There is no doubt that biases exist in both sampling and sequencing methodology. However, i don't think there's reason for being too pessimistic about the comparison power of the approach. (certainly not as pessimistic as the Amend et al. paper).

I hold this position mainly based on the many human time-series studies done on both 16S and complete genome metagenomics which show that samples from the same individual are much closer to each other than to other individuals. I like to consider these time-point samples as replicates (if you buy the assumption that the microbiota is decently stable over time) and their proximity (based on a distance measure applied to the relative abundances) gives me confidence that we're getting at least something right.

As to using the actual number of reads for comparison, that is obviously flawed due to the variation in sampling depth (i.e. total number of reads you get out of a sample). I do believe however that relative abundances are fine to use for between-sample comparisons.

Hi Carolina ! The question you ask is very similar to detecting transcript abundance based on NGS data in transcriptome sequencing. And the answer is yes read abundance could be correlated to differential abundance of organisms as it can be correlated to differential expression of genes in a transcriptome. The necessity being that you confirm the findings by an independent method such as qRT-PCR for a small sub-set. You also need to understand the resolution of this differentiation based on your data and then make conclusions considering the limitations.

Thanks for the reply Amey! Just found another article dealing with the biases in the whole pyrosequencing approach to explain community "diversity".

Article Groundtruthing Next-Gen Sequencing for Microbial Ecology–Bia...

Use of read abundances for calculating disimilarity matirx is not appropriate.

Article Fungal community analysis by high-throughput sequencing of a...

I am studying fungal ecology using 454 sequencing, and didnt found any corelation between fungal biomas in soil and read count. so we always use transformed data

Although read abundance is treated sometimes as species abundance, this is not absolutely true. It has been shown that read abundance is approximately quantitative within species, but not in the case of between-species comparisons - the paper you've cited: https://www.researchgate.net/publication/47677916_Quantifying_microbial_communities_with_454_pyrosequencing_does_read_abundance_count?ev=srch_pub

So, basically, if your are comparing abundances (and possibly dynamics and development) of a species you may use read abundance data.

Article Amend AS, Seifert KA, Bruns TD.. Quantifying microbial commu...

Hi Carolina,

you should not use absolute read abundances, since you have a great bias caused by unequal sampling effort (i.e. the differing total amount of reads in each sample). However, for many applications, such as ordination methods (e.g. NMDS) it is absolutely fine and sometimes necessary to use relative read abundances, since you preserve some of the structure of your data.

This is a really good review paper dealing with the basic concepts in microbial ecology.

Article Ramette A.. Multivariate analyses in microbial ecology. FEMS...

There are many bias in the evaluation of "true" abundances following any community typing or sequencing method. This is why relative abundance of each OTU is necessary prior to ordination. My thoughts are that it should be usefull to consider some kind of reference samples that are introduced in the NGS run and that have a peculiar treatment, so that target samples can be compared to reference. Such reference should probably be at the heart of the initial questioning that conducted you to NGS (e.g. any stress... : how do microbial community respond ?). Thus even though true abundances are biased, the structure of target communities are compared to the reference one...

Hi Carolina,

Depending how you treat your raw data, and what kind of interpretation you want to do with them later, in principle you can. Check this also http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0040467

There is no doubt that biases exist in both sampling and sequencing methodology. However, i don't think there's reason for being too pessimistic about the comparison power of the approach. (certainly not as pessimistic as the Amend et al. paper).

I hold this position mainly based on the many human time-series studies done on both 16S and complete genome metagenomics which show that samples from the same individual are much closer to each other than to other individuals. I like to consider these time-point samples as replicates (if you buy the assumption that the microbiota is decently stable over time) and their proximity (based on a distance measure applied to the relative abundances) gives me confidence that we're getting at least something right.

As to using the actual number of reads for comparison, that is obviously flawed due to the variation in sampling depth (i.e. total number of reads you get out of a sample). I do believe however that relative abundances are fine to use for between-sample comparisons.

I agrere with Andrei Yukrov. We have compared the abundance for a certain microorganism in sediments using both pyrosequencing reads and qpcr. We obtained a very good correlation (r2 >0.9) between both techniques. Obviously, when different species are compared, biases grow due to differences in primer annealing, and other PCR derived effects. Moreover, there are certain groups that may never be covered due to poor representativity of the primers. Pyrotags (and other ngs-tags) are based on conventional PCR, and as such they must be considered semi-quantitative. However, I believe that all the same it is currently the best approach we have for community structure analysis.

I tend to agree about the statements regarding the problems associated with sampling effort or biases arising due to the PCR techniques, but I still do think that if you have many fold coverage of the sample and your sequencing effort deep enough then you could rely on the abundance of microorganisms not just restricted to one species but to the community as a whole. But the relative quantification approach is a much safer one.

Relative read abundance is useful; but it isn't the whole story and it isn't completely accurate. Issues such as copy number are elephants in the room, for example. Certainly total quantity is another aspect of the ecology that relative abundance doesn't capture.

This is a common approach in microbial ecology, and although biases exist, they occur in whatever method you use to study bacterial communities (mainly due to PCR biases). It is important to standardise the number of reads per sample; you can do this in Mothur using the sub-sample command. This makes the comparisons much more reliable as different samples are likely to have very different read numbers.

Here's another cool paper where, among many other things, they compared microbial abundance as determined by qPCR and 454 pyrosequencing. Check supplementary figure S1 and you'll see that the results are fairly well correlated. http://www.pnas.org/content/109/42/E2846.short

As has been stated, there are issues associated with the use of absolute read count. Copy number and issues related to pooling of samples are two that come to mind (although the latter can be controlled with good technique). And I agree that using relative abundances does a "good job" (exactly what that means is a long topic!) of dealing with some of these issues. The only thing I can add is that I have been involved in several (close to 10) studies that involved multiple subjects being sampled over time. As Paul Costea observed in the HMP samples, there is much higher consistency among samples from the same subject than there is among subjects. This indicates to me that there is some valid signal getting through.

Hi,

We have found that, providing that the amplicons are relatively short (

I have samples of gDNA extracted from intestinal content of fish (200mg starting material per sample). I have run some test PCR's on this gDNA with universal bacterial primers 27F and 1492R just to ensure that I get a band. My DNA readings are between 50-300ng/ul as measured using Qubit. I have run test PCR's with NGS primers also and they have also worked. Is this DNA therefore of sufficient quality and quantity to use for pyrosequencing of the microbiome do you think?