DLX5 expression is monoallelic and Dlx5 is up-regulated in the Mecp2-null frontal cortex

To the Editor

There has been a controversy as to whether DLX5 is imprinted in human brain and lymphoblastoid cell lines (LCLs), and whether expression of Dlx5 and Dlx6 (Dlx5/6 exist as a bigene cluster [1]), is dysregulated in the Mecp2-null mouse frontal cortex.

In the September issue of Am J Hum Genet. (Vol 81: 492–506, 2007) [2], Schüle et al. published a paper entitled ‘DLX5 and DLX6 expression is bi-allelic and not modulated by Mecp2 deficiency.’ This study concluded that DLX5 was not imprinted in normal human LCLs nor in human brain, contradicting the conclusions of Okita et al. (Genomics 81:556–559, 2003) [3] and Horike et al. (Nature Genetics 37:31–40, 2005) [4]. Schüle et al. also claimed that expression of Dlx5 and Dlx6 in mouse brain varies greatly among individual mice and is not necessarily up-regulated in the frontal cortex of Mecp2-null mice. These findings directly contradict those of Horike et al. We have now repeated and verified our previous experiments that are relevant to the points raised by Schüle et al. Further, we have also expanded our analyses to determine why Schüle et al., were not able to reproduce our findings. Here are our findings in response to their assertions.

(1) Schüle et al. presented quantitative RT-PCR (qRT-PCR) data obtained from frontal cortex tissue samples of male wild-type and Mecp2-null mice from several litters. They reported that the expression levels of Dlx5, Dlx6 and the imprinted gene Peg3 were all highly variable among individual mice, independent of genotype (0.5–3.5-fold difference relative to a single wild-type mouse selected among several litters). Schüle et al. therefore concluded that the findings of Horike et al.– that there is a 2-fold increase in expression of Dlx5 and Dlx6 in Mecp2-null mice compared to wild type littermates – were the result of ‘biological noise.’

Considering that these genes are tightly regulated during development, we were surprised at their observations that their expression was completely random even among wild type animals from the same litter. We repeated these experiments using three sets of littermates of Mecp2-null and wild-type mice (42 or 51 days old) from an independent breeding colony of Mecp2-heterozygous mice (generated by Adrian Bird's laboratory) [5] kindly provided by Masayuki Itoh (National Institute of Neuroscience, Japan), who carefully bred the Mecp2±heterozygous mice to retain the original phenotype. In all five Mecp2-null mice examined using Gapdh as a control, we were able to reproduce the ∼2-fold (1.7–2.6-fold) increase in expression of Dlx5 in Mecp2-null mice, compared with wild-type mice (Fig. 1A), just as originally reported by Horike et al. We similarly verified up-regulation of Dlx6 in the same brain subregions of the Mecp2-null mice tested (representative data are shown in Fig. 1B). We further examined expression levels of Peg3, which Schüle et al. also report as having high variability in expression. In contrast to their findings, we observed consistent expression levels with a small standard error of the mean (S.E.M.) between mice of the same genotype and saw similar levels in the frontal cortex between wild-type and Mecp2-null mice (Fig. 1C). Expression levels of genes in brains are best compared among littermates, since mice from different litters often show some variability in gene expression. Nonetheless, in our study, difference in expression levels of the control Gapdh typically do not exceed more than 20%, even among mice from different litters. Therefore, our results consistently show non-random expression of Dlx5 and Dlx6 relative to Gapdh among littermates, accurately reflecting the genotype of individual mice within each litter. In the frontal cortex of Mecp2-null mice, Dlx5 and Dlx6 are moderately up-regulated compared to wild-type mice, while there is no change in expression for Peg3.

Fig. 1.

Upregulation of Dlx5/Dlx6, but not Peg3, in the frontal cortex of Mecp2-null mice. qRT-PCR results for (A) Dlx5, (B) Dlx6, and (C) Peg3 in the frontal cortex of Mecp2+/y (blue) and Mecp2-/y (red) littermate mice, relative to Gapdh. The absence of genomic DNA in the cDNA samples was confirmed. Inset in (A) shows electrophoresis of RNA used in qRT-PCR analysis. (D) qRT-PCR results for Dlx5 in the frontal cortex (ctx) and striatum (st) of wild-type mouse brain. (E) qRT-PCR results on Dlx5 in the striatum of Mecp2+/y (blue) and Mecp2-/y (red) littermate mice. Error bars indicate the S.E.M. Primers sequences for Dlx5: 5′-CAGAACGCGCGGAGTTG-3′ and 5′-CCA-GATTTTCACCTGTGTTTGC-3′, Dlx6: 5′-TGGCTGCTTCCTTAGGACTGA-3′ and 5′-CTTAGAGCGCTTATTCTGAAACCA-3′, and for Peg3: 5′-CCTAT-GAGTATGGGCCCTCCTA-3′ and 5′-CATTCGTACAGTGGGATGTGCTT-3′.

The differences between our results and those of Schüle et al. likely arose from differences in reproducibility in dissection of the frontal cortex and in the purity/quality of RNA; RNA degradation can dramatically alter qRT-PCR results. Because Dlx5 is expressed at high levels in striatum (greater than 4-fold compared to the frontal cortex (Fig. 1D)), we took great care to ensure that neighbouring striatum (or other tissues) did not contaminate the frontal cortex sample by isolating a narrow region corresponding to the primary and secondary motor cortex (M1 and M2). We also isolated striatal tissue and found Dlx5 was again moderately up-regulated in this region in Mecp2-null mice.

One caution is that after more than 2 years of breeding of Mecp2 heterozygous mice to ensure a higher chance of survival of Mecp2-null(-/y) pups in our laboratory, (either by selecting female mice that are better caretakers, or using a surrogate mother) we found that the phenotype of many Mecp2-null(-/y) male mice became milder, with a much extended life span and delayed onset of disease. Mecp2-null(-/y) males produced by such mouse lines often show a less than 2-fold increase in Dlx5 and Dlx6 (typically a 1.2- to at most 1.5-fold increase compared to control mice). This decline in the fold increase in these mice might be attributed to changes in epigenetic status at these gene loci, and we are currently studying this effect.

(2) Another major point of controversy is that Schüle et al. conclude that DLX5 expression is biallelic in human LCLs and human brains, whereas Horike et al.[4] and Okita et al.[3] independently conclude that it is imprinted in human LCLs, and Okita et al. previously found that it is monoallelically expressed in human brains as well. A total of 18 LCL samples were examined between the two groups (Horike et al. and Okita et al.), among which 16 showed monoallelic expression. In all eight LCL samples for which the parental origin of DLX5 was known, DLX5 was expressed from the maternal allele [3].

Schüle et al. have examined four human LCLs, and observed only one sample with monoallelic expression (Control 1 in Fig. 2 of Schüle et al.), while they observed biallelic expression in all the rest. One important point that needs to be raised here is that if DLX5 expression were truly biallelic in nature, one cannot expect to obtain data consistent with monoallelic expression, especially after many cycles of amplification. In contrast, if it were monoallelic in nature, one may still obtain sequencing data that could be misinterpreted as biallelic. There are at least two reasons for this: (1) the expression level of DLX5 is low in human LCLs and requires many PCR cycles, and (2) human LCLs are unstable and can become biallelic under non-optimal culture conditions (more passages than recommended and non-optimal freezing/thawing conditions). Therefore, even if a gene is predominantly expressed from one parental allele, a low-level expression from the other parental allele may still be amplified resulting in sequencing data suggesting biallelic expression for many LCLs. Therefore, the quality of both the LCLs and the RNA are absolutely critical in obtaining reliable data. We repeated the imprinting analyses, using two newly purchased normal human LCLs from Coriell Cell Repositories with minimum prior passages and processed them exactly according to the supplier's instructions. We again observed that DLX5 was monoallelically expressed in these normal human LCLs. Based on the clean, non-overlapping sequencing data of cDNA derived solely from one parental allele, and the equal contribution of sequences from genomic DNA of both parental alleles (reflected by equal intensity C and T peaks at the point where the sequence mismatch occurs between paternal and material alleles), it is apparent that Dlx5 is monoallelically expressed in normal LCLs (Fig. 2). We have reproduced the same quality of data (Fig. 2A) as those published previously by Horike et al.

Fig. 2.

Monoallelic DLX5 expression in normal individuals and biallelic DLX5 expression in two individuals with Rett syndrome (RTT). The C7 or C8 mononucleotide repeat polymorphism in the 3′ UTR of DLX5 is shown. Sequence traces of genomic DNA amplified by PCR and those from the RT-PCR products for DLX5 are shown for LCLs from normal individuals and individuals with RTT. See Methods section in Horike et al.[4]. A) LCL cell samples were prepared after minimum passages without additional freezing and thawing after purchased from Coriell Cell Repositories. B) The same normal LCLs samples used in A) were subjected to several additional passages and one-time freeze/thawing before use.

To show that LCLs are prone to losing the imprinting status at the DLX5 locus, we deliberately continued culturing these same cells under non-optimal conditions by several additional passages. After one freeze/thaw cycle, these cells grew at a much slower rate, and DLX5 expression became biallelic for both cell lines (Fig. 2B). Therefore, we always used freshly purchased LCLs from the supplier with minimum passages to obtain a sufficient number of cells for all experiments involving LCLs. In contrast to normal LCLs, LCLs from two individuals with Rett syndrome (RTT), expanded with a minimum number of passages, exhibited biallelic expression of DLX5 as reported by Horike et al., and the data were successfully reproduced in Fig. 2A.

In the Schüle et al. paper, it is difficult to draw definitive conclusions regarding the mono- or biallelic expression of DLX5 based on Fig. 2 of Schule et al.[2] (for human LCLs) and 3 of Schule et al.[2] (for human brain) because of the high level of background sequence noise in multiple cDNA and genomic DNA samples, and the unequal C and T signal intensities at the mismatch for some genomic DNA samples. In mouse brain, Dlx5 is mostly biallelic [4, 6]. However, in the human brain, Okita et al. from Oshimura's lab showed clean sequencing data in all three brain samples, indicating a mostly monoallelic nature of DLX5 expression. Again, such data could not have been obtained unless transcripts are derived mainly from one allele.

(3) Schüle et al. used somatic-cell hybrid lines to show that human DLX5 and DLX6 were expressed from both parental alleles. However, the particular cell lines they used have an inherent problem that make them unsuitable for imprinting analyses: individual cell lines have variable copies of multiple human chromosomes. One cell line has lost chromosome 7 that contains DLX5/DLX6. Furthermore, they all contain an additional active X chromosome, where Mecp2 is located. We do not know how cells respond to multiple copies of chromosomes especially at the level of gene expression; dysregulation of dosage-sensitive expression of genes may cause unpredictable outcomes. Furthermore, in their experiments, important positive controls were missing, viz., verification of the imprinted status of bona fide imprinted genes located on the human chromosome 7, such as PEG10 and PEG1/MEST. M. Oshimura and colleagues, on the other hand, established a series of human monochromosomal hybrids retaining a single chromosome of defined parental origin [7]. This in vitroassay system is suitable for imprinting analysis and has led to the identification of several imprinted genes [8–10], including DLX5, and the data were verified with human LCLs [3]. Collectively, the experiments of Schüle et al. using a somatic hybrid cell system do not support their claim that DLX5 is not imprinted.

We have carefully analyzed the study of Schüle et al., repeating many of their and our own experiments, and stand by our original conclusion that DLX5 is imprinted in human LCLs, and that this imprinting is lost in some but not all patients with RTT. Furthermore, we stand by our conclusion that chromatin looping defects occur in the brains of Mecp2-null mice. Our chromatin immunoprecipitation (ChIP) studies identified Mecp2 binding sites in the Dlx5-Dlx6 region. To verify the results, ChIP experiments have to be performed carefully with all proper controls. Our data, showing Mecp2 binding to the Dlx5-Dlx6 locus in mouse brain, were validated by a recent ChIP-chip study in a human neuronal cell line [11].