We're in the process of doing some of this at the moment - I'll let you know once my student has some results! How did you normalise your RNAseq data? Have you used multiple housekeeping genes for the qPCR and checked that they are behaving like HK genes in the RNAseq?

Are you using TaqMan or SYBR green? Did you test for DNA contamination in your RNA before performing reverse transcription?

Could let us know the range of correlation you did found? And how you did the correlation?

I'm trying to do exactly the same at the moment. Has the species you are working with been sequenced? If it has then look for splice forms of the genes that you have designed your primers for and see if you are covering multiple variants (like Nallasivam said). Also make sure that any form of normalisation of your transcriptomic data isn't skewing the overall picture.

In a paper published by my lab the authors sequence a miRNA library by solexa from different stages of a cockroach.

"The results indicate that when all miRNAs were considered into the analysis, there was no significant correlation between Solexa reads and qPCR data. However, a significant positive correlation was observed when we considered those miRNAs with more than 100 Solexa reads. These results suggest that Solexa sequencing data with less than 100 reads can only roughly represent the relative abundance of miRNAs"

http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0019350

I hope this information could be useful for you, although it wasn't transcriptomics.

Order the primers around the part that has been sequenced.

Similar to Jesus Lozano above, I have found that for genes with low expression, the Q-PCR and RNA-seq do not correlate well, but for genes that are more highly expressed the two techniques correlate better. These are anecdotal observations that are not published. Sorry I don't yet have a good number for "low" and "high" expression.

In my experience, comparing absolute quantitation of expression levels across platforms tends to result in less than ideal correlation. This is especially true if normalization is handled differently for the 2 approaches. Even comparing 2 NGS platforms using normalized counts can be problematic. Each platform has its own biases associated. For this reason, we tend to correlate relative expression across platforms. This is done by sequencing/doing qPCR for 2 samples with some replication. For RNA-Seq data (non-miRNA), I have used DESeq to calculate significant fold changes for individual genes then compare the same genes to the ddCT values with TaqMan qPCR. The correlation between fold changes per gene tends to be much better than comparing absolute values.

The primers you use will exert some influence on your results. You may design more primers to check whether the primers are different.

Definitely, qPCR is a better approach to validate the expression level.Three software applications--geNorm, NormFinder and BestKeeper--are used in most of the reports till date to estimate expression stability and provided congruent results. Considering the diverse technologies now available for transcriptome analysis, methods for standardising measurements between platforms will be paramount, hence, a revalidation with qPCR is always a better approach,As teh RNA standard and reference gene selction are oe of teh important criteria, we will found this paper useful, atleast I have found so,Methods. 2012 Jul 24. [Epub ahead of print]

Application of next generation qPCR and sequencing platforms to mRNA biomarker analysis.

Devonshire AS, Sanders R, Wilkes TM, Taylor MS, Foy CA, Huggett JF.

Source

Molecular and Cell Biology, LGC Limited, Queens Road, Teddington, Middlesex TW11 0LY, UK.

Thanks so much everyone! We wanted to validate 5 of the genes with qPCR, and only one showed the same direction of fold change as the RNA seq. I must add that we worked with very small sample numbers (3 - 6). Thanks for you inputs!

On the side of sequencing - try getting the counts for exons, and get fold change + DESeq/edgeR on the exons (or DEXseq, the exonic DESeq). On the side of PCR - design the primers in the same single exon that you compare with rnaSeq. Smaller granularity of genomic units should result in better comparability. Fingers crossed! :)

In our RNA-Seq based expression profiling studies we saw a very good correlation between RNA-Seq expression levels and both RT-qPCR and end product PCR. However, Sometimes the poor expression levels observed in PCR experiments could be due to specificity of your primers and change of both or one of the primers (as suggested by wenbing) might improve your results and thus better agreement between the expression levels detected using two tools. It could be that you have already tried it and you still see a poor correlation.

The other scenario is if it is a alternatively spliced gene it could be your primers pick only one spliced variant and its abundance could be different to what you have seen in RNA-Seq. It could be that you have already tried to detect possible splice variants. If not try to reanalyze data to detect splice variants and different sets of primers would help you to profile splice variants using PCR.

I forgot to mention about primer efficiency. I know that you know deferent primers have deferent efficiencies. You can assess primer efficiency and specificity in your RT-qPCR, you could check this with cDNA ("normal" PCR or RT-qPCR). You have to keep an eye on copy numbers in your analysis using cDNA. if your primers are sitting in two exons, you could design a new forward and rivers primers, that could be used with your old primer pair. If you see a difference in primer efficiencies compared to the primer pairs that show you better correlation between RNA-Seq and PCR expression levels, either you could try a new primer pairs for the poorly correlated once or in case of "normal" PCR you could play around a little bit with annealing time, temperature and Mg concentration and some other parameters. I think most probably you would have already thought about these.

Transcriptome expression results are generally fair estimates for genes which are abundantly expressed in a particular tissue. In our studies we have observed that for rare transcripts the expression estimates given by RNA sequencing does not correlate with qRT-PCR. Second, designing of primers is the important thing for expression validation. If the primers are designed from junctions which participate in alternative splicing then you will not get good correlation. Other thing primer sequences should be designed from the reads only and not from the reference annotation. In my observation if you take care of these things are very less chances that your RNA-Seq and qRT-PCR results will contrast each other.