We used the previously characterized construct pCAG-EGFP/RFP-miRNAint (G/R-miRNA) to express miRNAs in transgenic mice () 
. This construct first expresses enhance green fluorescent protein (EGFP), which enables rapid screen of transgenic mouse lines where the transgene is active in the desired tissue. Upon induction by Cre, the EGFP gene is excised, leaving the promoter to drive the expression of red fluorescent protein (RFP) and miRNA. The RFP provides a convenient indicator for the level and location of the miRNA expression 
. We targeted two genes with this construct. One was the progranulin gene and the other was the E1k subunit of á-ketoglutarate dehydrogenase complex (KGDHC). Both genes are involved in neurodegenerative diseases. Loss of function mutations in the progranulin gene cause frontotemporal dementia (FTD) and a decrease in KGDHC activity is associated with Alzheimer's disease (AD) 
. As a control for possible non-specific effects associated with the overexpression of miRNA, we used a construct that expresses a scrambled miRNA (miR-Scr) that does not target any specific gene.
Cre-loxP-based conditional miRNA expression causes microcephaly.
Transgenic mice were generated by pronuclear injection. Two constructs for progranulin knockdown (miR-PGRN1 and 9), one construct for E1k knockdown (miR-E1k) and one construct with scrambled miRNA sequence (miR-Scr) were injected. Transgenic lines were screened based on a semi-quantitative estimation of EGFP fluorescence in the brain. One transgenic line from each of the two miR-PGRN constructs, two lines from the miR-E1k and six lines from the miR-Scr were selected and propagated based on their relatively high levels of EGFP expression in the brain (). The transgenes were inherited at a frequency expected from Mendelian inheritance without any overt phenotypes. To investigate the effect of PGRN and E1k knockdown in the brain, we crossed the offspring of these lines with nestin-Cre driver mice, which express Cre in neural progenitor and glial cells beginning at ~E10.5 day 
. In half of the transgenic lines, the double transgenic progeny developed microcephaly. The brains of these double transgenic mice (Dtg) weighed approximately 50% of those of their single transgenic (Stg) and wild type (Wt) littermates (; ). The overall brain structures were maintained but were proportionally smaller than those of the Stg and Wt littermates. For example, the cortical cell layers appeared normal but thinner (). Similarly, the hippocampal architecture appeared intact but smaller compared with the non-transgenic brains (). The microcephaly phenotype was observable at birth. However, in most lines the number of the double transgenic progeny was not different from what was expected from Mendelian inheritance (), indicating that the microcephaly did not cause embryonic lethality in most lines.
Summary of basic characteristics of the ten transgenic lines.
Genotype composition of the progenies from the crossings of G/R-miRNA and nestin-cre.
The microcephaly might be caused by the knockdown of PGRN or E1k expression, but this appeared unlikely. In the miR-PGRN9 line which had microcephaly, there was only ~25% knockdown of PGRN mRNA (). This is unlikely to cause microcephaly because this phenotype was not observed in the progranulin knockout mice 
. In the miR-E1k lines, line 32 developed microcephaly but line 15 did not. However, line 32 had only ~20% knockdown of the E1k mRNA whereas line 15 had ~75% knockdown. Thus, target knockdown could not explain the cause of microcephaly.
It has been shown that high levels of siRNA or shRNA expression can cause toxicity in mice 
. Therefore, high levels of miRNA expression might also be toxic and cause microcephaly. To test this possibility, we first analyzed the RFP expression in the double transgenic progeny since the RFP expression was driven by the same promoter that drove expression of the miRNA (). We found that the levels of RFP did not correlate with the microcephaly phenotype. For example, the transgenic lines with the highest RFP levels, such as miR-E1k15, miR-Scr105, miR-Scr109 and miR-Scr113, did not develop microcephaly, whereas the lines with low levels of RFP expression, such as miR-E1k32 and miR-Scr103, developed microcephaly (). To verify the miRNA levels, we measured the transgene-encoded miRNAs by real time RT-PCR and confirmed that most transgenic lines with the highest miRNA levels did not develop microcephaly, whereas the lines with relatively low levels of miRNA did (). Some transgenic lines also had a significant level of leakage in miRNA expression before crossing with nestin-Cre (), but none showed microcephaly during line propagation in crossings with the wild type mice. Thus, the microcephaly phenotype did not correlate with high levels of miRNA expression.
To further seek the cause for this phenotype, we measured the transgene copy numbers in all the transgenic lines using real time PCR of the genomic DNA. Interestingly, four out of five transgenic lines with microcephaly phenotype had high transgene copy numbers, while all lines without microcephaly had low transgene copy numbers (). This observation raised the possibility that aberrant Cre-mediated recombination at the site of the multiple transgene integration might cause the microcephaly. One possible scenario is that a cellular toxicity might be derived from the presence of inverted loxP sites, i.e. loxP sites in a head-to-head or tail-to-tail configuration. For example, transgenes with inverted loxP repeats can cause chromosome loss and death of proliferating cells during Cre-mediated recombination 
. However, the prevalent assumption has been that multiple transgene copies are integrated in a tandem head-to-tail array with only rare exceptions of inverted repeats 
To test the hypothesis that Cre-mediated recombination of the loxP-containing transgenes that are integrated in an inverted manner leads to microcephaly, we used PCR to amplify all possible types of transgene-transgene junctions, including the head-to-tail and the inverted head-to-head and tail-to-tail junctions. We detected head-to-tail PCR products from six out of ten transgenic lines (; ). Four of these six lines (miR-PGRN9, miR-E1k32, miR-Scr103 and 97) had 9 or more transgene copies and also carried inverted transgene integration. The other two lines (miR-E1k15 and miR-Scr109) had only head-to-tail integration and carried 2–3 transgene copies. One line miR-Scr35 carried two transgene copies that are integrated head-to-head. Three lines (miR-PGRN1, miR-Scr105 and 113) had low transgene copies and no junctional PCR products. These lines were likely to carry a single copy of the transgene. To ascertain that the PCR products are truly derived from the expected junctions, we sequenced the junctional PCR products and were able to obtain sequences from most of these fragments (; ). Because of the frequent deletion of the sequences from the original constructs at the junctions, we often observed multiple sequences from one PCR band (). We were able to read sequences from some of the PCR bands that contained two sequences, but were unable to read sequences from several other bands. This was most likely due to the presence of multiple sequences within those bands.
Detection of inverted transgene arrays in transgenic lines.
Notably, all the lines that were positive for the inverted transgene integration, including miR-pgrn9, -miR-E1k32, miR-Scr103, miR-Scr97 and miR-Scr35, had the microcephaly phenotype. In contrast, all the lines that were negative for the inverted transgene integration did not have the microcephaly phenotype. Thus, the presence of the inverted transgene integration predicted the microcephaly. Additionally, of the seven lines that had two or more copies of the transgene, five had inverted transgene integration. The two lines that did not have inverted transgenes had low transgene copies. Thus, contrary to the prevailing assumption that the transgenes are integrated into the genome in a head-to-tail tandem array with rare exceptions 
, the inverted transgene integration is common in mice that have multiple transgene copies.
Finally, to determine whether the inverted transgene repeats caused cell death in proliferating cells, as has been shown previously 
, we examined the double transgenic embryos with both the G/R-miRNA and nestin-Cre transgene. We detected increased cell death in the ventricular zone and the cortex during the neurogenesis at 14.5 and 16.5 days post coitum (dpc) (). This observation confirms that the inverted transgene repeats containing loxP sites caused death of the proliferating cells during neurogenesis, which led to the microcephaly.
Increased apoptosis in the proliferating cell population in prenatal mouse brains.