Plant-nanoparticles enhance anti-PD-L1 efficacy by shaping human commensal microbiota metabolites

Abstract

Diet has emerged as a key impact factor for gut microbiota function. However, the complexity of dietary components makes it difficult to predict specific outcomes. Here we investigate the impact of plant-derived nanoparticles (PNP) on gut microbiota and metabolites in context of cancer immunotherapy with the humanized gnotobiotic mouse model. Specifically, we show that ginger-derived exosome-like nanoparticle (GELN) preferentially taken up by Lachnospiraceae and Lactobacillaceae mediated by digalactosyldiacylglycerol (DGDG) and glycine, respectively. We further demonstrate that GELN aly-miR159a-3p enhances anti-PD-L1 therapy in melanoma by inhibiting the expression of recipient bacterial phospholipase C (PLC) and increases the accumulation of docosahexaenoic acid (DHA). An increased level of circulating DHA inhibits PD-L1 expression in tumor cells by binding the PD-L1 promoter and subsequently prevents c-myc-initiated transcription of PD-L1. Colonization of germ-free male mice with gut bacteria from anti-PD-L1 non-responding patients supplemented with DHA enhances the efficacy of anti-PD-L1 therapy compared to controls. Our findings reveal a previously unknown mechanistic impact of PNP on human tumor immunotherapy by modulating gut bacterial metabolic pathways.

Fig. 1. Plant-derived nanoparticle (PNP) uptake by human gut bacteria.

A Schematic representation of the processing of PNP, labeling with PKH26, and administration to germ-free (GF) C57BL/6 J mice colonized with human fecal bacteria (hFB) as well as PNP/PKH26⁺ bacteria analysis (n = 5 mice per group). Nano10 and exosome-like nanoparticles (ELN) collected from centrifugation at 10,000 × g and 100,000 × g, respectively. B The fecal bacteria from healthy subjects (n = 15) pooled together and incubated with PKH26-labeled PNP in an anaerobic chamber at 37 °C for 2 h (indicated as Human LI). An aliquot of human fecal bacteria colonized GF C57BL/6 J mice (hFB). PKH26-labeled PNP administered to mice at 0.5 g/kg (body weight, n = 6) for 2 h. Fecal bacteria collected from large intestine (indicated as Mouse LI) and small intestine (indicated as Mouse SI). The PNP-PKH26⁺ bacteria from feces sorted by FACS, and DNA extracted for next-generation sequencing (NGS) of the 16S rRNA gene. The bar graph shows the percentage of taxonomy OTU in all sequence reads at the level of family. C Principal component analysis (PCA) of different sources and different types of PNP. The similar composition of bacteria is circled as a cluster. D Individual PNP-PKH26⁺ gut microbiota composition from human LI, hFB mice LI, and SI are shown on a principal coordinates analysis (PCoA) test according to their Bray-Curtis dissimilarity at the family taxonomic level. Source data are provided as a Source Data file.

Fig. 2. Composition analysis of PNP lipids and amino acids.

A Lipid composition of PNP identified with the liquid chromatography-mass spectrometry (LC-MS) (n = 2, each group was pooled from three independent samples). N10, Nano10. B Violin plots showing the composition of lipids in ELN (orange) and Nano10 (blue). The dots represent the mean proportion of each lipid in ELN and Nano10. Vertical lines indicate the interquartile range. Statistical significance between ELN versus Nano10 is indicated in the plot at p < 0.01 using the Kruskal–Wallis test in R. C Principal component analysis (PCA) of lipid composition in PNP using R. D Heatmap indicating amino acid (AA) level on the PNP membrane identified by LC-MS. N10, Nano10. E Principal component analysis (PCA) of AA levels in PNP using R. Source data are provided as a Source Data file.

Fig. 3. Correlation analysis of gut bacteria preferentially taken up and the composition of PNP lipids and amino acids.

A Heatmap indicating Spearman’s correlation coefficients between the lipid compositions or AA levels in PNP and the PNP recipient bacteria integrated from Human LI, Mouse LI, and Mouse SI in Fig. 1B. B, C Volcano plots of Spearman correlation coefficients (x-axis) and significance (y-axis) showing a correlation between the lipids (B) or AA levels (C) in PNP and the composition of PNP recipient bacteria. Significant associations (P < 0.05) are highlighted. List of top 10 p-values in correlation test aside from the plots. D DGDG and PI depleted from ginger and aloe-derived nanovesicles (GNV & ANV) respectively. FACS analysis of Clostridium perfringens (C. perf) incubated with PKH26-labeled GELN, GNV with DGDG depletion (DGDG^dep) and supplemented with DGDG (left panel), and Lactobacillus reuteri (L. reuteri) incubated with PKH26-labeled AELN, ANV with PI depletion (PI^dep) and supplemented with PI (right panel). Quantification of PKH26⁺ bacteria ⁽bar graphs). (P = 0.0095, P = 0.0019, P = 0.0045, P = 0.0098, two-way Chi-Square test, n = 5). E L. reuteri and Ruminococcus bromii pretreated with glycine (1 mg/ml) and valine (1 mg/ml) at 37 °C for 1 h, respectively, following incubation with glycine dehydrogenase (GLDC) (left panel) or valine dehydrogenase (VDH) (right panel) treated GELN/PKH26. FACS analysis of PKH26⁺ bacteria. Quantification of PKH26⁺ bacteria (bar graphs). (P = 0.0451, P = 0.0076, P = 0.0066, P = 0.0071, two-way Chi-Square test, n = 5). Data are representative of three independent experiments as the mean ± standard deviation (SD, error bars). Source data are provided as a Source Data file.

Fig. 4. The influence of PNP on the gut bacterial metabolites using metabolomics analysis.

A–D GF mice and hFB mice treated with PNP (0.5 g/kg, body weight) every other day for two months. Gut feces from GF and hFB mice suspended in PBS (1.0 g/ml). After centrifuging at 10,000 × g, the supernatant was used for metabolomics analysis using 2D-LC-MS/MS. Heatmap indicating top 10 bacterial metabolites up- or down-regulated by ELN (E) and Nano10 (N10) in hFB mice normalized to GF. Source data are provided as a Source Data file.

Fig. 5. The PNP-mediated influence of bacterial metabolites on host metabolic pathways.

A The relative impact on the microbiome metabolic pathway by PNP analyzed with MetaboAnalyst. B Heatmap of the metabolomics pathway in human gut bacteria colonized GF mice treated with PNP using gplots (R package). Source data are provided as a Source Data file.

Fig. 6. Ginger-ELN (GELN) improves PD-L1 antibody effect against melanoma mediated by gut metabolites.

A Schematic representation of the treatment schedule for B16F10 melanoma cell (1 × 10⁵, n = 5) inoculation of hFB mice with anti-PD-L1 antibody (Bioxcell BE0101, 15 mg/ml) via intraperitoneal (IP) injection as well as oral administration with GELN or gut fecal Sup containing metabolites (100 µl from 100 mg feces) from GELN treated mice. B Representative B16F10 melanoma primary tumor (top left) and lung (bottom left, metastatic nodules indicated by red arrows) from tumor-bearing hFB mice (n = 5) at 28 d post-injection of B16F10 cells along with anti-PD-L1 antibody w/o GELN or gut metabolites from GELN treated mice. Quantification of primary tumor volume and metastatic nodule number (> 1 μm) (right panel). (P = 0.0386, P = 0.0055, P = 0.0082; P = 0.0485, P = 0.0085, P = 0.0106; two-way ANOVA test, n = 5). C Representative examples of B16F10 melanoma primary tumor (top left) and lung (bottom left) from tumor-bearing hFB mice (n = 5). The treatment of GELN replaced with GELN-RNAs (0.5 mg/kg, body weight) encapsulated with GELN-derived nanovesicle (GNV). (P = 0.0426, P = 0.0076, P = 0.0068; P = 0.0442, P = 0.0063, P = 0.0070; two-way ANOVA test, n = 5). D Representative examples of B16F10 melanoma primary tumor (top left) and lung (bottom left) from tumor bearing GF mice (n = 5). The treatment of each group was the same as in (B). (P = 0.0418, P = 0.0084; P = 0.0281, P = 0.0045; two-way ANOVA test, n = 5). E Representative hematoxylin and eosin (H&E)-stained sections of melanoma tumor tissue (200x magnification) from tumor-bearing mice from (C) and (D). Scale bars, 200 μm. F Survival rate of mice in hFB (C) and GF (D) mice. (P = 0.033, P = 0.028; P = 0.5165; P = 0.0306; two-way Chi-Square test, n = 5). Data are representative of three independent experiments as the mean ± standard deviation (SD, error bars). Source data are provided as a Source Data file.

Fig. 7. GELN-enriched with aly-miR159a-3p modulates metabolism of DHA in bacteria by targeting PLC.

A Relative level of docosahexaenoic acid (DHA) and eicosapentaenoic acid (EPA) in LI and SI of GF and hFB mice, respectively, using HPLC (n = 3). (P = 0.0001, P = 0.0022; P = 0.0002, P = 0.0079, two-way ANOVA test, n = 3). B HPLC analysis of DHA and EPA in the gut feces from GF and hFB mice as well as medium of Clostridium perfringens (C. perf) (C. perf, ATCC 13124) treated with GNV-RNAs (0.5 mg/kg, body weight; 10 mg/1 × 108 bacteria) and ELN-RNAs mixed from garlic, aloe, and lemon. C Top five abundance miRNAs in GELN from NGS miRNA sequencing. D Schematic diagram of the putative binding sites of aly-miR159a-3p in the phospholipase C (PLC) of C. perf. Red arrows indicate primers for ChIP. The pink bar indicates the lipoxygenase domain. The aly-miR159a-3p seed matches in the PLC gene mutated at the positions indicated. E Schematic representation of DHA metabolism derived from EPA and α-linolenic acid (ALA) and converted to 4-oxo-DHA by PLC. The hypothetical model of GELN-derived aly-miR159a-3p regulates the metabolism of DHA and EPA by targeting PLC. F Analysis of interaction between miR159a-3p and PLC gene using the ChIP assay. Biotin-labeled aly-miR159a-3p incubated with DNA from C. perf (left panel) and feces from the colonized hFB mouse (right panel). PLC gene binding to aly-miR159a-3p pulled down with streptavidin beads and amplified with specific PLC primers. G qPCR analysis of PLC expression in C. perf treated with scrambled miRNA, aly-miR159a-3p, and miR159a mutant. (P = 0.0012, two-way ANOVA test, n = 3). H Western blot analysis of PLC expression. C. perf enterotoxin (CPE) served as a loading control. The size (kDa) of protein molecular weight (MW) indicated. I HPLC analysis of EPA (left), DHA (middle), and 4-oxo-DHA (right) in C. perf treated with aly-miR159a as well as control miRNAs. (P = 0.4478, P = 0.0074, P = 0.0062, two-way ANOVA test, n = 3). J Schematic diagram of PLC knockout (KO) in C. perf and replaced with ampicillin-resistant (Amp⁺) sequence using fusion PCR (left panel). Western blot analysis for wildtype (WT) and PLC KO C. perf (right panel). CPE was the loading control. The size (kDa) of protein MW indicated. K HPLC analysis of DHA in WT and PLC KO C. perf. (P = 0.0073, P = 0.0034, two-way t test, n = 3). Data are representative of three independent experiments as the mean ± SD (error bars). Source data are provided as a Source Data file.

Fig. 8. DHA improves PD-L1 antibody effectiveness against melanoma mediated by reduction of PD-L1 expression in tumor cells.

A Representative B16F10 melanoma primary tumor (top left) and lung (bottom left) from tumor-bearing hFB mice (n = 5) at 28 d post-injection of B16F10 cells along with anti-PD-L1 antibody w/o DHA and EPA via oval gavage (50 mg/kg, body weight, every other day). Quantification of primary tumor volume (P = 0.0322, P = 0.0287, P = 0.0197, two-way ANOVA test, n = 5) and metastatic nodule number (> 1 μm) (P = 0.0188, P = 0.0084, P = 0.0052, two-way ANOVA test, n = 5) (right panel). B Survival rate of mice in (A). (P = 0.0168, P = 0.0211, two-way Chi-Square test, n = 3). C PD-L1 analysis in tumor of B16F10 melanoma xenograft mice in (A) using qPCR. D Representative immunoblots analysis of PD-L1 in tumor of B16F10 melanoma xenograft mice. GAPDH served as a loading control. The size (kDa) of protein MW indicated. Numbers below western blots represent densitometry values normalized to the loading control. E Western blot analysis of PD-L1 in B16F10 cells treated with DHA (0–5.0 mg/ml) for 2 h. The size (kDa) of protein MW indicated. Numbers below western blots represent densitometry values normalized to the loading control. Data are representative of three independent experiments as the mean ± SD (error bars). Source data are provided as a Source Data file.

Fig. 9. Induction of PLC alleviates DHA level contributing anti-PD-L1 immunotherapy non-response in melanoma patients.

A Representative immunohistochemistry (IHC) analysis of PD-L1 level in PD-L1 antibody response (Resp, n = 35) or non-response (Non, n = 26) primary melanoma (T) specimens (n = 20) as well as their adjacent normal skin tissue (A) of patients (left panel). Scale bars, 300 μm. Quantification of PD-L1⁺ cell ratio in the specimens (right panel). (P = 0.0042, P = 0.0026, P = 0.0266, two-way Chi-Square test). B PD-L1 expression in the tissues from the melanoma patients using qPCR. (P = 0.0016, P = 0.0007, P = 0.0303, two-way ANOVA test). C PD-L1 expression in the tissues from the melanoma patients using ELISA. (P = 0.0272, P = 0.0178, P = 0.0352, two-way ANOVA test). D Representative immunoblot analysis of PD-L1 in the tissues from the melanoma patients. The size (kDa) of protein MW indicated. Numbers below western blots represent densitometry values normalized to the loading control. E Analysis of DHA in melanoma of patients using HPLC. (P = 0.0334, P = 0.0025, P = 0.0273, two-way ANOVA test). F Feces from subjects suspended in PBS (0.5 g/ml) following centrifugation, the supernatant was used for DHA analysis with HPLC. (P = 0.0082, two-way t test). G The bacteria isolated from the feces of melanoma patients and PLC level analysis using western blot (left panel) and ELISA (right panel). The size (kDa) of protein MW indicated. Numbers below western blots represent densitometry values normalized to the loading control. (P = 0.0006, two-way t test). The box & whisker plots in Figs. above with the box representing the SD, the middle line within the box representing the median, and the bars representing the range of minimum and maximum. H Schematic representation of the administration to GF C57BL/6 J mice colonized with fecal bacteria from subjects of Resp and Non following B16F10 inoculation and PD-L1 antibody therapy (n = 5 mice). I Representative B16F10 melanoma primary tumor (top left) and lung (bottom left) from tumor-bearing mice (n = 5) at 28 d subjected to injection of B16F10 cells along with anti-PD-L1 antibody w/o DHA via oval gavage (50 mg/kg, body weight, every other day). Quantification of primary tumor volume and metastatic nodule number (> 1 μm) (right panel). (P = 0.0362, P = 0.0158, P = 0.0103; P = 0.0401, P = 0.0377, P = 0.0155, two-way ANOVA test, n = 5). J Survival rate of colonized mice with B16F10 inoculation and administered bacteria and DHA by oral gavage. (P = 0.0226, P = 0.0156, two-way Chi-Square test, n = 5). K Monocytes isolated from the melanoma tissue. FACS analysis of IFNγ in CD8⁺PD-1⁺ cells (left panel). The strategy of gating used is the same as Supplementary Fig. 3C and Supplementary Fig. 9C. Quantification of IFNγ in CD8⁺PD-1⁺ cells^. (right panel). (P = 0.0053, P = 0.0255, P = 0.0072, two-way Chi-Square test, n = 5). L Monocytes isolated from the tumor. FACS analysis of apoptosis by flow cytometry using Annexin V-FITC staining in CD8⁺PD-1⁺ cells of tumor from mice. Numbers in boxes indicate a representative percent of CD8⁺PD-1⁺ apoptotic cells (Annexin V⁺7-AAD ⁻ ) (left panel). Quantification of the percentage of apoptotic CD8⁺PD-1⁺ cells (right panel). (P = 0.0181, P = 0.0303, P = 0.0121, two-way Chi-Square test, n = 5). Data are representative of three independent experiments as the mean ± SD (error bars). Source data are provided as a Source Data file.

Fig. 10. DHA interferes with transcription factor c-myc access to the PD-L1 promoter.

A Schematic diagram of the strategy to demonstrate the interaction of DHA and the promoter of PD-L1 (left panel). Oligo binding to DHA with the expected mobility shift on PAGE (SSGS, right panel). The transcription start site (TSS) is marked by the bent arrow. ATG; translation start code. B 10 pmol of synthetic DNA oligo PD-L1p8, PD-L1p9, and PD-L1p10 (60 mer/each) corresponding to the sequence on the promoter of PD-L1 incubated with DHA (1 µM) or EPA (1 µM) for 30 min at 37 °C. The oligos separated on 15% native PAGE and visualized with ethidium bromide. The red arrow indicates oligo migration shifted. C The shorter synthetic DNA oligos PD-L1p9A to PD- L1p9F (20 mer/each) correspond to the sequence of PD-L1p9. Representative PAGE for the oligos PD-L1p9B to PD-L1p9D with or without DHA. D Oligo PD-L1p9C incubated with gut supernatants (S) from hFB colonization mice without or with GELN treatment (S + GE). Representative SSGS showed the mobility shift of oligo PD-L1p9C and oligo mutant. Oligo PD-L1p9C mutant was used as the control. E Surface plasmon resonance (SPR) analysis of the interaction of biotin-labeled oligos PD-L1p9C and mutant PD-L1p9C-Mut with DHA (1 µM.) (P = 0.0062, two-way t test, n = 5). F The promoter sequences of PD-L1 and mutant inserted into a luciferase reporter p pEZX and transfection of microglia. Luciferase activity assessment 12 h after treatment with DHA (P = 0.0087, P = 0.6154, two-way t test, n = 5). G Representation of a 15% PAGE for the oligo PD-L1p9C, as well as mutants indicated in the figure. Each oligo contained a single base mutation. The size (kb) of the DNA length is indicated near the figure. The base in pink replaced by a different base, caused the abolishment of the DHA binding shift. H The sequence of mouse and human PD-L1 promoter indicating the distance from the TSS containing the DHA potential binding motif (pink) and c-myc binding site (box). I The analysis of luciferase activity for the c-myc KO B16F10 cells transfected with the luciferase plasmid containing mouse (m) and human (h) PD-L1 promoter sequence (P = 0.0005, P = 0.0023; P = 0.0065, P = 0.0078, two-way ANOVA test, n = 6). Treatment of DHA and/or recombinant c-myc protein is indicated in the figure. J 3D predicted structures of the interaction between DHA and oligo mPD-L1p9C or hPD-L1p (G- Red; T- Yellow; C- Green; A- Cyan) at position G-4 and G-11 by hydrogen bond respectively. K SPR analysis of the interaction between c-myc protein and biotinylated oligo PD-L1p9C covalently immobilized onto the sensor chip with or without DHA (P = 0.0002, two-way t test, n = 3). L Biotinylated oligo PD-L1p9C (WT) and mutant transfected into microglia for 12 h and incubated with gut supernatant for an additional 6 h. Metabolite analysis with LC-MS after pull-down with streptavidin beads (P = 0.0022, two-way t test, n = 6). Data are representative of three independent experiments. The box & whisker plots with the box representing the SD, the middle line within the box representing the median, and the bars representing the range of minimum and maximum. Source data are provided as a Source Data file.