There is NO easy way to determine subtle differences in expression when the gene is generally expressed at very low levels. At these levels it may happen, that you have some samples without a single target molecule, others with one, two, three and more molecules per sample. These variations (we talk about 300% and more!) may be biological but even in the absence of any biological variation you will have the stochastic variation (-> see "Poisson distribution") of this size. To reliably determina a, say, 50% change in (average) expression, you have only one option: use a very, very, very large sample (and concentrate the RNA/cDNA), or, what is equivalent in the amount of work and the final result: measure many, many, many samples. There is no way to get the answer you want if you have limited material in few samples. Either you have to analyse thousands(!) of samples or somehow create samples that are thousands of times as big that you can analyze.

(besides these problems the technical problems of qRT-PCR add another piece of trouble [efficiency, noise, detection limit, unspecific products,...])

When I did qPCR, I usually discarded data that was >30 cycles since it is unreliable.

I suggest lysing more cells or concentrating your samples to get at least 32-fold (~5 PCR cycles) more RNA per microliter.

A different approach altogether would be to go to RNA FISH. This way, you get a direct measure of the number of specific mRNAs/cell. You also get a picture of the distribution along the population (i.e. how many cells have 0, 1, 2, 3 etc.. copies/cell) vs. qRT-PCR which is an average. Another advantage is being able to detect transcription sites.

Thanks for your response Gal, the high cycle number was also my concern since they show a high error rate. I will try synthesizing cDNA from more RNA, I originally used 1µg RNA. Could you have a recommendation for some paper for RNA FISH with Plants? Thanks.

How do you know that your transcript is present only at very low levels? Most measures are only relative. Using qPCR you would need a dilution of a template with a known concentration (e.g. Plasmid or PCR product).

But back to your troubleshooting: I have experienced dramatic improvements (~15 cycles) by redesigning my primers - and that without any loss of specificity. I would suggest that you try five each forward and reverse primers in various combination and select the best combination.

Good luck.

Kristian, I have tried 3 different primer pairs, and the problem is not the amplification efficiency, I always have efficiencies between 1,93 to 2 which are very good. I know that the transcript is low because with the same cDNA concentration, my house keeping genes have a Ct value of 18-20 where are my GOI has a Ct value between 33-36, that means at least 10,000 to 15,000 fold difference between the housekeeping gene and the GOI expression.

Are you using oligodT or random primer for the cDNA? This could also affect your specific cDNA yield.

As for FISH in plants - I'm not familiar with any specific papers, but I assume it would be similar to yeast FISH: you would have to digest the cell wall.

I am using random hexamer, I will try using oligo dT.

PCR efficiency does not necessarily reflect the quality of the primers. Several additional factors come to play, particularly the amount of unspecific priming (observed as a smear when confirming band size by gel electrophoresis).

I cannot tell anything from your measures of 1.93-2.0 as PCR efficiencies are usually measured as percent. If your measures are 193;200%, then there is clearly an issue with unspecific priming. If your measures are 0.93(1.93-1.0);1.0(2.0-1.0), then it is INDICATIVE of a good PCR. Same if your measures are 2^(cycle*1.93);2^(cycle*2.0). But as you can see there are many interpretations of your measures, and no matter how they are interpreted they DO NOT guarantee a good/optimal PCR reaction - even if you have tried three primer pairs.

As to the oligoT, this approach results in overrepresentation of amplicons located in the 3'-end of the target genes ~ not a good idea.

If you want to change your approach go for TaqMan, or go old-school and do a Nouthern blot. With a good probe the Southern blot is still the most accurate method for measuring gene level simply because the is no cDNA synthesis and no amplification steps.

Kristian, the amplification efficiency of 2 means that with every cycle there is doubling of template i.e. it corresponds to 100% efficiency, an efficiency of 1.93 corresponds to an efficiency of 93%. The melting curve analysis and gel electrophoresis confirms a single band of expected length, moreover the product has been sequenced and shows the correct sequence. It is simply a gene with very low expression.

I like the idea of Northern blot, but as I understood Northern blot has a lower sensitivity than RT and qRT-PCR, so I do not see if it could help at all.

Oliver I have performed a promoter GUS assay for the said gene and we do not see any GUS staining, both our positive and negative controls work for the GUS staining. This experiment was what actually raised the question, we are able to amplify the cDNA for this gene but do not see GUS expression.

What you need is a control cell population where the gene is known to be expressed. If no protein has been found in all the cell types which have been analyzed, you can question the fact that you are dealing with a true gene coding for a transcription factor. Once you have this positive control, you must compare the relative level of transcripts in the cells you are interested in to that in your control. Then, if you find transcript levels that are 10.000 fold lower than in control cells, you can consider that it is background expression.

I would say more than 34 cycles, but you always have to compare with blank

Relative gene expression calculation with Ct value >30 are not reliable. Normally we run 40 cycle PCR, and the problem lies with the fact that Rf of low abundant gene does not go to plateau, which makes it difficult to set the upper threshold or differentiate from the background noise.

In that case you can compare with blank and try to set up the lower threshold but then your upper threshold is still not defined.

Either you would need to use more transcript to have the CT values for your GOI within the range of HKG. eg if your house keeping gene is expressed at 16CT cycles when using 100ng/ul of reverse transcribed RNA (i.e: 1ul from 1ug RNA/20ul), then you would need to use more RNA for your GOI. say CT per 200ng/ul.

The other alternative could be to check for isoforms of the genes if there are any or even compare with intronic sequences of the gene to establish that there is an expression. You should expect lower Ct values with intronic qPCR but primer design is crucial in such experiments.

Usually anything beyond 33 cycles, I would treat at par with low expression/not expressed. However there might be a low basal expression that can be better picked up with FISH than with qPCR. Some of the new probes (Stellar) allow counting no. of spots on the FISH as a quantification.

Your housekeeping gene is not the good one. You must have an housekeeping gene express at the same level than your gene of interest. But if your negative control is clear, and your positive control positive and not too high, you could use your results. If you want to increase your number of CT, you could increase the amount of RNA during your RT.

If possible, use Taqman assay for your gene of interest. If you dont detect expression in the test sample reproducibly, and DO detect it in positive controls at expected levels, you can be fairly confident in your results.

Fold change values obtained by Q-PCR are subjective and gene-and experimental condition-specific. Few genes do not show very drastic difference in their expression levels under two experimental conditions, say case and control. Therefore, as long as you see sufficiently low p-value, tghe fold change of 1.5 is sufficient enough to consider the change in gene expression.

There is NO easy way to determine subtle differences in expression when the gene is generally expressed at very low levels. At these levels it may happen, that you have some samples without a single target molecule, others with one, two, three and more molecules per sample. These variations (we talk about 300% and more!) may be biological but even in the absence of any biological variation you will have the stochastic variation (-> see "Poisson distribution") of this size. To reliably determina a, say, 50% change in (average) expression, you have only one option: use a very, very, very large sample (and concentrate the RNA/cDNA), or, what is equivalent in the amount of work and the final result: measure many, many, many samples. There is no way to get the answer you want if you have limited material in few samples. Either you have to analyse thousands(!) of samples or somehow create samples that are thousands of times as big that you can analyze.

(besides these problems the technical problems of qRT-PCR add another piece of trouble [efficiency, noise, detection limit, unspecific products,...])

I agree with what was said by Ram Gupta. Once you do proper quality control, your only problem is probably cDNA concentration. I have had this problem before and there are several quality control steps you need to be sure your results are valid. 1) You need high quality primers. Make sure you have good primer efficiency. Usually 90-100% is good, which is about 1.9-2.0 or nearly doubling to doubling every PCR cycle. Ideally the primer set should bridge an intron to be sure you are not getting accidental priming to contaminating DNA. 2) You need a different housekeeping gene that is expressed less or you need to dilute the cDNA for your housekeeping gene. A housekeeping gene expressed at cycle 19 is a poor control for a gene of interest expressed at cycle 32. If you cannot change housekeeping genes, dilute a little cDNA 1:10 or 1:100 for use with the housekeeping gene in the qPCR reaction q-PCR results are all relative so you can use a different concentration of cDNA for the housekeeping gene and the gene of interest as long as you are careful pipetting and always use the same dilutions for each gene. 3) You need to use more cDNA but just for your gene of interest. For valid results you need to have your gene of interest expressed by cycle 29-30. If you loaded 2uL of cDNA previously in your qPCR reaction, try 3-5uL in the reactions with the gene of interest. You can also go back to your cDNA synthesis and add more RNA (use the maximum amount for your cDNA synthesis reaction). Unless you used very little the first time, this will only be moderately helpful, adding more cDNA at the qPCR step is probably more useful. Hope that helps!

All have provided excellent comments and suggestions, but ultimately the solution boils down to sample size, as rightfully noted by Jochen. You can do power calculations to determine sample size but you'll also need very reliable technical performance of the selected assays. In general three replicates are considered enough, but in your case probably 5 or even more (if time and money permits) will be necessary to verify that observed variation between samples is biologically driven not a result of technical variation. The housekeeping genes is another issue; as pointed out you do need a gene which is expressed at similar levels as your target genes. The good news, as I see it, is that you can use any gene of similar expression levels as a control, as long as its expression doesn't vary between samples and exp. conditions.

To comment on your other question, the correspondence between transcription/ translation is medium at best, so even if you have low transcription, protein levels might be higher. This is speculative, but have you measured protein levels? May be these are better, although more laborious, alternative to PCR?

Thank you everyone for your comments and suggestions, I do not understand the problem associated with housekeeping gene selection. I do not see any reason why housekeeping genes expressed at higher levels are not applicable for the assay as long as they are expressed at the same levels across all my tissues analyzed and that the amplification efficiency of my primer pairs for the housekeeping gene and GOI are relatively similar and in the same dynamic range (over different template dilutions)

I will try making cDNA with more amount (5µg instead of 1µg) total RNA and use the a lower dilution for the template in the qPCR reaction. I will do 5 biological replicates and see what comes of it.

Amey:

I also do not understand the "problem" associated with housekeeping gene (proper term is "reference gene") selection mentioned by S. Roche and S. Doherty. Choosing a reference gene with a high Ct value, or diluting your cDNA(??), will just compound the potential for variability since now the reference gene Ct will be subject to the same type of stochastic variation described by J. Wilhelm. Better to have a reference gene Ct that is in the highly reliable and reproducible range. Otherwise, you are in the impossible position of trying to extract information from the difference between two highly noisy signals!

Maybe I can answer the RNA/protein question and clarify about housekeeping/ reference gene choice. 1) The relationship between RNA expression levels and protein levels is different for each gene according to its type of regulation (transcriptional, post-transcriptional, post-translational, etc...). Thus, how closely the mRNA and protein levels will correlate must be determined for any given gene under the experimental conditions of interest.

2) Imagine that your doubling efficiency for a qPCR reaction is 1.99 for the reference primer set you are using. If the reference gene crosses threshold at 19 cycles, you will have compounded that reduction in efficiency through 19 cycles. If your gene of interest has an efficiency of 1.93 and amplifies at 34 cycles, the difference in efficiency (slope of the curve) and numbers of cycle till the curve crosses the threshold will make comparison difficult because these are not linear relationships. In addition, reaction efficiency tends to be lower at higher cycle number because of changes in reagent concentrations that reduce efficiency (dNTPs get used up and Taq has been heated and cooled). These compounding factors can be hard to predict and make your experimental error greater than your fold difference (much more than pipetting a cDNA dilution). Most people don't take these subtle factors into account, but you need to do so when you are trying to observe small changes. Ideally, a dilution curve should be done for each gene, where amplification is determined at different cDNA dilutions to look for a relationship (evidence of true signal). Amplification for the control sample of the gene of interest should be at a dilution such that the it crosses threshold between 15-25 cycles where reaction efficiency is high and there a full 5-6 log fold amplification curve can be observed to ensure efficiency (slope) is being measured at the true linear asymptote. Having a reference gene crossing threshold at a similar cycle number and with similar efficiency gets rid of all of the above issues. Hope that helps!

For more, check out: http://www.ncbi.nlm.nih.gov/pubmed/19246619

Thanks for this reply Heather, I was aware of this, and made standard dose response with a dilution series for all my primer pairs for the reference genes and GOI. The problem was at the lowest diltion of 1:5 my Ct value for the GOI is 31-32 so I made a dilution series with my GOI cloned in a plasmid and calculated the amplification efficiency. I also calculate the amp. efficiency for every run based on the fluorescence cycle plot according to Ramakers et al., 2003.

I will look into determining the protein levels of the concerned TF in my tissues.

Sounds like you are doing the right things Amey. Maybe try a 1:4 or 1:2 dilution for your gene of interest. Best of luck!

You may try RT with gene specific primer. If it does not work, you may amplify your transcript using IVT before PCR. RT should be done either with two GSPs, one of them carrying T7 promoter, or amplify all RNA pool ( T7-SMART primer plus modified oligodT). Although IVT mostly provides a linear amplification rather than exponential as PCR, you need to control your assay. In order to produce a control, make the target transcript using IVT and then measure RNA. Make a serial dilutions and run them through the assay: RT, IVT amplification, and PCR. If you observe the linearity, you are in business. It also allows you to quantify the transcript.

First, I would like to say that I think you should not dilute the samples independently for your target and reference gene. This will end up inserting a source of variation between the genes expression that will not be corrected by the control expression, which can generate a bias in your analyzes.

Perhaps, it may be interesting to have samples where you know that the gene is expressed (overexpressing, maybe) and where you know there is not (a knockout mutant), to be used as positive and negative controls. Then you can get an idea if your target gene is expressed in the sample in question or not.

Dear Mr Bhide. We had the same problem, when the expression was very low due to the type of samples in our case. We solved this problem with preamplification reaction for the interested genes. I suggest you the article from EHP, where this method was used and validated. http://www.ncbi.nlm.nih.gov/pubmed/21247820 I believe that it will help you a lot. Best regards, Simona

Just an update for everyone interested and keeping track. I made cDNA using 3µg total RNA and random hexamer primers and used a 5 µl of a lower dilution (1:20) as template for the qPCR and I got Cts in the range of 27-30 with very low standard deviations for technical triplicates and biological duplicates. So the expression is very low but detectable and reproducible. I also changed my house keeping genes to genes that are about 100 fold higher expressed compared to the GOI as compared to my earlier ones that were expressed between 10-15000 times, but it does not change much with respect to relative quantification. Thank you everyone who helped me with this question.