different platelet-blocking agents (n = 3). Scale bar 20 µm. s, t Quantitative analysis of the platelet adhesion assay as platelet covered area/FoV in % (s) and platelets/FoV (t) (n = 9 (small symbols) were analyzed out of three independent experiments (large symbols)). Horizontal lines represent mean. Statistical significance was calculated by one-way ANOVA and Tukey's multiple comparisons test. u Quantification of PD-L1 on platelets after co-incubation with PD-L1-GFP-transfected HOP-62 cells with or without pre-treatment with platelet-blocking agents (n = 3). v Ratio PD-L1-GFP+ platelets/PD-L1-GFP+ tumor cells after co-incubation of platelets with PD-L1-GFP-transfected HOP-62 cells with or without pre-treatment with platelet-blocking agents (n = 3). a-d, f, p, q, u, v Data are mean ± SEM. Statistical significance was calculated by one-way ANOVA and Tukey's multiple comparisons test. g, j Correlation was determined by simple linear regression analysis. I, n Data are mean ± SEM. Statistical significance was calculated by two-tailed Student's t test. Source data are provided as a Source Data file.
response evaluation was performed 6-8 weeks after initial treatment. Remarkably, a significant drop in pPD-L1 Adj. levels was detected upon initiation of therapy in those patients whose tumors were later identified to have undergone at least partial remission (PR) (p = 0.02, Supplementary Fig. 11a). In contrast, patients who were later identified to have progressed despite therapy displayed a significant increase of pPD-L1 Adj. already in early measurements after therapy initiation (p = 0.04, Supplementary Fig. 11b). Of note, the predictive value of pPD-L1 Adj. was robust regardless of the used therapeutic regime. In two patients receiving ICI we determined pPD-L1 Adj. at multiple time points. Remarkably, pPD-L1 Adj. expression changes correlated with  Fig. 11c-f). As genomic alterations in EGFR and ALK represent independent factors influencing OS and progression-free survival (PFS), especially in patients receiving ICI, we additionally analyzed the role of pPD-L1 Adj. in the respective subgroups with or without such alterations. Whereas we were not able to detect a significant difference regarding OS ( Supplementary Fig. 12a-c), in pPD-L1 Adj. high patients harboring an EGFR or ALK alteration who received a platinum-based chemotherapy, PFS tended to be worse compared to patients without EGFR and ALK alteration ( Supplementary Fig. 12d, e). This might be explained by the fact that these patients had already shown a tumor progression upon first-line treatment with a tyrosine kinase inhibitor (TKI). In patients with EGFR and ALK alteration who received TKI, pPD-L1 Adj. was not predictive for PFS (Supplementary Fig. 12f). Since in our cohort none of the patients receiving ICI showed EGFR or ALK aberrations, the role of pPD-L1 Adj. could not be investigated in this cohort. Finally, we set out to probe whether the pre-therapeutically determined pPD-L1 Adj. level can predict the therapy response of NSCLC patients to immune-checkpoint blocking antibodies. To do so we analyzed the PFS of patients either treated with only conventional chemotherapy or immunocheckpoint blockade. pPD-L1 positive and negative subgroups were defined according to the median pPD-L1 Adj. level. In patients receiving conventional chemotherapy we observed a significantly higher PFS when pPD-L1 levels were low (Fig. 7m). Interestingly, in patients treated with ICI (Pembrolizumab or Nivolumab), high pPD-L1 Adj. predicted a PFS benefit (HR 4.74, p = 0.003) (Fig. 7n and Supplementary  Fig. 12j). Strikingly, when the predictive power of pPD-L1 Adj. was compared to conventional histological PD-L1 quantification (TPS > 50% and ≥1% in tumor biopsies), pPD-L1 Adj. was found to much better predict therapy response towards ICI (Fig. 7o,  Supplementary Fig. 12k, l). In summary, our data suggest that pre-therapeutically measured pPD-L1 Adj. levels accurately predict the therapeutic response towards immune-checkpoint blocking antibodies. Prospective clinical trials are warranted to validate our findings and to justify the implementation of pPD-L1Adj as a biomarker in clinical routine.

Discussion
Human cancers are heterogenous and biomarkers based on histopathological analyses of single tumor biopsies are often lacking robustness. Histological quantification of intratumoral PD-L1 expression is routinely performed on NSCLC biopsy material as an attempt to predict responses towards immune-checkpoint inhibition, however, the correlation between expression levels and the overall response rate (ORR) is limited 12 . In our present study we show that blood platelets are in frequent contact with lung cancer cells in vitro and in vivo and take up PD-L1 from the cancer cells in a fibronectin, integrin α5β1 and GPIbα dependent manner. Our data provides mechanistic explanation for recent reports describing PD-L1 on platelets from patients suffering from different types of cancers 32,35,36 . Interestingly, while there is comprehensive literature describing tumor cell induced platelet aggregation (TCIPA) and tumor cell-associated thrombus formation 37-41 , our herein presented data suggest that platelettumor cell contact can occur without substantial platelet activation and degranulation. Since pPD-L1 has not only been detected on the surface of activated platelets but also in resting platelets, it is tempting to speculate on an equilibrium between intracellularly stored pPD-L1 in α-granules and cell surface pPD-L1. Indeed, a similar mechanism has been described for the uptake and redistribution of fibrinogen and immunoglobulins 42 .
Importantly, as pPD-L1 is found to inhibit T cell function, it is likely that pPD-L1 plays a distinct role in systemic immunomodulation. Of note, pPD-L1 has recently been described in patients suffering from tumors which were classified as PD-L1 negative in biopsies 35 . Our herein presented data as well as other published studies on tumor heterogeneity suggest that immunohistochemistrybased quantification of protein expression on tissue sections from single biopsies should be interpreted with caution, as protein expression might differ spatially and temporally 43 . Obviously, while our herein presented data suggest a highly efficient uptake of PD-L1 from lung cancer cells into platelets, it does not exclude that some pPD-L1 might be derived from other sources such as endothelial or other non-malignant cell types.
As the total blood volume is circulated up to 1000 times through the body each day, we reasoned that platelets might mirror the collective PD-L1 payload of a tumor and thus might open up venues for novel biomarker strategies. In this regard it is striking that pPD-L1 not only correlates with tumor stage/grade and the occurrence of metastases but is found to be superior in predicting response towards immune-checkpoint inhibition when compared to standard histological PD-L1 quantification on tumor biopsies. Since in particular lung cancer represents one of the most frequent and lethal cancers worldwide 10 , further clinical investigation of pPD-L1 as a biomarker in NSCLC does not only hold the promise to unburden our health systems by avoiding costly and unnecessary therapies with ICI but, even more Immunofluorescence staining for CD41 (green) and PD-L1 (red) on platelets in a PD-L1− NSCLC patient tumor sample (n = 3). Scale bar left 500 µm, center left 50 µm, center right and right 10 µm. c Upper, Representative micrographs of NSCLC adenocarcinoma (H&E) (n = 3). Scale bar left 250 µm, center 50 µm, right 500 µm. Lower, Immunofluorescence staining for CD41 (green) and PD-L1 (red) (n = 3). Scale bar left 500 µm, center left 50 µm, center right and right 10 µm. d, e Quantitative analysis of the platelets/FoV (d) and PD-L1+ platelets (%/FoV) (e) in healthy lung tissue, PD-L1+ and PD-L1-NSCLC patients (n = 9) (small symbols) were analyzed out of three independent experiments in three healthy donors and six NSCLC patients (large symbols). Horizontal line represent mean. f Percentage of PD-L1+ platelets in healthy donors (n = 64) and NSCLC patients (n = 128). Each dot represents a single donor. Boxes represent median and 25th to 75th percentiles, whiskers are minimum to maximum. g Total amount of PD-L1 (pg/mL) in healthy donors (n = 21) and NSCLC patients (n = 64) analyzed by ELISA. Protein level were analyzed in 64 out of 128, randomly assigned patients of the NSCLC cohort. Data are mean ± SEM. h Total amount of PD-L1 (pg/mL) in PRP, platelet lysate, platelet releasate, and serum (n = 6). Data are mean ± SEM i Left, Representative PD-L1 immunofluorescence staining of platelets from a healthy donor (upper) and a NSCLC patient (lower). Platelets were counter stained with phalloidin. Right, Expression pattern of PD-L1 on a platelet derived from a NSCLC patient counter stained with phalloidin (upper) or CD41 (lower). Scale bar left 10 µm, right 2 µm (n = 1). j Platelets of a NSCLC patient, assessed by transmission electron microscopy. PD-L1 stained with post-embedding immunogold-labeling. PD-L1 gold particles densely accumulating on the platelet membrane (black dots). Upper scale bar 2 µm, lower 100 nm (n = 1). d, e, h Statistical significance was calculated by one-way ANOVA and Tukey's multiple comparisons test. f, g Statistical significance was calculated by Mann-Whitney test. Source data are provided as a Source Data file.
important, will avoid side effects of ICI in patients who would not benefit from this kind of therapy.
It should be mentioned that our exploratory study suffers from some limitations. Owing to the fact that we used an exploratory cohort of NSCLC patients with unequal representation of tumor stages and different cycles of various treatment regimens for the development of the calculation algorithm, pPD-L1 Adj. may not have yielded its maximum performance. Expectedly, while pPD-L1 can robustly detected in patients treated with TKI or platinum-based chemotherapy, pPD-L1 could not be detected in patients treated with anti-PD-L1 mAbs. Of note, this observation is in line with previous data 32 Fig. 4 PD-L1 on platelets shows functional relevance via decreasing T-cell activity. a IFNγ ELISPOT assay of peptide-specific T-cells co-incubated with PD-L1+ platelets with or without anti-PD-L1 mAb pre-treatment (n = 3). b Quantification of the IFNγ ELISPOT assays (n = 3). c Flow cytometry-based quantification of indicated cytokines and surface markers for peptide stimulated CD8+ T-cells co-incubated with PD-L1+ platelets with or without anti-PD-L1 mAb pre-treatment (n = 3). d Representative FACS plots showing the gating strategy and the T-cell subpopulations after pre-sensitization, enrichment, and expansion. e Quantitative sub-phenotyping of NY-ESO-1 specific T cells using flow cytometry (n = 2). f Representative FACS plots displaying CD4+ TEM activity levels measured by INFγ expression after co-incubation with PD-L1+ platelets with or without anti-PD-L1 mAb pre-treatment (n = 3). g Quantification of INFγ+CD4+ TEM (n = 3). h IFNy fold change in CD4+ TEM (n = 3). i Representative FACS plots displaying CD4+ TEM activity levels measured by TNFα expression after co-incubation with PD-L1+ platelets with or without anti-PD-L1 mAb pre-treatment (n = 3). j Quantification of TNFα+CD4+ TEM (n = 3). k, TNFα fold change in CD4+ TEM (n = 3). b, c, g, h, j, k Data are mean ± SEM. Statistical significance was calculated by two-tailed Student's t test. Source data are provided as a Source Data file. anti-PD-L1 mAbs to PD-L1 expressing platelets. However, this complicates an exact determination of pPD-L1 in these patients. Even if we did not observe a significant correlation of pPD-L1 and genetic alterations beyond KRAS, this exploratory study cohort of consecutively analyzed NSCLC patients might not be ideal to study the predictive role of pPD-L1 in NSCLC patients harboring genomic alterations including EGFR, ALK and ROS1. Nevertheless, besides the tremendous potential of pPD-L1 Adj. as a biomarker, we believe that platelet PD-L1 might also represent a potential target for therapeutic intervention. This presumption is supported by our observation that pPD-L1 in NSCLC patients correlates with the number of T cells in TME and the number of infiltrating T cells. Similar observations in a mouse model support this finding 35 . Along these lines it is tempting to speculate that pPD-L1 might be involved in formation of the premetastatic niche by generating an immunotolerant environment at sites distant from the primary tumor (Supplementary Figs. 1, 2). Inhibition of pPD-L1 could prevent the formation of metastasis and such a concept would warrant the investigation of immune-checkpoint blocking antibodies in order to prevent metastasis when tumors with high metastatic risk are treated in a curative intention. Of note, clinical trials investigating the perioperative administration of ICI in NSCLC have reported reduced relapse and metastasis and our herein presented data might offer a mechanistic explanation for the observed results 44,45 . Last but not least, as pPD-L1 Adj. is shown to be prognostic and predictive in NSCLC, pPD-L1 might additionally serve as a liquid biomarker for early tumor detection or recurrence, an approach which warrants future clinical testing.

Methods
Study design and selection of patients. During 2016-2019, 173 consecutive patients with non-small lung cancer (NSCLC) treated in the Department of Medical Oncology and Hematology and Department of Internal Medicine VIII, University Hospital Tuebingen, Germany were prospectively included in the study (screening cohort = SC). In order to preclude the influence of anticoagulants like aspirin (ASS), low molecular weight heparin (LMWH) or other heparinoids and non-vitamin K antagonist oral anticoagulants (NOACs), long-term medication of each patient was considered. In our cohort 12 patients with LMWH and 28 patients taking ASS and/or clopidogrel were excluded. In Supplementary Fig. 4, a detailed flowchart of patient selection is given. In all cases, sample collection was performed prior to the next application of the respective therapy. Tumor characteristics are based on baseline clinical staging. In order to take disease progression better into account the occurrence of metastasis was double checked at the time point of study inclusion. Our cohort comprised 71 male and 57 female patients with a mean age of 65.7 years (range  Preparation of platelets. Platelets were obtained from healthy donors (not taking any medication for at least 10 days) and NSCLC patients after informed writing consent. Citrated blood was briefly centrifuged for 20 min at 120 × g, the upper fraction was harvested as platelet-rich plasma (PRP). Platelets were washed twice with citrate wash buffer (128 mmol/L NaCl, 11 mmol/L glucose, 7.5 mmol/L Na 2 HPO 4 , 4.8 mmol/L sodium citrate, 4.3 mmol/L NaH 2 PO 4 , 2.4 citric acid, 0.35% bovine serum albumin, and 50 ng/mL prostaglandin E 1 (PGE1)). To avoid the influence of PGE1 on platelet-tumor cell and platelet-immune cell interaction, we did not use PGE1 in our co-incubation experiments. For platelet activation 10 μM of the Thrombin Receptor Activator Peptide 6 (TRAP-6), a protease-activated receptor 1 (PAR 1 ) agonist, 2.5 μM ADP or 5 μg/mL Collagen was added to the platelets for 2 min. Platelets were fixed by 2% paraformaldehyde for 10 min and washed twice with PBS containing 1% FCS.
Flow cytometry. Flow cytometry was performed using fluorescence-conjugates or specific mAb and their controls followed by species-specific conjugate (Supplementary Table 2) using a FACS CantoII flow cytometer (Beckman Coulter) or a LSRFortessa (Becton Dickinson) from the flow cytometer facility Tuebingen. Positive cells in percentage (%) were calculated as follows: Surface expression in percent obtained with the specific antibody-surface expression in percent obtained with isotype control. Platelets were preselected by CD41a + and CD62P − (resting) or CD62P + (activated). Data analysis was performed using FlowJo software (v.10). In order to verify the reproducibility of our flow cytometry system, we performed a Bland-Altman analysis ( Supplementary Fig. 8f). For immunophenotyping of PBMC subsets of lung cancer patients and healthy control donors were identified by counterstaining with CD3-PECy5 (BD biosciences, San Diego, CA), CD19-APC/Fire750, CD4-Pacific Blue, CD8a-BV605, CD56-PECy7, CD14-BV785, HLA-DR-BV650 (Biolegend, San Diego, CA) and CD16-FITC (invitrogen). PD-1 and PDL-1 expression as well as activation levels were analyzed using a PD-1-APC or PDL-1-APC and a CD69-PE antibody (BD biosciences), respectively. Isotype controls were obtained from BD biosciences. Dead cells were excluded from analysis with LIVE/DEAD™ Fixable Aqua (Thermo Fisher Scientific, Waltham, MA).
Histopathology, immunohistochemistry and immunofluorescence staining of paraffin-embedded tissue samples. Tissue samples were fixed in 4% formalin and paraffin-embedded (FFPE) at the Department of Pathology (University Hospital Tuebingen). The sections were cut briefly in 3 μm sections and stained with Hematoxylin/Eosin and CD61 (clone: 2C9.G3) following standard protocols. For immunofluorescence microscopy, sections were deparaffinized and hydrated in a first step. The heat-induced antigen retrieval method was performed using sodium citrate buffer (pH 6.0) for 30 min. Antigen blocking was performed with Blocking solution (Zytomed) for 60 min. Primary antibodies that were included anti-CD41, mouse, 1:250 (clone: HIP8) and anti-PD-L1, rabbit, 1:200 (clone:  Cyclic immunofluorescence staining of NSCLC patient samples. Paraffinembedded patient samples were cut in 2-5 μm slices and collected on object slides. Subsequently, sections were subjected to deparaffinization and rehydration. Slides were treated with xylene for 10 min, followed by rehydration using an ethanol dilution series of 100%, 95%, 70%, 50% for 5 min each. One last change was performed using deionized water. Heat-induced antigen retrieval was performed using a Sodium-Citrate buffer (10 mM Sodium citrate, 0.05% Tween 20, pH 6.0) and boiling the samples for 20 min. Samples were cooled down and stored in MACSima™ Running Buffer (Miltenyi Biotec, 130-121-565) until initial DAPI staining (Miltenyi Biotec, 130-111-570). The MACSima TM device is an ultra-high content cyclic IF device which allows for fully automated IF imaging. Iteratively, the device performs fluorescent staining with multiple labeled antibodies, image acquisition, and bleaching per cycle. Images were generated according to the manufacturer's instructions and analyzed with the Qi Tissue Image Analysis Software. For quantification at least two ROI were selected based on manual prestaining of DAPI.
Tracking platelet-tumor cell interaction using live-cell imaging. For live-cell imaging analysis A549 cells (cultured as stated above) were used. Tumor cells were co-incubated with platelets at a platelet-tumor cell ratio of 1:1000. Platelets were added to the tumor cells directly prior image acquisition. Platelet-tumor cell interactions were analyzed using phase-contrast live-cell microscopy with frame intervals of 30 s for up to 40 min (Leica Microsystems, Thunder Imager 3D Assay; HC PL APO 40 × /0.95) using adaptive focus control. Cell positions were assigned by their center-of-mass coordinates.
Electron microscopy and immunoelectron microscopy. For transmission electron microscopy, platelets from one representative pPD-L1 high expressing NSCLC patient were used. Platelets were centrifuged and the resulting pellets were fixed for 24 h in Karnovsky's fixative. As previously described, Ultrathin sections were examined with a LIBRA 120 (Zeiss) operating at 120 kV 46 . For immunoelectron microscopy, platelets were fixed and embedded in Lowicryl K4M (Polysciences) 47 . Samples were stained with anti-PD-L1 antibody (Abcam) and examined using a LIBRA 120 transmission electron microscope (Zeiss) at 120 kV.
ELISA. Protein levels of PD-L1 were measured using a human PD-L1 ELISA kit (Abcam, clone:  according to the recommendations of the manufacturer. All concentrations are expressed as means ± SEM of triplicates. Western blot. Whole-cell extracts were prepared using RIPA buffer and protein concentration was analyzed using the BioRad Dc assay. 25-50 μg of protein were transferred to 10-15% SDS-Page and blotted on a PVDF membrane (Millipore) with a wet blot system. The membrane was blocked for 1 h at room temperature with Roti-Block, followed by overnight incubation with the following antibodies: anti-PD-L1, rabbit (  IFNγ ELISPOT assay in CD8 + T cells and platelet T-cell co-incubation. Freshly thawed (ex vivo) PBMCs from healthy donors were analyzed by enzyme-linked immunospot (ELISPOT) assay in duplicates. Interferon γ (IFNγ) ELISPOT assays in our study were performed as described previously 49 . In brief, 96-well nitrocellulose plates (Millipore) were coated with 1 mg/mL anti-IFNγ mAb (Mabtech) and incubated overnight at 4°C. In a next step, plates were blocked with human serum (10%) for 2 hours at 37°C. PBMCs (2.5 × 10 5 cells per well) were pulsed with an EBV/CMV epitope mix containing the frequently recognized peptides BRLF109-117 YVLDHLIVV (A*02) peptide and CMV pp65 (A*02) peptide NLVPMVATV and incubated with or without platelets (ratio 1:50) for 24 h. Phytohemagglutinin was used as positive control. HLA-A*02 (KLFEKVKEV)-and B*07 (KPSEKIQVL)-restricted control peptides derived from benign tissues (HVexclusive HLA ligands) served as negative control. Prior co-incubation with T cells PD-L1 positive platelets from NSCLC patients were pre-treated with the anti-PD-L1 mAB Atezolizumab for 30 min and washed twice with PBS containing 1% FCS. Readout was performed according to the manufacturer's instructions. Spots were counted using an ImmunoSpot S5 analyzer (CTL).
Cytokine and cell surface marker staining. Peptide-specific T cells were further analyzed by intracellular cytokine and cell surface marker staining. PBMCs were incubated with 10 μg ml −1 of peptide, 10 μg ml Generation of NY-ESO-1-specific CD4 + T cells and platelet T-cell co-incubation. The generation of NY-ESO-1-specific T cells was performed using as described previously 50 . In brief, PBMCs from a healthy donor (1 × 10 7 /mL) were stimulated using pools of NY-ESO-1 overlapping peptides (1 μg/mL). The NY-ESO-1 overlapping peptide pool of 15 amino acid length (11 amino acid overlap) was purchased via Miltenyi Biotec. The cells were cultured in RPMI 1640 containing 10 % human AB-serum and 1% L-glutamin in the presence of 10 U/mL recombinant IL-2 and 10 ng/mL Il-7. Culture medium was replaced every third day. After a pre-sensitization period of 7-14 days, NY-ESO-1 specific, IFNγ + T cells were enriched after re-stimulation with NY-ESO-1 peptide pool for 6 h using CliniMACS® (Miltenyi Biotec) technique as reported previously 51  Six hours prior analysis T cells were co-incubated with platelets of NSCLC patients or healthy donors (ratio 1:200) and re-stimulated with NY-ESO-1 peptides (1 μg/ mL). In order to investigate the functional role of PD-L1 on platelets surfaces, PD-L1 positive platelets from NSCLC patients were pre-treated with Atezolizumab (100 μg/mL) for 30 min and washed twice with PBS containing 1% FCS. As a negative control a Myelin oligodendrocyte glycoprotein (MOG) peptide mix was used (1 μg/mL). SEB (Toxin Technology, Sarasota, FL, USA) at 10 μg/mL was used as positive control. NY-ESO-1 specific T cell activity was determined by intracellular TNFα and IFNγ quantified via flow cytometry as described above.
In situ proximity ligation assay (PLA). HOP-62 and NCI-H23 cells were grown on glass bottomed plates. After two washing steps cells were fixed in 1% PFA in PBS (pH 7.4) for 10 min at −20°C. After three washing steps in PBS cells were incubated with a BSA blocking solution (5% BSA, 0,2% Triton X-100, 0,1% Tween) for 30 min. In situ PLA was performed using the Duolink PLA kit (Sigma-Aldrich) according to the manufacturer's instructions. In brief, after blocking cells were incubated with anti-PD-L1, rabbit (1:250, clone: 28-8) and anti-fibronectin, mouse (1:200, clone: P1H11) for 2 h at room temperature. After three washing steps with PBST (phosphate buffered saline, 0.1% Tween), anti-mouse PLUS and anti-rabbit MINUS PLA probes were linked to the primary antibodies for 1 h at 37°C. After three times washing steps with buffer A (0.01 M Tris, 0.15 M NaCl, and 0.05% Tween-20), PLA probes were ligated for 60 min at 37°C. After two washing steps with buffer A, amplification using Duolink In Situ Detection Reagents (Sigma) was performed at 37°C for 120 min. Following amplification, cells were washed three times for 5 min with wash buffer B (0.2 M Tris 0.1 M NaCl). Cells were then coated with Duolink Mounting Medium containing DAPI. Image acquisition was performed using an Olympus BX63 microscope and a DP80 camera (Olympus).
Establishment of an activation-independent calculation matrix for platelet PD-L1. Since platelet pre-activation levels differ due to sample collection/preparation and protein surface expression depends on the platelet activation state, accurate determination of total protein expression on platelet surfaces is challenging. As a result, the platelet pre-/activation level acts as a confounding factor and thus impairs the suitability of pPD-L1 as a promising biomarker in NSCLC. To circumvent this dilemma, we established an activation-independent calculation matrix of platelet PD-L1. The matrix is based on our cohort of 128 NSCLC patients and investigates the activation-dependent expression change of PD-L1 (ΔpPD-L1) during controlled platelets stimulation ex vivo. Patients were categorized into pPD-L1 quartile groups