Inconsistency between 16s sequencing with Illumina and qPCR

[Ayaka]

Hi all, I would like to ask your opinion about inconsistent results regarding the abundance of certain bacteria between sequence data and qPCR using the same DNA samples.
I have 2 treatments (A and B) and I want to know the effect of those treatments in the composition of the microbiota. In a first instance we did qPCR for certain bacterias and observed differences in the samples before and after the treatments with Wilcoxon signes rank test (paired test) p<0,05) so we went on sequencing the same DNA samples.
The sequencing data was analyzed using proportion ( tested also with Wilcoxon signes rank test) and using the DESeq2 (differential abundance).
The sequence data shows significative difference with the proportion analysis for treatment A but not forB (p-value=0.2754 ) and for treatment B but not A with DESeq2 ( Doing the analysis at OTU level, all the OTUs of treatment A had p-values >0.2).

What can be the causes of this inconsistency between the analysis done with sequencing data and qPCR?
and
What is the most reliable result, qPCR or sequence data?


thanks, Cheers.


PS: We did pair end sequencing 300x2 with primers for the V3-V4 region with 70.000 sequence per sample ( 82 samples).
The primers used for the qPCR were checked with BLAST and their specific for the bacterias of interest.

[thermophile]

1-sequencing isn't completely quantitative. It's sort of quantitative but with big caveats (pcr bias, primer bias, gene copy number, etc)

2- are you sure that you're comparing apples to apples? How are you determining which OTUs are the OTUs that your qPCR is measuring (just taxonomic identification is not sufficient, you should be doing in silico PCR to determine which reads are which groups). All sequences that hit a particular qPCR primer should be combined for that analysis-don't run each OTU that hits each primer separately

[bunce]

Hi - I agree with Thermophile. One of the simplest things you can do is an in silico analysis of the binding sites in your bacterial targets for the V3-V4 primers. Over a number of PCR cycles even slight efficiency difference in primer binding can skew composition. You should also do multiple replicates (possibly across multiple dilutions). The way you generated the library can also skew the composition (e.g. using a 2-step PCR to add adapters). There are few things to try here but I would trust the quantitativeness of qPCR over read abundance. Cheers Mike

[Michael Love]

a side note: I would recommend looking at effect size (LFC) rather than p-values. In fact, this is one of our points in the DESeq2 paper.

short version: the p-value tells you a lot about power, and this is certainly going to be different across technology.