Final Report Summary - CANCERBIOME (Cancerbiome: Characterization of the cancer-associated microbiome)
In Cancerbiome we aim to characterize associations of microbiota with cancer by using metagenomics. Furthermore, we want to explore whether potential microbial markers are only at the tumor surface or also in non-invasive samples like faeces which are suitable for diagnostics. We used colon cancer as a test case to establish the various protocols needed to arrive at reliable microbial profiles.
The first phase of the project was consequently focussing on pipeline-development, ranging from sample preparation to analysis in order to minimise batch effects and to be able to compare data from different cohorts. A complex part of the technology development was to establish fast and accurate taxonomic and functional profiling, which involved many bioinformatics steps starting from raw reads generated by an Illumina sequencer and ending with normalised and comparable profiles. We first developed a metagenomic annotation pipeline, MOCAT (Kultima et al, PlosOne, 2012, Bioinformatics 2016) and had then to devise a species-based profiling method with a minimum of biases. We therefore developed an accurate and fast tool that can automatically delineate species from genome sequences using 40 single copy marker genes (Mende et al., Nature Methods, 2013) and applied the concept to gut metagenomes (Sunagawa et al., Nature Methods, 2013). The resulting profiling method does not rely on reference genomes (only for an estimated 45% of gut species a genome sequence exists) and is thus widely applicable.
Another major hurdle was the sampling itself where commercially available kits varied a lot and did not give the yields needed. Therefore, sampling protocols for biopsies and faecal samples have been established, whereby in biopsies the large amount of human DNA still prevents current large-scale application as the microbial fraction could only be increased from 1-10%, meaning that, with a preferred sequencing depth of 6-9 Gb microbial DNA, ten times as many samples have to be taken as planned, which is not affordable within the budget limit; we thus only obtained 16S-based profiles from tumor surfaces and compared them to 16S and metagenomic profiles from faecal samples (where the human contamination is in the order of only 0.1%).
Taken together, despite unexpected difficulties, a reliable and robust pipeline has been developed, and we have taken the colon cancer metagenomes forward to characterise the strength and reliability of potential microbial markers. Although originally only 40 individuals were budgeted to be studied, it turned out that many more individuals were needed to obtain a significant association that correlates with different cancer stages.
To exclude a specific association within one city or country, we analysed two study populations, one from Paris (collaboration with Iradj Sobhani, Henry Mondor Hospital) and one from Heidelberg (collaboration with Magnus von Knebel-Doeberitz and Neli Ulrich, Univ. Heidelberg). A total of 159 French faecal samples, covering healthy individuals, partially with benign polyps, but patients in all four cancer stages, have been processed and analysed. Indeed a classifier based on 16 species shows a strong distinction between health and disease states, comparable to the commercially available FOBT test. Furthermore, validation in a German cancer cohort of 38 individuals showed almost the same signal strength and comparison with hundreds of supposedly healthy individuals (faecal metagenomes from published studies) also confirmed the potential of the approach (Zeller et al., Mol. Syst. Biol., 2014). After these results, several cohorts from different countries have been analysed with similar outcome and consequently a cheap and robust qPCR based method for non-invasive CRC detection has been developed, ready to progress towards clinical applications.
The first phase of the project was consequently focussing on pipeline-development, ranging from sample preparation to analysis in order to minimise batch effects and to be able to compare data from different cohorts. A complex part of the technology development was to establish fast and accurate taxonomic and functional profiling, which involved many bioinformatics steps starting from raw reads generated by an Illumina sequencer and ending with normalised and comparable profiles. We first developed a metagenomic annotation pipeline, MOCAT (Kultima et al, PlosOne, 2012, Bioinformatics 2016) and had then to devise a species-based profiling method with a minimum of biases. We therefore developed an accurate and fast tool that can automatically delineate species from genome sequences using 40 single copy marker genes (Mende et al., Nature Methods, 2013) and applied the concept to gut metagenomes (Sunagawa et al., Nature Methods, 2013). The resulting profiling method does not rely on reference genomes (only for an estimated 45% of gut species a genome sequence exists) and is thus widely applicable.
Another major hurdle was the sampling itself where commercially available kits varied a lot and did not give the yields needed. Therefore, sampling protocols for biopsies and faecal samples have been established, whereby in biopsies the large amount of human DNA still prevents current large-scale application as the microbial fraction could only be increased from 1-10%, meaning that, with a preferred sequencing depth of 6-9 Gb microbial DNA, ten times as many samples have to be taken as planned, which is not affordable within the budget limit; we thus only obtained 16S-based profiles from tumor surfaces and compared them to 16S and metagenomic profiles from faecal samples (where the human contamination is in the order of only 0.1%).
Taken together, despite unexpected difficulties, a reliable and robust pipeline has been developed, and we have taken the colon cancer metagenomes forward to characterise the strength and reliability of potential microbial markers. Although originally only 40 individuals were budgeted to be studied, it turned out that many more individuals were needed to obtain a significant association that correlates with different cancer stages.
To exclude a specific association within one city or country, we analysed two study populations, one from Paris (collaboration with Iradj Sobhani, Henry Mondor Hospital) and one from Heidelberg (collaboration with Magnus von Knebel-Doeberitz and Neli Ulrich, Univ. Heidelberg). A total of 159 French faecal samples, covering healthy individuals, partially with benign polyps, but patients in all four cancer stages, have been processed and analysed. Indeed a classifier based on 16 species shows a strong distinction between health and disease states, comparable to the commercially available FOBT test. Furthermore, validation in a German cancer cohort of 38 individuals showed almost the same signal strength and comparison with hundreds of supposedly healthy individuals (faecal metagenomes from published studies) also confirmed the potential of the approach (Zeller et al., Mol. Syst. Biol., 2014). After these results, several cohorts from different countries have been analysed with similar outcome and consequently a cheap and robust qPCR based method for non-invasive CRC detection has been developed, ready to progress towards clinical applications.