Actually I am confused about the meaning of biological replications ,at the moment I am trying to design a method and I want to study the gene expression .I want to use one animal under specific condition .So my question is
If I want to do 4 biological replications, should I extract the RNA from the whole animal (homogenize the whole animal and take 4 isolation )
Or should I extract RNA from different 4 organs from the same animal?
Or should I use 4 animal under the same conditions?
Given that RNA expression is tissue specific the 'holistic' approach is likely to wash out any signal. If you're unprepared to do something difficult properly then don't do it at all.
"Given that RNA expression is tissue specific the 'holistic' approach is likely to wash out any signal. "
Your statement is very wide and bold, so wide and bold it is likely to wash out any signal. ;-)
My comment about "holistic" was more a joke than serious, it is about spinning a shortcoming into an advantage.
Anyway, in an ideal world we really would like to have a finer-grained picture considering tissue, spacial location of cell on said tissue, type of cell within tissue, etc. However, there are several constraints on what we can do - the most obvious being time and money, but also number of skilled people working, difficulties in obtaining samples, and so on. And as John said on his comment, there are experiments where effects are global (see here and here for two papers I like), and dissecting is a pain.
I fundamentially disagree. I note that the second paper you mention is one of mine and is very different to what you're proposing.
In our A. thaliana study we only took wild-type two-week old seedlings which are early in development with little variability and few tissues. We obversed what the global transcriptome structure looked like in terms of 3' end formation. We made no assessment of differential gene expression.
Likewise in the other experiment only the sex chromosomes were observed which are of course exquisitely tissue specific, so taking whole organisms is probably ok.
I disagree fundamentally because if an experiment is too hard or expensive do to properly then it shouldn't be performed. You'll save yourself and everyone else a lot of wasted time chasing down false positive data. If you're interested in tissues-specific gene expression differences then look at the tissues not the whole animal.
Well, there are times when it's the right thing to do - like as h.mon mentioned Drosophila are often ground up and sequenced because you are looking for global changes, or extracting the cell is a pain, or you dont know the cell type affected yet (give high-methyl diets to flies and see what gets turned on/off). But certainly you have to be mindful of diluting out signal. I think as long as you respect that, its OK.
i am confused about the meaning of biological replications
So am I, and I think the distinction biological vs technical doesn't make any sense. You should randomly replicate the level(s) of variation that you want to estimate, whether you call that level "biological" or "technical" doesn't change anything.
If you are interesting in the variation between mice within a certain breed, than you should randomly replicate mice within that breed. If you are also interested in the variation between different explants from that same organ, you should randomly split the organ and process the different bits independently. Or again, you might be interested in seeing how different library preparation protocols performs, so you could replicate library preps within the same RNA extraction.
In this experiment, does it matters which level is biological and which is technical? I don't think so, what matters is the question you are interested at.
You're right that one should focus on the variability of interest; but the presence of a stronger variability in another dimension can ruin your experiment. Consider, for example, you want to see how the left bit of the liver is different from the right. You take one sample of each side, and spend all the time and money to finish RNA-Seq and differential expression, and get a gene list with fold-change. You don't know yet if this holds for a second animal, or even if your samples were of a similar material. What if oopsy your left bit of liver had a vein running through it, and that's really the source of the difference seen. You dont know if the difference is really the liver gradient you wanted to measure, or just a sample problem. This can be remedied by taking two samples at each point. Then the same argument applies to the whole animal. They are much more different than you know. Really have to do two or three whole separate animals, and maybe then one bit of left and one bit of right liver will capture the real diversity present. Until you do the measure you don't know.
If you were interested in technical variability, then you should measure the same bit of tissue twice. Here the problem is possibly the variability is present in a gene that is not being transcribed in that animal, so without a biological replicate, you have underestimated the technical variation.
It's important to think about the possible sources of variation when designing the experiment. If money was free, we should always do more samples from more independent animals. When money is limited, we have to choose carefully how many samples of what kind are needed.
Yes, I agree, especially with "presence of a stronger variability in another dimension can ruin your experiment". I was just ranting about the neat distinction "technical" vs "biological". Are mice belonging to the same strain and almost genetically identical technical or biological replicates? People from ecology and evolutionary biology most likely would say "technical" while molecular biologists would say "biological" (apologies about the rough categorization...). Really, I don't think it matters. What is important, as you say, is to be aware of the possible sources of variation and interpret the results consequently, for example by acknowledging that the findings hold for this particular mouse strain.
I agree that greater variability in one dimension can confound the variability of interest - but that's precisely why the distinction between biological (what's going on in the animal/tissue/cell that I want to measure?) vs. technical (what did I do to obtain that measurement?) is critical. Usually, the biological variability of transcription will greatly exceed the technical variability of RNA-Seq. For technical replicates (defined as a single homogenized biological specimen, divided into multiple samples, each of which is processed in parallel for RNA extraction, library construction, and sequencing), we typically see correlation coefficients of 0.98-0.99. That's a main reason why most experiments forgo technical replicates. Plus, there are methods to detect many of the sources of technical variation (i.e., spike-in controls, data analysis for batch effects). While the researcher can exert considerable control over technical reproducibility, s/he is at the mercy of the biological noise in the system. Biological replicates are needed to estimate that noise, so that you can determine, with some statistical confidence, whether your treatment is responsible for the changes that you observe.
You don't mention what the biological question is? Do you wish to study the gene expression under different conditions or treatments or between different tissues?
In designed a robust experiment a clear biological questions is required. Then, and only then, can you start looking at the level of replication required and under which conditions.
Your three options will give very, very different results and perhaps none is suitable for what you wish to do. Having said that, option 1 is the least likely to result any useful data.
OK. So what does the 'treatment' constitute? A drug treatment or environmental change or genetic modification? Decide which tissue is the most relevant to the study (e.g. blood, brain, liver), then decide what suitable controls are and then you need to work out what a suitable number of animals you require per sample (replicates).
You don't mention which animal you're looking at, but if it's something that requires a licence and ethical approval (like we do in the UK) you need to balance the power of the experiment vs avoiding unnecessary animal experimentation. Ethically, there's no point doing two replicates per sample as you will never have any power to identify anything significant and so is a waste of animals. Similarly 20 replicates is unnecessarily high. You probably need to consider doing something like 6 or 7 reps per condition, but that will be very specific to the experiment in question and check with local experts.
You can try a first batch of 3-4 replicates just to see if there is something, but you can't easily do legitimate statistics by combining different batches (your experimental conditions change). 6-7 reps would be what is needed for a publication.
I am usually confident about technical errors on a properly calibrated instrument or with a proven technique (unless you are developping a new technique or you are trying to see why an experiment fails repeatedly). Now, why should we bother doing biological replicates?
Let's suppose you are doing an experiment with two conditions A and B. You want to find if some particular RNA is overexpressed in A. You do your experiment and you find in A it is twice more expressed than in B. What can you conclude?
Not a lot. Maybe the levels of your RNA fluctuate a lot naturally. Maybe your animal had a hidden condition affecting it. If you do biological replicates, you can determine if you have similar levels of expression across individuals. You can then determine if the difference between conditions is mainly due to normal fluctuations between individuals or to a bona fide influence of your conditions.
EDIT: in one of my experiments, I was comparing two groups of mice: in one a two-parts construct (driven by TetO TetA) was active, in another it was missing one half and should have been very weakly expressed.. The experiments gave weird results, because in my inactive construct group, some mice expressed my transgene ten times more than others. They all had one particular half of the transgene, which turned out to have a faint basal expression due to an unlucky insertion in the genome.
If I had tested only a couple of mice, or only mice with one particular half of my transgene, I would have concluded everything was fine: my controls were a lot weaker than my mice with the active transgene. However, the low basal expression was also found in my mice with the full construct, and was enough to invalidate my results.
Biological replicates are parallel measurements of biologically distinct samples that capture random biological variation, which may itself be a subject of study or a noise source. Technical replicates are repeated measurements of the same sample that represent independent measures of the random noise associated with protocols or equipment. For biologically distinct conditions, averaging technical replicates can limit the impact of measurement error, but taking additional biological replicates is often preferable for improving the efficiency of statistical testing.
You should think long and hard about what exactly you want to get out from an experiment that homogenizes a whole animal and looks at the RNA..
It is true, but for small organisms (e.g. Drosophila) it may be hard to dissect tissues and extract RNA, specially for large sample sizes.
On the other hand, you can view these whole organisms extractions as an "holistic" approach. ;-)
Given that RNA expression is tissue specific the 'holistic' approach is likely to wash out any signal. If you're unprepared to do something difficult properly then don't do it at all.
Your statement is very wide and bold, so wide and bold it is likely to wash out any signal. ;-)
My comment about "holistic" was more a joke than serious, it is about spinning a shortcoming into an advantage.
Anyway, in an ideal world we really would like to have a finer-grained picture considering tissue, spacial location of cell on said tissue, type of cell within tissue, etc. However, there are several constraints on what we can do - the most obvious being time and money, but also number of skilled people working, difficulties in obtaining samples, and so on. And as John said on his comment, there are experiments where effects are global (see here and here for two papers I like), and dissecting is a pain.
I fundamentially disagree. I note that the second paper you mention is one of mine and is very different to what you're proposing.
In our A. thaliana study we only took wild-type two-week old seedlings which are early in development with little variability and few tissues. We obversed what the global transcriptome structure looked like in terms of 3' end formation. We made no assessment of differential gene expression.
Likewise in the other experiment only the sex chromosomes were observed which are of course exquisitely tissue specific, so taking whole organisms is probably ok.
I disagree fundamentally because if an experiment is too hard or expensive do to properly then it shouldn't be performed. You'll save yourself and everyone else a lot of wasted time chasing down false positive data. If you're interested in tissues-specific gene expression differences then look at the tissues not the whole animal.
Well, there are times when it's the right thing to do - like as h.mon mentioned Drosophila are often ground up and sequenced because you are looking for global changes, or extracting the cell is a pain, or you dont know the cell type affected yet (give high-methyl diets to flies and see what gets turned on/off). But certainly you have to be mindful of diluting out signal. I think as long as you respect that, its OK.