Buffy coats of al least 10 healthy donors (HD) were provided by Centro Trasfusionale Universitario Azienda Policlinico Umberto I in Rome, Italy (the study was approved by the ethical committee of Istituto Superiore di Sanita`, Rome, Italy, and donors gave written-informed consent to participate). Human PBMC were isolated from buffy coats by Ficoll-Histopaque 1077 gradient (Sigma-Aldrich, St. Louis, MO). Activated NK cells were, obtained starting from a coculture of PBMC (4 X 10⁵ cells/ml) with cobalt-irradiated (4000 rad) RPMI 8866 cells (B lymphoblastoid cell line) (1 X 10⁵ cells/ml) as described in detail in (17). The expanded NK cells subset was incubated with human rIL-2 (100 U/ml; Hoffman-La Roche, Nutley, NJ) for 3 days (cell viability > 90%). This culture method ensures an average of 25-fold increase in activated NK cell number, as well as more than 85-90% of NK cells population expressing CD16+ CD56+ molecules.

The culture supernatants of ex vivo expanded human NK cells were thawed at the time of experimental use and subjected to differential centrifugation as described in (18). Briefly, conditioned cell culture medium was centrifuged for 5 min at 300 × g and 20 min at 1,200 × g to remove cells and debris. NK-derived microvesicels (NKMVs) were pelleted for 30 min at 10,000 × g and washed in phosphate-buffered saline (PBS), while NK-derived exosomes (NKExo) were collected by ultracentrifugation at 100,000 × g for 90 min at 10°C using a Sorvall WX Ultra Series centrifuge in an F50L-2461.5 rotor (Thermo Scientific, Germany). The resulting pellet was first washed in PBS and then ultracentrifuged at 100,000 × g for 60 min. NK-derived extracellular vesicles (NKEv), that comprised both MVs and Exo, were resuspended in PBS, RPMI 1640 medium or phisiological solution depending on the experiments. The characterization of NKEV by Nanoparticle Tracking Analysis (NTA), scanning and transmission electron microscopy, and by proteomic analyses has been reported in our recent study on the immune modulating capability of human healthy donor-derived NKEV (18).

SU-DHL-4, Human diffuse large follicular B-cell lymphoma cell line, (from ATCC) was cultured in RPMI 1640 medium (Life Technologies, Grand Island, NY), supplemented with 10% FBS, 100 U/ml penicillin, 100 mg/ml streptomycin (Life Technologies) and 2 mmol/l glutamine in a 5% CO2 environment at 37˚C. Cells were routinely tested for mycoplasma contamination by a PCR Mycoplasma detection kit (Venor GeM; Minerva Biolabs, Berlin, Germany).

Proteomic analysis was previously reported by our group (18). All the proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD014894. The entire set of proteins identified in Exo and/or MV (3259 proteins in total) were analyzed by DAVID (23) to evaluate the enrichment in KEGG pathways. Only top 40 pathways with FDR <0.001 were taken into account.

SU-DHL-4 cells were plated at 5 × 10⁴ cells per well in buffered RPMI medium. After 24 h, 10 µg Exo, 10 µg MVs or 20 µg EVs (Exo + MVs) were added to the cells. Following 24 h or 48 h of co-culture, cells were collected and washed twice with PBS1X. Apoptosis was evaluated by staining the cells with Annexin V-FITC and propidium iodide (PI) following manufacturer’s instructions (BioVision Incorporated, Milpitas, CA). After staining with Annexin V/PI for 15 min, cells were fixed in 1% PFA and analyzed on a Becton Dickinson FACScalibur, at least 10,000 cells per sample were analyzed using CellQuest software (Becton Dickinson Systems). All experiments were performed in triplicate, each being repeated at least three times. The mean fluorescence intensity (MFI) of samples was used in all cytotoxicity calculations. The maximum level of PI uptake was determined in target cells lysed by 0.5% Triton X-100. Cytotoxicity was expressed as the percentage of cell deaths among target cells: (number of dead cells/[number of dead cells + number of live cells]) x 100.

All animals utilised in the present study were housed and treated in accordance with protocols approved by institutional authorities, in agreement with European Community Directives and with the Italian Law. All efforts were made to minimize animal suffering, to reduce the number of animals used and to adopt alternatives to in vivo testing whenever possible. The animals used in this experimentation were included in the research protocol “Exploitation of human NK cell-derived exosomes as cell-free support in tumor immunotherapy” approved by the experts from Service for Biotechnology and Animal Welfare and authorized bythe Italian Ministry of Health with the Decree nu 225/2016-PR of 1st March 2016.

CB.17 SCID/SCID female mice aged 4–5 weeks (Harlan, Milan, Italy) were kept under specific pathogen free conditions and fed ad libitum. Mice were injected subcutaneously in the right flank with 15 X 106 SU-DHL-4 human follicular B-cell lymphoma cells in 0.15 ml of saline solution (Baxter s.p.a.) and 0.15 ml of Matrigel (Corning^® Matrigel^® Growth Factor Reduced (GFR) Basement Membrane Matrix, Sigma-Aldrich). Mice were then randomly divided into three groups, of seven animals each. The Co-NKEV group was injected s.c. with 10 µg of NKEV simultaneously to the B Lymphoma cells injection, and then treated twice a week with an intratumoral injection of 10 µg of NKEV/mouse. The Post-NKEV group start to receive the NKEV treatment (10 µg) a week after the B Lymphoma cells injection, and treated twice a week with an intratumoral injection of 10 µg of NKEV/mouse, as the Co-NKEV group. The control group received an intratumoral injection of 0.2ml/mouse of saline solution twice a week. Tumor growth was estimated twice per week with caliper using the following formula (24):

After 30 days from the injection of human B Lymphoma cells, all mice were sacrificed by cervical dislocation, following the guidelines of the Italian National Institute of Health. Tumors were analyzed with a two-way repeated measure ANOVA, with treatment (SAL, Co-NKEV, Post-NKEV) x time (7 points) design. To make tumor variances homogeneous at all time points, a square root transformation was applied to data. Time was a within subject factor, the treatment was a between subject factor. Post hoc analyses were conducted using Bonferroni test.

In vivo MR examinations were performed during treatment (from day 14 to day 24 after tumor implantation). MRI and MRS analyses were conducted at 4.7 T on a Varian/Agilent Inova horizontal bore system (Agilent, Palo Alto, USA) equipped with an actively shielded gradient coil (200 mT/m in 150 µs). In order to meet the requirements of spatial homogeneity and signal sensitivity, a volume coil was used for homogeneous transmission in combination with a receiver surface coil (RAPID Biomedical, Rimpar, Germany). MRI evaluation was performed by T1-weighted (T1W: TR/TE=600/18ms, thickness =0.8mm, FOV 20x20 mm2, matrix 256x128, 21 slices, 4 averages) and T2-weighted (T2W: TR/TE=3000/70ms, thickness =0.8mm, FOV 20x20 mm2, matrix 256x128, 21 slices, 4 averages) multislice spin echo MRI. Diffusion weighted MRI (DWI: TR/TE=2000/50 ms, thickness =1.2 mm, FOV 20x20 mm2, matrix 64x64, 12 slices, 2 averages and b-values= 0, 31, 69, 99, 200, 314, 707, 1105 s/mm2) was performed to measure the magnitude of diffusion of water molecules within tissue. This is performed by means of the apparent diffusion coefficient (ADC) parameter that is calculated by using the mono-exponential decay of signals for b-values over 100 s/mm2. Water motion in the capillary network (perfusion) influence the DWI signal. To separate perfusion- and diffusion-related effects can be used the intra voxel incoherent motion (IVIM) model by assuming biexponential behaviour of signal decay. IVIM model can estimate the true diffusion coefficient (D) and perfusion-related coefficient (D*) and perfusion fraction (f). Changes in ADC and in IVIM parameters may be useful for early tumor response assessment as observed in animal and human studies (25, 26). We perform both monoexponential fit for b values over 100 s/mm2 to estimate the ADC and the biexponential IVIM analysis to measure perfusion related parameters. In order to study the biologic heterogeneity of tumor by classifying domains of different diffusivity, which may have prognostic and predictive implication (27), in addition to the estimation of the average ADC, we performed histogram analysis of ADC values. ADCmean, ADCmedian, kurtosis and skewness were determined from ADC histogram analyses. Kurtosis measures how sharp is the peak relative to a standard bell curve: sharp peaks indicate homogeneous tumors. Skewness indicates the departure from horizontal symmetry, which suggests the presence of areas of higher ADC (indicative of necrosis) or lower ADC (proliferating areas) within the tumor. Histogram analyses with their related parameters have been also performed for b value less than 200 s/mm2 i.e. for fast diffusing spins (ADCperfusionmean; ADCperfusionmedian, kurtosisperfusion and skewnessperfusion) (28). All the MR-related parameters (i.e. detectable metabolites, T2, D, D*, f, ADCmean, ADCmedian, kurtosis, skewness) were evaluated with two-tailed t-test and multivariate analysis. Post hoc analyses were conducted using Bonferroni test.

The quantitative in vivo MRS protocol (PRESS, TR/TE = 4000/23 ms) included the T2-corrected water signal as internal reference and the LCModel fitting routine (29). Fourteen metabolites (selected on the basis of in vitro analyses of aqueous tissue extracts of these tumors) were included in the basis set: alanine (Ala), creatine (Cr), phosphocreatine (PCr), glycine (Glyc), glucose (Glc), glutamate (Glu), glutamine (Gln), glutathione (GS), glycerophosphocholine (GPC), phosphocholine (PCho), myo-inositol (m-Ins), lactate (Lac), scyllo-inositol (Scyllo-Ins), taurine (Tau). Spectra of lipids (Lip) and macromolecules were also included in the basis set as detected in our previous experience in other tumor types (29). Both GABA and NAA metabolites have not been usually observed in lymphomas (30), so we excluded their basis from the basis set utilized by LCModel for our quantitative analyses. We included in the analysis only those metabolites that were estimated to have Cramer-Rao lower bounds (CRLB) less than 20% (which corresponded to an estimated concentration error <0.2 µmol/g).

Deuterated reagents (methanol (CD3OD), chloroform (CDCl3)) and deuterium oxide (D2O) (Cambridge Isotope Laboratories, Inc.) and 3-(trimethylsilyl) propionic-2,2,3,3-d4 acid sodium salt (TSP) (Merck & Co, Montreal, Canada) were used for extraction of aqueous and organic metabolites. All samples were stored at −80°C until metabolomics analysis by NMR spectroscopy. All frozen tissues pieces were weighted and placed into pre-cooled (dry ice) homogenization tubes. Ice-cold extraction solvent (methanol/chloroform/water (1:1:1)) was added to each tube and the tissues were then homogenized by homogenizator (Ika Homogenizer T10, Sigma-Aldrich, Milan, Italy) three times over 30s with 30s pause intervals to ensure constant temperatures during homogenization. At least 24h after, polar and lipid phases, containing water soluble and organic cellular metabolites respectively, were separated by centrifugation at 20,000 x g at 4°C for 30 min. Afterward the polar methanol/water phase was lyophilized by using a rotary evaporator (Savant RTV 4104 freeze dryer), while the organic lipid phase was collected in tube and chloroform evaporated under nitrogen gas flow and residues stored at -20°C. The aqueous and lipid fractions were then reconstituted in 700 µl D2O using TSP (0.1mM) as NMR internal standards or suspended in a CD3OD/CDCl3 solution (2:1 v/v) with internal control tetramethylsilane (TMS, 0.05%), respectively.

High-resolution 1H-NMR analysis was performed at 25°C at 400 MHz (Bruker AVANCE spectrometer, Karlsruhe, Germany) on aqueous and organic extracts using a 60°flip angle pulse, preceded by 2s presaturation for water signal suppression (interpulse delay 2s, acquisition time 1.70s, spectral width 12 ppm, 16,000 data points, 320 scans) as previously described (28). Free induction decays were zero-filled to 32,000 data points and Fourier-transformed; a cubic splines model function was applied for baseline correction. Metabolites were identified according to our previous studies (by spiking with metabolite standards) and the databases published on the Human Metabolome Database site (http://www.hmdb.ca). The absolute quantification of metabolites was determined by comparing the integral of each metabolite to the integral of reference standard TSP, and corrected by respective proton numbers for metabolite and TS at equilibrium of magnetization. Metabolite quantification was expressed as nanomoles/g tissue and then converted into metabolite percentage (relative to total metabolites evaluated in each sample) to enable comparisons among the samples. The relative quantification in organic fraction of metabolites was determined by comparing the integral of each metabolite to the total metabolites evaluated in each sample and expressed as percentage relative to total metabolites.

A portion of the resected tumors was processed for immunofluorescence analyses. In particular, tumors specimens were fixed with PFA 4% and 0.12M sucrose in PBS, for 24 h at 4°C. After fixation, samples were rinsed three times with 5% sucrose/0.15mM CaCl2 in PBS for 10 min each, and then dehydrated overnight with 30% sucrose/0.15mM CaCl2; thus, samples were frozen at -80°C. Frozen tumor specimens were then mounted in Tissue-Tek^® O.C.T. Compound (Sakura, Gentaur, Kampenhout, Belgium), serial 10 μm-thick sections were cut using a Leica CM 1860 cryostat (Leica Biosystems, Buccinasco, MI, Italy), and mounted on positive charged microscope slides (SuperFrost Plus, Menzel Glaser, Thermo Fisher Scientific, Rodano, MI, Italy). Sections were incubated in 1% horse serum in PBS for 30 minutes at room temperature and then overnight with the following primary antibodies: CD19 (1:50 BD Biosciences, Milan, Italy), CD38 (1:50, Immunological Sciences, Rome, Italy), Ki67 (1:100, BD Biosciences, Milan, Italy), Bax (1:100, BD Biosciences, Milan, Italy), Bcl2 (1:50, Dako, Agilent Technologies Italia, Cernusco sul Naviglio, MI, Italy). Slides were washed 3 times with PBS and then incubated 2h at room temperature with the appropriated Alexa Fluor^® fluorochrome-conjugated secondary antibodies. For the TUNEL assay the DeadEnd™ Fluorometric TUNEL System, (Promega, Milan, Italy) was used according to the manufacturer’s instructions. DAPI solution (300 nM in PBS; Sigma-Aldrich, Merk Life Science, Milan, Italy) was added to each slide and incubated 10 minutes at room temperature. After this step slides were rinsed with PBS and mounted with Vectashield mounting Medium (Vector laboratories, Burlingame, CA, United States). Images were captured with a Zeiss LSM980 apparatus (Zeiss, Oberkochen, Germany), equipped with a Pln Apo 40x/1.4 oil objective, 0.55NA, Airyscan2 and excitation spectral laser lines at 405, 488, 543, 633 nm. Image acquisition and processing were carried out using Zen Blue edition 3.3 (Zeiss). CLSM images were obtained by Z-projection with up to 25–30 serial sections of 0.15-0.20 µm thick, taken from the bottom to the edge of the tissue sections, several fields were analyzed for each condition and representative results are shown. Quantitative analyses were performed using the Zen 3.3 software and mean fluorescence intensity (MFI) ± SEM of signals were determined in at least 6-7 independent experiment per marker (≥300 cells per experimental condition, in each repeat), and plotted following the formula: Final MFI = MFI (region of interest) - MFI (background).

Statistical analyses were performed by paired and unpaired Student’s t-test, ANOVA one way or two way test for significant differences between groups, and by post hoc analyses using Bonferroni or Tukey’s, as indicated. Data are expressed as mean ± SEM or mean ± SD, as indicated, using GraphPad Prism 5, and p<0.05 or less were considered to be significant.

The studies involving human participants were reviewed and approved by the Ethical Committee of Azienda Policlinico Umberto I, University Sapienza, Rome, Italy. The patients/participants provided their written informed consent to participate in this study.

Activated NK cells were obtained starting from a co-culture of healthy donors PBMC with cobalt-irradiated B lymphoblastoid cell line (RPMI 8866 cells) at the ratio 1:4, as described in detail in (17). After a 10-day expansion phase of this human healthy NK cells subset, human rIL-2 was added to the cell culture and incubated for additional 3 days (cell viability > 90%). At the end of the culture, supernatants were subjected to differential centrifugation, as previously described in (17, 18).

NK-derived extracellular vesicles (NKEV) were isolated by centrifugation at 100,000 x g, while NK-derived microvesicles (NKMVs) and NK-derived exosomes (NKExo) were obtained by centrifugation at 10,000 x g and 100,000 x g, respectively. All these NKEV were extensively characterized, as reported in our previous work (18).

The present study is an extension of our previous works (17, 18), we herein investigated the capability of human healthy NKEV in reducing viability and inducing cytotoxicity in cancer cells, both in vitro and in vivo in a B lymphoma model. The entire set of proteins identified in Exo and/or MV (3259 proteins in total) and reported in Federici et al. (18) was analyzed by DAVID to evaluate a possible enrichment in KEGG pathways. As shown in Figure 1 , several pathways were found enriched. In particular, we focused our attention in NK cell mediated cytotoxicity and apoptosis pathways (red bars in Figure 1 ), that are described in the cartoons in Figures 2 , 3 , respectively. Many proteins involved in these two pathways have been identified in our set (red stars in Figures 2 , 3 , see also Supplementary Tables 1 , 2 ), and most of them were detected both in Exo (solid circles Figures 2 , 3 ) and MV (empty circles Figures 2 , 3 ). In the KEGG pathway enrichment analysis performed by using DAVID we also found the necroptosis, the Neutrophil Extracellular Trap (NET) and the biosynthesis of amino acids pathways ( Supplementary Figures 1 – 3 ). This functional enrichment observed in the NKEV protein composition suggest that they could exert a cytotoxic effect and that their presence in plasma could play a key role to support the efficiency of the immune system.

In order to evaluate the cytotoxic effects exerted by NKEV, NKExo and NKMVs on tumor cell viability we performed flow cytometry analyses on SU-DHL-4 B lymphoma cells after incubation with NKEV for 24h and 48h. In particular, cytotoxicity was evaluated using the AnnexinV-FITC/PI dual staining assay ( Figure 4 ). As reported, NKEV were able to induce up to 60% of cell death, mostly by apoptosis, both after 24h and 48h of co-culture with SU-DHL-4 cells ( Figures 4A, B ). Representative cytometry dot plots are shown in Figure 4C . We also investigated the single contribution of NKExo and NKMVs in inducing tumor cell death ( Figures 4D, E ). We observed that NKMVs are able to induce at least 50% and 40% of cell death by apoptosis after 24h and 48h, respectively ( Figure 4E ). Whereas, NKExo induced up to 30% of apoptosis after 24h and 20% after 48h of co-culture ( Figure 4E ). Representative cytometry dot plots are shown in Figure 4F . Moreover, non-apoptotic cell death remained at low levels during NKEV incubation (less than 10% at both 24h and 48h, Figure 4B ). Besides, co-culture with NKExo and NKMVs increased non-apoptotic cell death, up to 30% after 24h of incubation with both kind of nanovesicles, while we observed a slight reduction after 48h (up to 10% with NKExo and 20% with NKMVs) ( Figure 4E ).

In the light of the results obtained with NKEV in vitro, mouse xenograft models of SU-DHL-4 cells were used to determine in vivo effects of NKEV in B lymphoma tumors ( Figure 5A ). SU-DHL-4 cells were subcutaneously inoculated in SCID immunodeficient mice, afterwards randomly divided into three groups (n=7 per group). One group of mice (the Co-NKEV group) received NKEV (10 µg) as a co-injection with tumor cells, and then these mice were treated twice a week with an intratumoral injection of 10 µg of NKEV/mouse ( Figure 5A ). The Post-NKEV group, instead, started to receive the NKEV treatment (10 µg) 7 days after SU-DHL-4 cells injection, and treated twice a week with an intratumoral injection of 10 µg of NKEV/mouse, as for the Co-NKEV group ( Figure 5A ). An intratumoral injection of 0.2 ml of saline solution/mouse was administered twice a week to the control group. Tumor growth was monitored twice per week during the follow-up of treatment by standard external caliper measurements, until animal sacrifice.

NKEV-treated mice displayed a significant reduction in tumor growth at all-time points, starting from day 13 after the injection for the Co-NKEV group (simultaneous inoculation of B lymphoma cells with NKEV, Figure 5B , red line), and day 19 for the Post-NK group (NKEV treatment a week after the B lymphoma cells injection, Figure 5B , green line), as compared to control group ( Figure 5B , blue line). The antitumor effects exerted by NKEV were more pronounced in the Co-NKEV group (-64% tumor growth respect to control group at day 13 after the injection), compared to the Post-NKEV group (-44% tumor growth respect to control group at day 19 after the injection) [ANOVA repeated measurement, with treatment (saline, Co-NKEV and Post-NKEV) x time (7 points) design, p=0.003, Bonferroni post hoc analyses]. Moreover, we also abserved differences in the engraftment among the groups. After 10 days from inoculation, in the Co-NKEV group the tumor mass had grown 27% less than control group and 11% less than the Post-NKEV group.

Next, we wanted to highlight the differences in the texture inside the tumor more than volume reduction, so we performed in vivo MRI of all xenograft tumor-bearing mice. The analyses showed homogeneous intensity in both T1 and T2, without any significant morphological alteration. T2-weighted images and the ADC maps of representative tumors are shown in Figure 5C . One-way ANOVA did not detect any significant changes in the T2 (measured by MRS) and in the diffusivity parameters derived from DWI, neither with IVIM method nor with the monoexponential model. Data (mean ± SD) and the corresponding level of significance (p) resulting from the ANOVA analyses are summarised in Table 1 . The homogenous tumor allowed the acquisition of good quality spectra. Examples of spectra obtained from a control and a Co-NKEV-treated xenograft are shown in Figure 5D . The in vivo quantification (mM) of the MR visible metabolites in the three group of xenografts showed a significant increase in the tumor lipid/lactate signal at 1.33 ppm (one-way ANOVA, p=0.0015) and in taurine signal (one-way ANOVA, p=0.0018) in the Co-NKEV group, compared to control and to the Post-NKEV group, which could be attributed to increased cell apoptosis (31, 32) ( Figure 5E ). A trend of increase of glutathione in the Co-NKEV group with respect to control and to the Post-NKEV group was also found ( Figure 5E ).

In order to better characterize the intratumoral metabolomics profile and to define the metabolic effects of NKEV treatment on B-cell lymphoma xenograft model, we performed an ex vivo analysis by using high resolution (HR)-NMR approach which offers quantitative determination of aqueous and lipid metabolites linked to different biological pathways. First, we measured the organic fraction of tissue extracts and we found significant changes in the pool of fat acid (FA) and in their degree of unsaturation (UFA). In particular, we observed a significant (p ≤ 0.05) increase in FA pool and decrease in UFA and mono-unsaturation (MUFA) in both Co-NKEV group (n=6) and in Post-NKEV group (n=7) as compared to control group ( Figure 6 ; Supplementary Table 3 ). In parallel, the content of fraction of saturated FA was significantly different in both treated groups as well (p ≤ 0.05). No differences were found in the levels of triacylglycerols (TAG), phospholipids, pool of phosphatidylcholine (PC) plus Lyso-PC and total cholesterol between both treated groups and control tumors. This active FA metabolism found in organic fraction following both treatments could be supported by average increase (although not significant) in some metabolites linked to branched aminoacids (valine and isoleucine), succinic acid and acetic acid found in aqueous fraction of tissues extracts of both Co- NKEV and Post-NKEV compared to control tissue extracts ( Supplementary Table 4 ).

We then analysed SU-DHL-4 human follicular B-cell lymphoma xenografts to better characterize the effects of NKEV treatment on tumor tissues. We first performed immunofluorescence staining using anti CD19 and CD38 Abs, markers for neoplastic B cells and haematological tumors in general ( Figure 7 ). Our results showed a drastic decrease in the CD19 expression after treatment with both Co-NKEV and Post-NKEV, while CD38 expression was slightly reduced, as also confirmed by the quantification of mean fluorescence intensity. The effects of NKEV treatment were particularly evident in the B-cell lymphoma proliferation rate, as indicated by the decrease in the expression of Ki67 marker ( Figure 8 ). Moreover, we evaluated the typical apoptosis-related markers Bax and Bcl2 ( Figure 8 ). Intriguingly, pro-apoptotic Bax expression was increased with the Post-NKEV treatment, while the Co-NKEV treatment did not alter Bax expression, compared to the untreated tumors. In contrast, anti-apoptotic Bcl2 expression decreased mainly in the Co-NKEV treatment. TUNEL assay showed an increase in apoptosis in tumors treated with NKEV a week after the B-cell lymphoma injection (Post-NKEV group) ( Figure 8 ). Taken together these results indicate that NKEV treatment induced a reduction in the B-cell lymphoma proliferation rate, as well as an induction of apoptosis, with a more pronounced effect obtained with the Post-NKEV treatment.