On 2016 Jul 14, Andrea Fuso commented:

The analysis was initially made on these two genes since those are the genes we were studying in relation to muscle differentiation (myogenin) and Alzheimer's Disease (PSEN1). In these two models we gained in the years sufficient knowledge to finely modulate methylation and to foresee that different treatments correspond to differential methylation. Moreover, we are interested in "rapid" methylation changes and it is our hypothesis that genes without CpG islands can be modulated by dynamic changes in DNA methylation that are more rapid than in CpG islands (in wich methylation-dependent silencing/activation is more stable) and possible also in non replicating tissues (brain). We are now checking promoters with and without CpG islands. Of course, designing MIPs in CpG islands could be very difficult or even impossible. That's a limitation.

On 2016 Jul 14, Long-Cheng Li commented:

Dear Andrea,

I am sorry for not having not responded earlier and thank you for your reply.

If the possibility of incomplete conversion can be excluded as you have explained above, I think the findings in your work is significant in that they makes researchers better aware of the non-CpG methylatioon issue. I do have another comment though which is why you chose two genes (myogenin and PSEN1) without CpG islands in their promoters instead of genes with CpG islands where differential methylation often occurs.

On 2016 Jul 13, Andrea Fuso commented:

None

On 2016 Jul 13, Andrea Fuso commented:

Dear Long-Cheng, thank you for your critic comments. Discussing this issue is very important and I'm sincerely happy you took some time to read our manuscript and share your thoughts. When we published the paper, we were hopeful it could raise some discussion useful to a better understanding of the non-CpG issue.

First of all, I would like to emphasize that our paper was not a criticism of MethPrimer. As a matter of fact (as disclosed in the paper) different primer design softwares, not just “MethPrimer”, allow to design bisulfite primers based on the assumption that non-CpG cytosines are transformed. MethPrimer is simply the most used and known and we defined “methprimers” the oligo designed using these different softwares. Our criticism were, on the contrary, addressed toward the concept that stands at the base of all these primer design softwares: i.e. that non-CpG cytosines are (almost) always demethylated. The idea of releasing MethPrimer was great and in line with the widely accepted concept of that era on non-CpG methylation. We should now critically consider that concept and make some effort to better understand extension and significance of non-CpG methylation and, in case, revise our tools for the study of DNA methylation. The adaptation of MethPrimer to the picking of primers in regions with few cytosines goes in this way and it is highly appreciable.

As for your comments about incomplete conversion: your considerations are correct but you are not correct when you affirm that “no efforts were made to verify the completeness of bisulfite conversion of the template DNA”. If you look at the methods and at the previous papers cited in the methods (previously published description were not repeated and just cited) you can see that we took many efforts to demonstrate that our analysis was not affected by artifacts. Our studies on myogenin methylation (some years ago) started with classical bisulphite conversion (no kits) but in the years we started to place side by side classical and kit-assisted bisulphite conversion, obtaining comparable results. The present study was made by using the Qiagen Epitect Kit but specific (high and low methylated samples) controls were modified using the ZymoResearch kit and the in-lab prepared bisulphite. These controls were real samples, not just PCR products. According to your idea that our data could be due to incomplete conversion, you should explain why incomplete conversion occurs only (or mainly) for high methylated DNA but not in low methylated DNA, since the difference between methprimers and MIPs is less when analyzing low methylated samples. In our analyses, in which samples with different (expected) methylation levels are modified in the same run, we can indeed observe both methylated and unmethylated sequences, according to the expected methylation. I think that this observation, more than any “control” witness the quality of the bisulphite conversion we make. If you suggest, or suspect, that bisulphite conversion kit (like the Qiagen and the Zymoresearch) that are today used in almost all the methylation studies, and that are preferred to the in-lab prepared bisulphite just to guarantee reliability and avoid incomplete conversion, are responsible for incomplete conversion….well, the problem is much bigger. I will be happy if you and others want to add more to this interesting discussion.

In conclusion, I completely agree with you that much more care should be put in the control of the bisulphite analysis…and as a reviewer of papers dealing with DNA methylation, I can see that few researchers take care of performing rigorous controls. But, I also invite you and the other colleagues to consider that new and unexpected results, such as relevant non-CpG methylation, deserve to be considered with unbiased attitude and methods.

On 2016 Jul 04, Long-Cheng Li commented:

As the author of MethPrimer, I thank the authors of this paper for looking into the bias issue of our program. However the conclusion reached that "MethPrimers can result in underestimation of the high methylation levels eventually biasing the detection of differences" could be misleading based on the following reasons:

1) MethPrimer was developed in an era when there was little knowledge about the existence of non-CpG methylation and under the assumption that all non-CpG cytosines are not methylated. One of the important principles used by MethPrimer in picking primers is to select primers that preferentially amplify bisulfite converted DNA, given the inherent issue of incomplete conversion of unmethylated cytosines during bisulfite treatment. This is implemented by forcing picking primers in regions which must contain a minimum number of non-CpG cytosines.

2) Using “methylation-insensitive primers” (MIPs) in which non-CpG Cs are degenerated, the authors detected higher levels of methylation compared to primers designed by MethPrimer in the promoter regions of two mouse genes (myogenin and PSEN1). However, no efforts were made to verify the completeness of bisulfite conversion of the template DNA. Although control samples with complete methylated and unmethylated DNA were included in the analysis, the use of PCR products as a control for unmethylated DNA is problematic since incomplete conversion is less an issue for PCR products compared to genomic DNA. In other words, complete conversion of the control DNA does not necessarily mean that genomic DNA was also completely converted. If that were true, the non-CpG methylation detected could be merely an artifact.

I do caution that researchers should keep in mind potential non-CpG methylation. Before more evidence is available about the prevalence of non-CpG methylation and its function, researchers should care more about potential incompleteness of bisulfite conversion and subsequent overestimation of methylation levels in the DNA.

On 2016 Jul 04, Long-Cheng Li commented:

As the author of MethPrimer, I thank the authors of this paper for looking into the bias issue of our program. However the conclusion reached that "MethPrimers can result in underestimation of the high methylation levels eventually biasing the detection of differences" could be misleading based on the following reasons:

1) MethPrimer was developed in an era when there was little knowledge about the existence of non-CpG methylation and under the assumption that all non-CpG cytosines are not methylated. One of the important principles used by MethPrimer in picking primers is to select primers that preferentially amplify bisulfite converted DNA, given the inherent issue of incomplete conversion of unmethylated cytosines during bisulfite treatment. This is implemented by forcing picking primers in regions which must contain a minimum number of non-CpG cytosines.

2) Using “methylation-insensitive primers” (MIPs) in which non-CpG Cs are degenerated, the authors detected higher levels of methylation compared to primers designed by MethPrimer in the promoter regions of two mouse genes (myogenin and PSEN1). However, no efforts were made to verify the completeness of bisulfite conversion of the template DNA. Although control samples with complete methylated and unmethylated DNA were included in the analysis, the use of PCR products as a control for unmethylated DNA is problematic since incomplete conversion is less an issue for PCR products compared to genomic DNA. In other words, complete conversion of the control DNA does not necessarily mean that genomic DNA was also completely converted. If that were true, the non-CpG methylation detected could be merely an artifact.

I do caution that researchers should keep in mind potential non-CpG methylation. Before more evidence is available about the prevalence of non-CpG methylation and its function, researchers should care more about potential incompleteness of bisulfite conversion and subsequent overestimation of methylation levels in the DNA.