Chronological Changes in MicroRNA Expression in the Developing Human Brain

Objective MicroRNAs (miRNAs) are endogenously expressed noncoding RNA molecules that are believed to regulate multiple neurobiological processes. Expression studies have revealed distinct temporal expression patterns in the developing rodent and porcine brain, but comprehensive profiling in the developing human brain has not been previously reported. Methods We performed microarray and TaqMan-based expression analysis of all annotated mature miRNAs (miRBase 10.0) as well as 373 novel, predicted miRNAs. Expression levels were measured in 48 post-mortem brain tissue samples, representing gestational ages 14–24 weeks, as well as early postnatal and adult time points. Results Expression levels of 312 miRNAs changed significantly between at least two of the broad age categories, defined as fetal, young, and adult. Conclusions We have constructed a miRNA expression atlas of the developing human brain, and we propose a classification scheme to guide future studies of neurobiological function.


Introduction
MicroRNAs (miRNAs) form a growing class of endogenous non-coding RNA molecules that modulate gene expression posttranscriptionally. Like transcription factors, an increase in the number of miRNAs strongly correlates with greater organismal complexity [1]. miRNAs form elaborate and sophisticated regulatory networks, where a given miRNA can influence the stability or translatability of hundreds of mRNA targets [2], and numerous miRNAs can act in concert to repress a common target. While transcription factors act as ''switches'' to initiate broad developmental transitions, miRNAs may act downstream to finetune genetic regulatory programs. An influential model posits that highly expressed miRNAs act to quell the deleterious effects of leaky transcription, while moderately expressed miRNAs buffer fluctuations in expression at weak or suboptimal promoters [3].
Generation of cellular diversity during mammalian brain development requires precise coordination of gene regulatory networks, with integral involvement of miRNAs. Lineage-specific expression signatures of cultured astrocytes and neurons [4] implicate miRNAs in neural cell fate specification. Neurobiological functions have been attributed to specific miRNAs. For example, miR-124 promotes neuronal differentiation [5], and miR-134 is involved in dendritic branching [6]. Global ablation of miRNAs is achieved in model systems by knocking out the essential miRNA processing enzyme, Dicer. In zebrafish, maternal zygotic Dicer knockout embryos undergo axis formation and body patterning, but exhibit profound defects in brain morphogenesis [7]. Likewise, selective inactivation of Dicer produces morphogenetic CNS abnormalities in conditional knockout mice, including microcephaly and reduced elaboration of dendritic branches [8]. While neuronal progenitors remain viable in the absence of miRNAs, cytoarchitectural abnormalities arise from failures to differentiate and propagate newborn neurons [9]. Dysregulation of miRNAs has been implicated in the etiology of schizophrenia [10,11,12,13], as well as age-related neurodegenerative disorders [14,15,16].
Early expression studies uncovered a subset of brain-specific or brain-enriched miRNAs [17]. Molecular profiling studies in rodents showed that miRNAs are dynamically regulated during brain development [18], with a ''chronological wave'' of sequentially expressed miRNA classes [19]. Since these landmark studies, hundreds of additional miRNAs have been discovered, including many that are specific to humans. In the present study, we have performed comprehensive miRNA expression analysis in human postmortem brain samples representing fetal, early postnatal, and adult time points. We measured the expression of all known miRNAs (miRBase 10.0), as well as 373 novel, putative miRNAs. Classifying miRNAs based on temporal expression profiles may provide insight into their regulation and potential neurobiological functions.

Sample selection and preparation
Frozen brain tissue samples were obtained from the NICHD Brain and Tissue Bank for Developmental Disorders, housed at the University of Maryland. Brain sectioning (described at http:// medschool.umaryland.edu/btbank/NICHD-BTB-ProtocolMethods. asp) began with separation of the cerebral hemispheres, followed by removal of the midbrain/pons/cerebellum. The remaining cerebrum (from the left hemisphere) was sectioned coronally at approximately 1 cm intervals. A total of 48 frozen cerebral tissue samples were obtained. Fetal samples originated from subjects ranging in gestational age from 14 weeks to 24 weeks. Additional tissue samples were obtained to represent early postnatal time points. The majority of tissue samples originated from African American individuals ( Table 1). Three samples were excluded from microarray expression analysis, two due to poor RNA integrity measures and one due to uncertain sample identification. All remaining samples had RNA integrity measures .7. The adult time point is represented by two commercially available total RNA products. FirstChoice Human Brain Total RNA (Ambion) is a high-quality RNA sample derived from an 81 year old adult male, and FirstChoice Human Brain Reference RNA (Ambion) is pooled from twenty-three male and female Caucasian donors with an average age of 68.
Tissue homogenization and RNA extraction procedures were performed using the mirVana PARIS kit (Ambion) according to the manufacturer's protocol. RNA yield and quality were determined using a NanoDrop ND-1000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA) and Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA).

Microarray preparation, data acquisition, and analysis
Dual-channel microarray expression analysis was performed using the NCode Human miRNA Microarray V3 (Invitrogen, Carlsbad, CA), which contains 710 probes for validated human miRNAs from miRBase release 10.0 as well as 373 Invitrogen novel human miRNAs. On each array, a single Cy3 labeled total RNA sample was competitively hybridized against a Cy5 labeled control pool. The control pool was a mixture of all RNA samples, where each developmental time point was equally represented, and each individual sample was equally represented within a given time point. For each channel, 1 mg of total RNA was labeled with the Label ITH miRNA Labeling Kit, Version 2 (Mirus Bio, Madison, WI) according to manufacturer's protocol. Samples were hybridized to the array at 37uC overnight using the hybridization buffer included with the labeling kit.
Image analysis was performed on a GenePix 4000B microarray scanner. A marked difference in pixel intensity was observed between Sanger miRNAs and the novel features exclusively present on the NCode V3 arrays, with many of the latter approaching saturation. Photomultiplier tube (PMT) settings were adjusted so as to balance signal intensity in the red and green channels, and maximize pixel intensity without saturation at 10% power. Arrays were then rescanned at 100% power without PMT adjustment. The 10% and 100% scan data were subsequently merged using methods outlined in Methods S1 in File S1, which resulted in substantial improvements in data quantity and quality. Images were exported and processed using the Acuity 4.0 software package. Microarray data has been deposited in the Gene Expression Omnibus (GEO) database under accession number GSE45126. Data filtering parameters were established to include only features with (1) at least 70% of pixels greater than 2 standard deviations above background in either the red or green channel, (2) regression R 2 greater than 0.5 to ensure feature uniformity, (3) less than 20% of pixels saturated in both the red and the green channels, and (4) no flags upon visual inspection. Normalization was performed with the LOWESS (locally weighted regression) method, and expression log ratios were used for statistical analysis. Merging and filtering resulted in a final set of 464 genes. Prior to significance testing, distribution assumption checks were performed, as outlined in Methods S2 in File S1.
464 genes which passed the filtering criteria were analyzed using ANOVA (analysis of variance) and backward model selection, and genes whose expression differed between any subset of the fetal, young, and adult time points were identified (method described in more detail in Methods S3 in File S1). A simulation based false discovery rate (FDR) estimation similar to the SAM procedure [20] was used to adjust for multiple testing. 2500 simulated datasets were used, and it was found that a p-value of 0.001901 controled for the FDR at a rate of 1%.

Preparation of TaqMan miRNA arrays
Sample pools for each unique developmental age were constructed by combining equal amounts of extracted total RNA (samples used in pools indicated in Table 1). Reverse transcription reactions were performed using 600ng total RNA input and reagents from the TaqMan miRNA Reverse Transcription Kit and Megaplex Primer Pools (Applied Biosystems). Reactions were performed (without preamplification) according to the Megaplex Pools protocol (Applied Biosystems, Carlsbad, CA), and thermal cycling was performed in a PTC-200 DNA Engine (MJ Research). Real-time PCR reactions were performed using the TaqMan Human miRNA Array set v2.0 in a 7900HT real-time PCR system (Applied Biosystems, Carlsbad, CA). A single array set was run for each sample pool.

Analysis of TaqMan miRNA arrays
Relative expression analysis was performed using the SDS 2.3 software. Using the 2 2DDCt method [21], expression levels of all miRNAs were normalized to the geometric mean of RNU44 and RNU48 snoRNAs and mammalian U6 snRNA. A stringent threshold was set to exclude miRNAs which did not show robust expression in at least one time point. This threshold was arbitrary, and corresponded to a raw Ct of .29 at all time points. Next, the fold change in expression was calculated relative to the 14 week time point in order to identify miRNAs with differential expression within the prenatal time points. Similar analysis was performed, using the 5-98 day time point as a reference, for the postnatal time points. Additionally, the 14 week time point was used as a reference to calculate fold-changes at all prenatal and postnatal time points. The distribution of these fold changes was used to identify miRNAs that exceeded two standard deviation fold change at any given time point. Only mature, guide-strand derived, miRNAs were included in these analyses.

Results
MicroRNA expression analysis was performed using the NCode Human miRNA microarray (Invitrogen), which contains 710 Model classes were refined to take into account the direction of expression differences between sample types, resulting in 13 unique model classes ( Table 2). Model classes 1 and 2 are models for which the young and adult mean expression levels coincide, and the fetal mean expression differs. A substantially greater number of miRNAs belong to the class where fetal expression levels are lower than that of young and adult samples (60 miRNAs     Figure S1 in File S1), and those features suspected of being larger than 200 bp are indicated in Table 2.
Analysis of microarray data from fetal samples alone was performed to determine if the gestational age (in weeks) had a significant impact on expression. After adjusting for multiple testing it was found that there were no miRNAs with significant changes in expression during the fetal time period. This may be due, in part, to the small sample sizes representing each developmental time point or to the lack of sensitivity and limited dynamic range of glass-slide microarrays. Similarly, an analysis was run on the young samples alone to determine if age (in days) had a significant impact on miRNA expression. After adjusting for multiple testing, there were no miRNAs with significant changes in expression.
Samples comprising each distinct time point were mixed to form 10 sample pools ( Table 1), and further miRNA expression analysis was performed using the TaqManH MicroRNA Array v2.0. Many of the validated, Sanger miRNAs that were detected and quantified on TaqMan arrays fell below the threshold of detection on the NCode microarrays. Therefore the TaqMan data only partially validated the NCode array data, but also allowed for analysis of other miRNAs and for more refined analysis of expression changes within the pre-and post-natal time points. All raw real-time qPCR data is presented in Table S1 in File S1.
The numbers of miRNAs with particular expression fold changes between groups are presented in Table 3. Similar to what was observed in the microarray data set, these data demonstrate that global miRNA expression is lower in fetal than young and adult brain tissue, given the greater number of miRNAs with decreased fold change in the fetal versus young and adult samples. Although pooling of samples precludes a wide-range of formal statistical analyses, we sought to identify meaningful patterns of change within the pre-and post-natal time points. MiRNAs which show peaks of expression at a particular gestational or post-gestational time point may be viable candidate miRNAs involved in specific developmental processes. A Q-Q plot of fold changes for all miRNAs at all time points (see Figure S2 in File S1) revealed expected trends. After setting an arbitrary threshold to exclude miRNAs that did not show high expression in at least one time point, the standard deviations for fold changes were calculated. In analysis of pre-natal time points, miRNAs were identified which showed a .2 standard deviation fold change at any time point (relative to the 14 week time point) resulting in a list of 14 miRNAs ( Table S2 in File S1). Particular candidate miRNAs emerge from this analysis. For example, hsamir-449a/449b show a dramatic change in expression during 14-20 week gestation, with a peak at week 17. Another pattern is represented by 216b, which shows highest expression at week 14, then decreasing expression until it reaches a low at week 17, followed by increasing expression through weeks 18, 19, and 20. This trend for increasing expression does not extend into postgestational timepoints, where expression is negligible.
In the individual analysis of post-natal time points, the earliest time point (5-98 days) served as the reference against the 1.5 and 4.5 year time point. Again, miRNAs with .2 SD fold change from 5-98 day time point were identified, resulting in a list of 15 miRNAs ( Table S3 in File S1). Particular miRNAs show clear patterns of change during early childhood. For example, hsa-mir-135b shows dramatically decreasing expression during early childhood, while at prenatal time points hsa-mir-135b showed much higher but somewhat consistent expression. Mir-542-5p on the other hand, shows consistently low expression at prenatal time points, spikes at earliest childhood time point, then declines.
Examination of the pre-and postnatal time points collectively, using 14 week as reference time point, revealed 56 miRNAs with .2 SD fold change between pre-and postnatal samples ( Table S4 in File S1). This larger list is a reflection of the more robust collective differences in miRNA expression between fetal stages and early childhood, as compared to the more subtle changes within the fetal and early childhood periods. This analysis reveals interesting candidates, including the mir-17-92 cluster located on chromosome 13, and the paralogue clusters on chromosome 7 (mir-25/93/106b) and the X chromosome (mir-106a/18b/20b). Collectively, this miRNA family shows a trend for modestly decreasing expression at prenatal time points, and dramatically lower and decreasing expression at postnatal time points. In contrast, mir-132 displays modest and steady expression levels during prenatal time points, with a striking increase in expression during early childhood.

Discussion
We reported 312 out of 464 detected features with temporally distinct expression patterns in the cerebrum using microarray technologies. Additionally, we presented comprehensive miRNA expression profiles of pooled samples at each time point using TaqMan technology. Utilizing the microarray data, we found that differentially expressed species formed ten unique model classes with contrasting fetal, young, and adult expression profiles. Notably, many miRNAs fit a model that did not depict a consistent trend in expression along the fetal to adult timeline, but instead sharply increased or decreased at early postnatal stages. For example, the ubiquitously expressed let-7 family members are conserved from nematodes to primates, and are well-known markers of a terminally differentiated state [22,23,24]. In our study, expression levels of most let-7 family members dramatically peaked at the adult time point, as expected, but we also observed a surprising decline in expression at early postnatal stages relative to fetal. We would also like to draw attention to several miRNAs with well-characterized roles in CNS development. miR-124 and miR-9 are brain-specific miRNAs [25] that are highly up-regulated at the onset of embryonic neurogenesis. Over-expression of these miRNAs in cultured cells promotes adaptation of a neuronal fate, while down-regulation has the opposite effect [26]. Developmental expression profiling of the murine CNS revealed 12 miRNAs (miR-9, miR-17-5p, miR-124a, miR-125a, miR-125b, miR-130a, miR-140, miR-181a, miR-199a, miR-205, miR-214, miR-301) with significantly higher expression at embryonic versus postnatal time points. Expression profiling was not performed in the fully mature mouse. In the present study, 4 miRNAs (miR-9, miR-124, miR-125b, miR-181a) fit the model F = A.Y. The decrease in expression at early postnatal stages is consistent with observations in the mouse; however, expression levels appear to rebound in the adult human brain. It is intriguing to note that, with regard to several neuronal miRNAs as well as the let-7 family, the early postnatal brain appears to adopt a more immature, dedifferentiated state. However, we did not observe expression of the pluripotency markers miR-290, -295, -293, -294, or the -302 family [27] at any developmental time point. With the exception of certain neurogenic brain regions, such as the dentate gyrus of the hippocampus [28], neurogenesis is mostly complete by the 18 th week of gestation. Our data support a role for miRNAs in later stages of neuronal development, such as the formation and pruning of synapses [29], which continues throughout infancy and early childhood. At these developmental stages, we observe transient down-regulation of several well-characterized neuronal miRNAs, with concomitant spikes in expression of 31 other Sanger miRNAs (model classes 5 and 9). We investigated 373 novel, putative miRNAs that are exclusively present on the NCode Human miRNA microarray V3 (Invitrogen), and found that 194 of these exhibited significant temporal expression changes. Sequences were mapped to genomic hairpins and validated by qPCR with probes for the ''mature'' region. Subsequent bioinformatic analysis was performed using the miRDeep algorithm, which evaluates enrichment of sequence reads derived from the stem versus loop region of genomic hairpins, suggestive of miRNA processing [30]. This analysis revealed that most of the small RNA sequences were unlikely to be classic miRNAs. By comparing expression profiles of various RNA size fractions, we determined that a significant proportion of the novel sequences were likely derived from larger non-coding RNAs. Several classes of ncRNA are dynamically regulated during development, and they may serve especially prominent roles in the developing brain (reviewed in [31]). For these reasons, all microarray features were retained for significance analysis, regardless of whether they represented miRNAs or other types of RNA. Model classes 1 (F,Y = A) and 7 (F,Y,A) are comprised of 67 and 10 genes, respectively. Of these 77 genes with lowest expression during fetal stages, 74 are novel ncRNAs. From our RNA size fractionation experiments, 29 of the 312 genes with significant developmental changes in expression level were flagged as potentially being larger ncRNAs, and 26 of these flagged species were novel ncRNAs from either model class 1 (F,Y,A) or 3 (F = Y,A). Taken together, these data implicate a class of longer ncRNAs that are progressively up-regulated during development and may serve important neurobiological functions in the adult brain.
TaqMan expression profiling was more sensitive than the glassslide microarray technology which we used, and allowed us to examine trends within the prenatal and postnatal time points. Several miRNAs were identified which have noteworthy developmental expression patterns. For example, the mir-17-92 cluster on chromosome 13 and paralogue clusters on chromosome 7 and the X chromosome show decreasing expression throughout pre-and postnatal timepoints, consistent with a report in the porcine developing brain [32]. Mir-132 shows a marked increase in post-natal compared to pre-natal time points, and appears to increase throughout early childhood. In vivo experiments in the mouse hippocampus have implicated mir-132 in the maturation of excitatory synapses formed by newborn neurons [33], and mir-132 dysregulation has recently been implicated in schizophrenia [34]. Mir-449a/b show a striking expression pattern during prenatal development, with a peak of expression at week 17. Mir-449 has been shown to control ciliogenesis in the human airway epithelium [35], and has been proposed to regulate choroid plexus development based on expression patterns in the mouse brain [36]. The peak in mir-449 expression at week 17 may be related to ciliogenesis in the ventricular system, which if impaired, can impact ion transport and cerebrospinal fluid production and flow [37].
In summary, we have established the first miRNA and noncoding RNA expression atlas of the developing human brain. Expression signatures varied significantly between fetal, early postnatal, and adult brain tissue samples, analogous to patterns of temporal regulation previously observed in animal model systems. Despite attempts to match samples with respect to demographic variables, factors pertaining to sample selection and handling can skew the results of post-mortem expression studies. Experimental and statistical methods were adopted to optimize sensitivity and accuracy of measured miRNA expression levels, yet some caveats still remain. The coronal sections of cerebrum from which RNA was extracted are heterogeneous in terms of cell type. Expression levels were averaged across multiple samples from a given developmental age to minimize regional effects, but the relative contribution of specific cellular populations to detected expression levels could conceivably vary across time points. Also, average post-mortem interval was higher for postnatal compared to prenatal samples (Table 1), however our previous work demonstrates that few microRNAs (8%) have expression which is influenced by post-mortem interval [38]. Additionally, the preand postnatal sample set used is biased towards African-American individuals while the adult timepoints were from Caucasian individuals, which could conceivably confound results if there are large ethnic differences in microRNA expression in the brain. However, data from other tissues indicate that ethnic differences are likely to be modest [39]. Despite these caveats, the valuable expression data presented here and the proposed classification scheme should guide future studies of miRNA expression and function in both normal and pathogenic development.