much different result from shotgun metagenomics and 16S amplicon

[chloe1005]

Hi, community,

I am analyzing the taxonomic profiling of my shotgun data. Which are 100bp paired-end reads from Illumina Hiseq. Now I am using Metaphlan2 to do the metagenomics profiling. However, the profiling result is far away from Illumina 16S miseq results. Since I have also been using Illumina 16S Miseq to test the taxonomy of my samples for several years. I have two the control samples and treatment samples. In Metaphlan2 results, it gave me around 30% archaea and 70% bacteria for control samples, while Miseq 16S reads tell me that only around 15% archaea and 85% bacteria for control samples. For treatment, shotgun profiling told me 60% archaea and 40% bacteria, while Miseq gave me 20% archaea and 80% bacteria. For my experience, this kind of sample could not achieve that much archaea abundance than bacteria. Furthermore, some(not all) bacteria and archaea composition are different between Miseq result and Metaphlan2 result.

Why is the result so different? Are there any suggestions why the two method result differs so much?

I am confused. Looking forward to a help.

[kbseah]

Hello,

I've seen similar issues with my own data, and in general I think that taxonomic profiles should always be taken with a pinch of salt. Off the top of my head, a few possibilities:

1. Metagenome read profiling methods can be quite sensitive to the database used and the cutoffs for assigning a given read to a taxon. It might be worth trying a different pipeline like Kraken (https://www.ncbi.nlm.nih.gov/pubmed/24580807) to see if you get similar results.

2. rRNA operon copy number can vary between different microbial species. E.g. if species A has two copies of the 16S gene per genome, and species B has only one, one, then A might appear to be twice as abundant as B. You could try profiling only the 16S sequences from the metagenomic shotgun libraries to see if this gives a better fit to your amplicon libraries, e.g. with Emirge (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3219967/) or Matam (https://www.ncbi.nlm.nih.gov/pubmed/29040406). My colleagues and I are working on a pipeline for quick screening and comparison of metagenome libraries for SSU using Emirge and other tools (https://github.com/HRGV/phyloFlash).

3. Amplicon libraries can be quite heavily influenced by amplification and primer biases during PCR (e.g. see https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3592464/)

Hope this helps!

-- Brandon

[chloe1005]

Hi,
Thanks so much for the reply. These make sense to me. And now I am totally agreed. During these days after I posted the thread, I have been trying many different method and software. I found just taxonomic profiling cannot be accurate, and the different result got from the comparison between 16S amplicon is expectable.
Kraken gives me 2% reads hit NCBI. It is lucky to meet PhyloFALSH, which can be used for extract 16S reads and give me the taxonomy result. 0.107% reads hits to SILVA database. Still waiting for the publication of PhyloFLASH.
I have also tried a software- Kaiju, which got 47% reads hits to NCBI nr database, 31% reads hits in RefSeq Complete Genomes database, 38% reads hits in proGenomes database.
Interesting, challenging but confusing. Maybe for environmental samples, assemble is necessary.
Looking forward to more suggestions shared from you.
Best.