Coupling culturomics and metagenomics sequencing to characterize the gut microbiome of patients with cancer treated with immune checkpoint inhibitors

Human samples were collected from 3 different biobanks, including the CRCHUM NSCLC ethical number CE17.035, CRCHUM healthy volunteers’ biobank ethical number CE20.300 and from the PRIMM study ethical number NCT03643289. Samples were collected by the patients in their homes according to International Human Microbiota Standard (IHMS), and stored in the fridge for 1 day, then brought to the research centre, and immediately at − 80 °C. Responder status was defined by either a partial or complete response to ICI as assessed by the investigator. Patients were considered non-responders if they had stable or progressive disease as assessed by the investigator.

Microbial culturomics was used to explore the bacterial diversity of stool samples and identification was facilitated using the MALDI-TOF MS. For the culture, we used two steps: Firstly, a direct inoculation of the stool sample was performed with 0.3 g of stool resuspended in 1 mL of 1 × PBS. Ten serial dilutions of this suspension were performed, and 50 µL of each dilution was spread on 5% Columbia agar (COS) (Nepean, Ontario, Canada) enriched with sheep blood (ThermoFisher, Montreal Canada), plates were incubated under aerobic and anaerobic conditions using Zip bag (Becton Dickson Mississauga, Ontario, Canada) containing an anaerobic generator, GasPak (Becton Dickson Mississauga, Ontario, Canada) atmospheres at 37 °C for 48 h. Secondly, enrichments were made by adding 200 µL of each sample in liquid broths BACTEC™ vials (Becton Dickson Mississauga, Ontario, Canada) supplemented with 5% of defibrinated sheep blood and 5% of 0.22 µm filtered rumen fluid under both aerobic and anaerobic conditions. For anaerobic conditions, serial dilutions were prepared from the anaerobic BACTEC™ vials and then spread on COS agar at different time points (24 h, day 3, day 7, day 10, day 15, day 21 and day 30) over a period of one month, at 37 °C, using Zip and GasPak generators. For aerobic conditions, serial dilutions were also prepared from aerobic BACTEC™ vials and inoculated on COS agar at the same time points, but under an aerobic atmosphere. The colonies obtained were subcultured after incubation for 48 h at 37 °C, and the purified colonies were identified using MALDI-TOF mass spectrometry. For each colony, a double deposit was made on a 96 MSP microplate and then coated with 2 μL of matrix solution, prepared from saturated α-cyano-4-hydroxycinnamic acid powder mixed with 50% acetonitrile and 2.5% trifluoroacetic acid. The spectra of each colony were then measured using the MicroFlex LT/SH spectrometer and automatically recorded using FlexControl v.3.4 and MALDI Biotyper Compass v4 software for assay preparation and biotyping. The spectra obtained were compared with the MBT Compass BDAL library (Bruker) and our local database. Colonies with a score > 1.9 were identified to the species level. Colonies not identified by MALDI-TOF MS (score < 1.9) were subjected to genomic sequencing for identification.

gDNA was quantified using the Quant-iT™ PicoGreen® dsDNA Assay Kit (Life Technologies). Libraries were generated from 50 ng of gDNA using the NEBNext Ultra II DNA Library Prep Kit for Illumina (New England BioLabs) as per the manufacturer’s recommendations. Adapters and PCR primers were purchased from IDT. Size selection of libraries containing the desired insert size was performed using SparQ beads (Qiagen). Libraries were quantified using the Kapa Illumina GA with Revised Primers-SYBR Fast Universal kit (Kapa Biosystems). Average size fragment was determined using a LabChip GXII (PerkinElmer) instrument. Libraries were normalized and pooled, then denatured in 0.05 N NaOH and neutralized using HT1 buffer. The pool was loaded at 225 pM on an Illumina NovaSeq S4 lane using Xp protocol as per the manufacturer’s recommendations. The run was performed for 2 × 150 cycles (paired-end mode). A phiX library was used as a control and mixed with libraries at 1% level. Base calling was performed with RTA v3.4.4. Program bcl2fastq2 v2.20 was then used to demultiplex samples and generate fastq reads. Shotgun metagenomic sequencing was performed at a read depth of 15 Gb/sample. FASTQ files were processed using the MetaPhlAn4 pipeline as previously described [11].

To allow comparison with the culturomics results, we performed frequency comparison within each group, which considers only the presence or absence of each species. For diversity analyses and Linear Discriminant Analysis (LDA), we used the MicrobiomeAnalyst pipeline using the default parameters (https://www.microbiomeanalyst.ca/), which relies on relative abundance or the number of read for each species. Alpha and beta diversity were also assessed using this last pipeline, with a p-value ≤ 0.05 considered statistically significant.

Twenty-two patients with cancer were included in the study and compared with a control group of seven healthy volunteers (HV). In the cancer group, 13 patients had NSCLC, and 9 had cutaneous melanoma, with a median age of 66 and 61 respectively, and 69% of NSCLC and 67% of melanoma were male (Table 1). All patients with cancer in both cohorts had advanced disease. For the NSCLC cohort, 85% patients had adenocarcinoma with programmed death-ligand 1(PD-L1) tumor proportion score (TPS), a standard prognostic biomarker for patients with NSCLC) [20]. Most patients were treated with single-agent anti- programmed death-1 (PD-1). For the melanoma cohort, 78% of patients had a diagnosis of cutaneous melanoma with BRAF wild type status (BRAF is the most commonly altered genetic alteration in melanoma and used for treatment decisions), and most patients were treated with combination anti-PD-1 and anti-CTLA-4. In the HV group, the median age was 37, with 43% being male.

Through culturomics analysis of 29 samples, we identified 221 different species from 102 genera with Bacteroides, Enterococcus, Streptococcus and Alistipes being the most common genera present, including three novel species; Alistipes montrealensis sp.nov, Tractidigestivibacter montrealensis sp. nov and Gabonibacter chumenis sp.nov [18, 19, 21] (Table S1).

We first compared diversity metrics between the cancer and HV groups. The HV group had significantly more isolated species compared to the cancer group, with 110 different species (mean of 34, range 21–43) compared to 182 different species (mean of 20, range 5–55) (p = 0.0024), respectively, with the exception of one outlier in the cancer group (Fig. 1A, Table 2). Additionally, differences were observed between cancer histologies with NSCLC harboring more isolated bacteria than melanoma, with 145 different species (mean of 26, range 16–55) versus 75 different species (mean of 11, range 5–24) (adj p = 0.0067), respectively (Figure S1A, Table S1). Beta diversity analysis also showed clustering of HV compared to cancer patients and differences between HV and distinct cancer histologies (Fig. 1B, Figure S1B), (p = 0.004, p-adj = 0.003 and p-adj = 0.003, respectively).

Bild vergrößern

Next, we performed heatmap representation to determine significant bacteria enriched in both groups. Several species were significantly enriched in the HV group compared to the cancer group including members of the Bifidobacterium genus (B. longum, and B. adolescentis), Bacteroides genus (B.ovatus, B.uniformis, B.nordii, B.vulgatus, B.fragilis and B. thetaiotaomicron) as well as Alistipes onderdonkii (Fig. 1C). Conversely, in the cancer group, Coprobacillus cateniformis, Klebsiella oxytoca and Veillonella parvula were more abundant (Fig. 1C). Subsequently, a separate analysis of the top 20 bacteria revealed that 18 species were significantly enriched in HV group. Frequency difference analysis confirmed the enrichment of B. longum, B. adolescentis and Bacteroides thetaiotaomicron in the HV group (Fig. 1D). In addition, using adjusted p-value 14 bacteria remained significantly increased in the HV group (Figure S1C). In the cancer group, we observed a trend for enrichment of two Enterocloster spp. (aldenensis, clostridioformis), Veillonella parvula as well as Alisitpes putredinis, Odoribacter splanchnicus and Erysipelatoclostridium ramosum reclassified as Thomasclavelia ramosa. For further characterization, we performed an overlap analysis and found that 71 bacteria were common to both groups, while 39 were unique to HV and 111 were exclusively isolated from cancer patients (Fig. 1E, Table 2). There were several genera exclusive to cancer including: Enterocloster, Peptoniphilus, Bacillus, Erysipelatoclostridium, and Veillonella. In the same way, Eikenella, Oscillibacter and Bifidobacterium genera were only found in HV (Table 2). Altogether these results demonstrate an enrichment in the number as well as specific bacteria previously associated with a healthy gut microbiome in HV, and the presence of specific genera exclusively present in cancer.

To identify cultivated bacteria associated with responder (R, n = 11) and non-responder (NR, n = 10) status of cancer patients (melanoma and NSCLC) based on objective response rate (ORR) to ICI, we first examined diversity metrics between these two groups. No significant differences were observed in the number of isolated species or in beta diversity (p = 0.17, and p = 0.12, respectively) (Fig. 2A, B). However, Hungatella hathewayi and Cutibacterium acnes were significantly enriched in NR. Conversely Bacteroides ovatus, B. vulgatus, Escherichia coli, Eggerthella lenta and Veillonella parvula were significantly increased in R patients (Fig. 2C). Subsequently, addressing the 20 most frequent isolated bacteria in each group, NR patients exhibited a total of two Enterocloster species (E.bolteae and E.clostridioformis), three Clostridium species (C.culturomicsense, C.cadaveris and C.perfringens) and Hungatella hathewayi and Cutibacterium acnes were the only significantly increased bacteria in this group (Fig. 2D). Conversely, the following species were significantly enriched in the R group with the predominance of Bacteroides genus (ovatus, vulgatus, cellulosilyticus, stercoris, and fragilis,) and as well as Escherichia coli, Eggerthella lenta, and Veillonella parvula were also enriched in this group (Fig. 2D). Of note, using Benjamini–Hochberg method to calculate the adjusted p-value, there no longer a significant difference between both groups (Figure S2A).

Bild vergrößern

Next, we performed the same analysis in the subgroup of NSCLC patients (n = 13), and identified similar findings as for the pooled NSCLC and melanoma results. NR (n = 6) exhibited an enrichment of three Enterocloster spp. (aldenensis, bolteae and clostridioformis) as well as Hungatella hatewayi and Clostridium perfringens (Figure S2B&C). In contrast, the R (n = 7) group showed an enrichment of Bacteroides genus (Figure S2B). These findings align with previous studies that differentiate the gut microbiomes of R from NR, further reinforcing the role of the gut microbiome in segregating patient phenotypes to ICI.

To understand the commonalities and differences between culturomics results and metagenomics, we performed shotgun metagenomics in all NSCLC patients. Metagenomics sequencing detected a higher number of species with 721 taxa (mean of 164 range, 32–253) compared to 154 different species (mean of 26, range 16–55) (p-value < 0.001) identified by culturomics (Tables S2-S3). Among the 721 taxa, 425 were unassigned at the species level (Table S2). Of the 296 species identified (assigned at the species level) by metagenomics (mean of 85, range 24–122), 61 species were common with culturomics, while 93 species were found only by culturomics (p-value < 0.001) (Fig. 3A, Tables S3-S4).

Bild vergrößern

We then analyzed the metagenomics data taking into consideration the patient clinical outcome. As previously published in the literature [2, 3], alpha diversity was significantly higher in R compared to NR (p = 0.001) (Fig. 3B), with no difference in beta diversity (p = 0.27) (Figure S3A). Linear discriminant analysis (LDA) between R and NR NSCLC revealed that Faecalibacterium prausnitzii, Eubacterium rectale, Roseburia inulinivorans, Coprococcus comes, and Agathobaculum butyriciproducens species were enriched with R status (Fig. 3C). Conversely, only Clostridia_unclassified_SGB4367 species was increased in NR NSCLC patients (Fig. 3C). In addition, we performed frequency comparisons between R and NR groups which revealed only for non-adjusted p-value additional enriched species, including Lacrimispora amygdalina, Evtepia gabavorous, Streptococcus parasanguinis and Agathobaculum butyriciproducens in R, and Clostridia-unclassified-SGB4367, Ruminococcaceae-GGB58485-SGB80143, Enterococcaceae-GGB33516-SGB5434 and Anaerofilum-sp-BX8 in NR (Figure S3B and C).

These results reinforce previous findings demonstrating higher alpha diversity as well as specific bacteria associated with response to ICI in NSCLC such as Faecalibacterium and Roseburia and also highlight the complementary nature of these two orthogonal gut microbiome profiling techniques.