One important caveat to keep in mind when dealing with 16S data is that it can be severely skewed by 16S gene copy number in the genomes. For example, Paenibacillus species can have more than a dozen 16S genes, and their 16S levels will be more than an order of magnitude higher than you would expect from pure cell counts. That means that 16S data often agrees very poorly with more reliable estimates of species abundance based on shotgun metagenomics, or even FISH analysis.

I came across this discrepancy in my own work on compost derived enrichment cultures, and showed that correcting for expected 16S gene copy number can correct for this bias (see http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0068465, supplementary file S2). Jonathan Eisen's team independently came up with a nice tool to estimate expected 16S gene copy number to do this correction:

Kembel SW, Wu M, Eisen JA, Green JL. Incorporating 16S gene copy number information improves estimates of microbial diversity and abundance. PLoS Comput Biol. 2012;8(10):e1002743.

http://www.ploscompbiol.org/article/info%3Adoi%2F10.1371%2Fjournal.pcbi.1002743

Hey Philip, your question is quite common to me. I study the bacteria in sponges. It's not surprising when an OTU totally outrages others.

Thank you Fang - I have seen it reported in the literature from time to time. I will wait and see if this pattern is replicated in my next samples

Hi Philip

I can't comment on the particular environment I'm afraid, as I've no experience in that area. Perhaps your fish gut is indeed dominated by that particular organism. However, you do bring up a valid point regarding primer bias. All techniques involving amplifications are indeed subject to bias. The problem is amplified of course when you have novel sequences which mismatch the primers used, and would thus be missed in the analysis. Another issue is DNA extraction bias. In addition to the suggestions of Stephen above, if you are truly concerned, and indeed a good way to help verify your results, would be to do FISH (not the organism!) with probes complimentary to the dominant organism you have observed in your library thus far. As a control I'd use a probe matching the primer you originally used. If you use a general probe such as DAPI in addition, you can gauge whether there was significant mismatch between your primers and members of the gut commnity.

Hello Philip, I would suggest the very similar experiment of qPCR to check the primer affinity bias via artificially formulated consortia with high to low noise ratio of other species along with the dominant one that you are obtaining from your metagenomic data. If this results corroborate with your result then you can probably go ahead with the well normalized and representative community distributions.

I am sure that a well sized 16S rRNA gene library could even further clarify the event.

All the best.

Jagat

Hi,

for a start, if you let me know your primer sequences I could do a checkup using other general primer sets in order to identify whether your primer pair yields a significantly different picture than other pairs.

Best,

Blaz

It is common for certain environments have a predominance of one or more bacteria. I know that working with molecular biology is expensive, but if possible, I suggest you try using repetitions to confirm whether or not it was a noise.

Hi Phillip,

First thigh you should do is to check, using rarefaction curves, how much your microbiota has been represented in your 16S dataset. It would be positive If it reach the saturation or it is close.

If the number of reads in this specific sample is much lower than the others I would sequence again this sample.

Moreover, I would be careful to assign your reads to species level. It depends on the length, most of all. I recommend you to analyze your data at genus level.

Best

Miguel

Some bacteria have multiple copies of the ribosomal RNA operon while others don't. This, combined with the exponential amplification by the PCR could seriously distort the results. I guess you included a dereplication step in the filtering procedure, if you didn't, I highly recommend that.

One important caveat to keep in mind when dealing with 16S data is that it can be severely skewed by 16S gene copy number in the genomes. For example, Paenibacillus species can have more than a dozen 16S genes, and their 16S levels will be more than an order of magnitude higher than you would expect from pure cell counts. That means that 16S data often agrees very poorly with more reliable estimates of species abundance based on shotgun metagenomics, or even FISH analysis.

I came across this discrepancy in my own work on compost derived enrichment cultures, and showed that correcting for expected 16S gene copy number can correct for this bias (see http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0068465, supplementary file S2). Jonathan Eisen's team independently came up with a nice tool to estimate expected 16S gene copy number to do this correction:

Kembel SW, Wu M, Eisen JA, Green JL. Incorporating 16S gene copy number information improves estimates of microbial diversity and abundance. PLoS Comput Biol. 2012;8(10):e1002743.

http://www.ploscompbiol.org/article/info%3Adoi%2F10.1371%2Fjournal.pcbi.1002743

I agree with Stephen Jackson and Laura Kelly. Additionally, animal stress before collecting samples, the protocol of samples collection and sample storage until DNA extraction and the 16S regions covered in your libraries could provide more and more bias. Is it happening to all your biological replicates? I think you should compare sample collection and extraction protocols, avoiding keeping animals under stress conditions and using liquid nitrogen to do samples storage. What kind of DNA extraction protocol you used? Kit or classical DNA extraction?

Are you doing DNA extraction from gut content only? Do you intend to verify bacteria attached to gut mucosae? The bacteria population can be quite different from one site to another when you compare gut content and gut wall.

You could design qPCR for all these protocolos comparisons.

Best

José

Among all the answers above I like Patrik's and Jose's. But personally, I won't go with qPCR. qPCR and 16S rDNA pyrosequencing are two different techniques. And the primers for qPCR may generate more bias, especially when we facing environmental samples.

It's not odd that the phylogenetic distribution of symbiotic bacteria is far off evenness. But what you do need to know is the host-specificity of that dominate group.

Since you have lots of sequences of this OTU, you can design a group-specific primer/probe and combine with other universal primers to amplify the longer fragments of 16S rDNA. BTW, which hyper variable region of 16S rDNA are you targeting? With the longer fragments, you can apply phylogenetic analysis to get a better idea about this predominate group.

I am not sure if FISH is feasible for you samples. But it would be more convincing if you can use FISH to visualise the distribution of that OTU.

DGGE can give the community profile..Run the gel with your amplified PCR product ..the seperation pattern of the bands will give the idea..the thick bands could be the over represented species.. for confirmation we can cut the bands gel elute, reamplify and sequence..

Hi Philip,

I like many suggestions about testing with other methods, but the basic point is whether the pattern you see is correct or not. It is true that many symbiotic niches do have poor evenness. However, as a first step, you should first confirm the sequences that form this OTU. At what level of taxonomic hierarchy you are defining OTUs and whether the sequence divergence between all sequences within this OTU reflect divergence at that level? You may also look at the microdiversity within this OTU and see whether the differences occur in the hypervariable regions only and if so, whether these are common to operons (something which Patrik suggested above).

We have used this approach to look at OTUs that get selected in certain soil types and not others. See the attached article.

Best,

Kamlesh

Article Relative impacts of land-use, management intensity and ferti...

The dominance of one OTU may be true but we also need to consider the possibility of primer bias. It has been shown that V1-V2 is good for gut microbiota while V4 region is good for soil microbiome. Although both of them have bias, they give relatively reliable results. http://www.nature.com/nrg/journal/v13/n1/full/nrg3129.html

I very much support the idea of doing direct FISH on your samples - assuming your total bacterial densities are sufficiently high - and cell integrity can be maintained, such highly abundant OTU should be easily revealed microscopically. Beauty of FISH is that it does NOT involve DNA extraction - and such biases do not come at play. Neither does 16S copy number interfer (though it will/may make your cell have higher FISH signals). qPCR - anohter suggested technique would be equally, potentially biased by DNA extraction biases and copy nr. biases. Good luck.

This is an interesting problem that has several layers of complexity, each of which needs to be considered. The first is the expected species distribution of the normal gut community, the second is the sensitivity and selectivity of the methodology used to measure the community and estimate species diversity, third is the methodology of calling species (e.g., what is the reference set used to make assignments, what is the level of taxonomic granularity applied), fourth is the degree of sampling error that one would expect from the method applied, and fifth, what are the inherent (hidden) biases?

Many of these points are touched on in the this thread. One point that is not touched on is why you would expect a high degree of evenness in species distribution in the fish gut? Why would you not assume a log-normal distribution? Likewise, why would you not expect to find the gut flora to be predominated by anaerobic species?

Thank you all for your help. I really appreciate it.

Philip - despite all the good answers already given to either methodical or bias correction problems, I've the feeling the first part of your question is a non-existent problem.

Low species evenness and high species diversity are not contradictory. First of all, what is a high level of species diversity anyway and how do you define this? Given the Shannon index, diversity is a plain number which accounts for both the total number of species within your sample and their abundance. But the final number does not necessarily tell you something about the abundance distribution among the different species (the evenness). The more species your sample contains the higher can be the maximum Shannon index (the same is true for more evenly distributed species). But the actual index is dependent on the distribution. So indexes from two different samples can have comparable values, see this easy example:

1. community consist of 2 species and 100 individuals evenly distributed

2. community consist of 12 species and 1000 individuals, where the first two species make up for 495 individuals each and the remaining 10 species are singletons

The Shannon index for both are:

1 = 0.6931

2 = 0.7652

So is this a high or low species diversity? Albeit not equal i would say the are within the same order of magnitude.

Anyway, the evenness index now allows you to interpret the results better by relating these values to the maximum possible diversity, i.e. if all species would have equal abundances:

1: max = 0.6931, evenness = 1

2: max = 2.4849, evenness = ~0.31

Long story short, if you have a rather big community dominated by some species you can have easily "high" species diversity indices but a very low evenness and the more diverse the sample is (in terms of richness) the harder it is to achieve high rates of evenness.

Btw: This is a good example how singletons can influence your community metrics.