Why can the mRNA strand made during transcription be thought of as a mirror image

Chromatin: steroid hormone-induced covalent modifications of histone tails

New mRNA and protein synthesis in hypothalamic neurons are required for estrogenic facilitation of lordosis behavior (reviewed in Pfaff, 1980). This discovery led to a long series of experiments in which genes that had two properties were identified: (1) their mRNA levels were increased in hypothalamic neurons after E administration and (2) their gene products would foster female reproductive behaviors (summarized in Pfaff, 1999). In turn, the emphasis on gene–behavior relationships sprung two very surprising observations: the loss of one gene (ERalpha) could cause females to be treated like males and to respond like males (Ogawa et al., 1996), and secondly, an individual gene could have opposite behavioral effects in male brains as compared to female brains (summarized in Ogawa et al., 2004). All of these and other observations (e.g. Ogawa et al., 1998a, 1998b) caused us to focus on hormone-facilitated gene transcription.

The initiation of transcription is widely understood to involve changes in chromatin, the set of proteins that guard access to DNA by transcription factors such as ERs (illustrated in Figs. 3–5). Chromatin, the principal structural protein associated with DNA molecules, has a potential regulatory role in gene expression. Chromatin is the principal structural protein around which DNA is wrapped and it is made up of four canonical proteins termed H2A, H2B, H3 and H4, the structure of which have been largely conserved over evolutionary time. Broadly, DNA exists in two states, (1) euchromatin, wherein DNA is relatively loosely coiled around chromatin and which is associated with transcriptionally active genes. This is in contrast to (2) heterochromatin where DNA is tightly wrapped around histone peptides and thus the underlying DNA is inaccessible to the transcriptional machinery. These states were long considered developmentally programmed and static. However, these broad categories are significantly complicated by dozens of covalent modifications of histone tail regions which can more subtly regulate gene expression. Modification of histone tails with the additions of acetyl, phosphate, methyl, sumo and ADP ribosyl groups at specific amino acid residues alone and in combination has specific regulatory effects on the genes associated with modified chromatin. The complexity of this system has lead Professor C. David Allis and colleagues here at Rockefeller to suggest that these covalent modifications of histone peptides taken together form a ‘histone code’ which provide important epigenetic regulation of gene expression. Therefore, with respect to female sex behavior, it was natural for us to try and draw histone modification into E-caused transcriptional changes in the VMH neurons essential for E-dependent lordosis behavior (Weil et al., 2009). Results showed that treatment with 17β-estradiol rapidly methylates H3K9 in the VMH, and also that treatment with 17β-estradiol rapidly increased the acetylation status of histone H4 proteins in the VMH. These changes are highly likely to be relevant to lordosis behavior because double-identified ER-alpha positive cells in the ventromedial hypothalamus are shown by our new results to exhibit H4 acetylation. We have further reason to believe that histone acetylation within VMH neurons is causally linked to female reproductive behavior, because microinjections of the histone deacetylase inhibitor into the VMH potentiates estrogen-induced lordosis behavior and reduces rejection behaviors (Weil et al., 2009).

Similar considerations apply to maternal behavior. The optimal pattern of hormone exposure for the onset of maternal behaviors in female laboratory mammals comprises a sudden drop of progesterone levels in the blood coupled with high levels of estrogens in the blood. Surprisingly, however, it had not been determined exactly what duration of estrogen exposure would be minimally sufficient for the facilitation of maternal behaviors. We have just finished the appropriate experiments by ovariectomizing female mice 11 days before tests of maternal behaviors and, at the same time, implanting progesterone capsules subcutaneously, which will subsequently be removed two days before behavioral testing. The question is: how long must estrogens circulate in progesterone-withdrawn mice, for maternal behaviors to occur? The answer was surprising: estrogens administered as little as two hours before testing could enhance maternal behaviors in progesterone-withdrawn female mice (Murakami et al., unpublished data, 2010). The advantage of this short requirement of estrogen treatment is that it permits us to concentrate on a short time window during estrogen-triggered behaviorally important transcriptional changes. In turn, this knowledge tells us exactly when to look for histone chemical modifications in preoptic area neurons, a project that we are finishing now (Murakami et al., unpublished data, 2010). From unpublished experiments, we know that the schedule of E administration consistent with the initiation of maternal behaviors is associated with global changes of histone chemistry in the preoptic neurons essential for maternal behavior, and that these changes include both methylation and acetylation (Murakami et al., unpublished data, 2010), but we neither have tried to draw these histone modifications into transcriptional arguments, nor have we proven that these histone modifications are essential for the initiation of maternal behavior. Those experiments will be initiated soon in this laboratory.

Although the data and molecular concepts of histone modifications fit very clearly with access to estrogen-sensitive genes by ligand-activated transcription factors, ERs, phenomena in neuroendocrine neurons are not always so simple. The changes in methylation status of specific histones caused by acute stress in hippocampal neurons (Hunter et al., 2009) have not, so far, been mimicked by corticosterone injections as they might have been expected to be. Alternate routes of stress effects, therefore, are being explored currently.

Our current emphasis on the chemical modifications of histone tails in hypothalamic and preoptic neurons brings us through the next mechanistic step in demonstrating how gene–behavior causal relations occur. After all, Ogawa et al. proposed that deletion of a single gene could cause a female mouse to be treated like a male and to behave like a male (Ogawa et al., 1996). Altogether, knocking out the gene for ER-alpha caused a panoply of changes in sexual, aggressive and maternal behaviors (Ogawaet al., 1997, 1998a, 1998b). Taken together, the Ogawa et al. (2004) findings proved that the same gene could have opposite behavioral effects (on aggressive behavior) in females as that same gene has in males.

In sum, because of the clear relationships between patterns of gene expression and behavior in neuroendocrine systems, we predict that sex steroid effects on reproductive behaviors – both sexual and maternal – will provide some of the most chemically detailed and functionally important examples of histone modifications in nerve cells.

Regulation of mRNA Synthesis

In prokaryotes, mRNA synthesis can be controlled simply by regulating initiation of transcription. In eukaryotes, mRNA is formed from a primary transcript followed by a series of processing events (e.g., splicing, capping, polyadenylation). Eukaryotes regulate not only transcription initiation, but also the various later stages of processing.

An important aspect of mRNA regulation is determined by the degradation of mRNA molecules; i.e., translation can occur only as long as the mRNA remains intact. In bacteria, mRNA molecules have a lifetime of only a few minutes, and continued synthesis of mRNA molecules is needed to maintain synthesis of the proteins encoded in the mRNAs. In eukaryotes, the lifetime of mRNA is generally quite long (hours or days), thereby enabling a small number of transcriptional initiation events to produce proteins over a long period of time.

Metabolic pathways normally consist of a large number of enzymes; in some cases, the individual enzymes are used in a particular pathway and nowhere else. In these cases, it may be efficient to regulate expression of either all or none of the enzymes in the pathway. In bacterial systems, all enzymes of the pathway are encoded in a single polycistronic mRNA molecule called an operon, and synthesis of the single mRNA produces all the enzymes. In eukaryotes, common signals for transcription of different genes may be used, or the primary transcript can be differentially processed to yield a set of mRNA molecules, each of which encodes one protein. In eukaryotes, regulation of synthesis of the primary transcripts simultaneously regulates synthesis of all the gene products.

In prokaryotes, gene expression fluctuates in response to the environment because bacteria must be able to respond rapidly to a changing environment. However, due to the differentiation of cells of the higher eukaryotes, changes in gene expression are usually irreversible—for example, in the differentiation of a muscle cell from a precursor cell. Eukaryotes can change the number of copies of a gene during differentiation and, in this way, regulate the level of gene expression.

Synthesis of genomic and subgenomic rnas

Genome replication and sg mRNA synthesis (“transcription”) proceed through minus-strand intermediates. The genome serves as a template for the synthesis of full-length minus-strand RNA, from which in turn new genome copies are produced, but it is also believed to be the template for the synthesis of sg minus-strand RNA species (vide infra). The synthesis of viral RNAs is highly asymmetrical as plus-strand RNAs are produced in fast excess.

A hallmark of nidovirus transcription is the production of a 3′-coterminal nested set of sg mRNAs. The sg mRNAs of corona-, arteri- and bafiniviruses are chimeric, that is, comprised of sequences that are non-contiguous in the viral genome. Each carries a short 5′ leader sequence of 55–92, 170–210 nt, and 42 nucleotides, respectively, which is identical to the 5′ end of the viral genome. It was established early on that leader and “body” sequences are not joined through splicing, but via a process of discontinuous RNA synthesis. A key observation was the presence of mirror-copy nested sets of sg minus-strand RNAs in corona- and arterivirus-infected cells. Combined experimental evidence from biochemical and reverse genetics analyses indicates that these sg minus-strand RNAs are in fact the templates for sg mRNA synthesis. Replicative intermediates (RI)/replicative forms with sizes corresponding to the different sg mRNAs were shown to be actively involved in transcription. According to the prevailing 3′-discontinuous extension model, the discontinuous step occurs during the production of sg minus-strand RNAs and entails attenuation of RNA synthesis at the TRSs, followed by a similarity-assisted copy choice RNA recombination event. In corona-, arteri- and bafiniviruses, a TRS is present immediately downstream of the genomic leader sequence. It is believed that, during minus-strand RNA synthesis, the replicase complex upon encounter of an internal TRS dissociates from the template and is transferred to the 5′ end of the genome, guided by sequence complementarity between the anti-TRS on the nascent strand and the genomic TRS. Reinitiation and completion of RNA synthesis would then result in a chimeric minus-strand that in turn would serve as a template for uninterrupted (continuous) synthesis of 5′ leader-containing sg mRNAs.

Discontinuous sg RNA synthesis is not a trait of all nidoviruses. Ronivirus sg mRNAs lack a common 5′ leader and thus apparently arise from non-discontinuous RNA synthesis. Toroviruses employ a mixed transcription strategy; of the four sg RNAs, only RNA 2 carries a 15–18 nt 5′ leader derived from the 5′ end of the genome, whereas the others do not. It is likely that sg mRNAs are transcribed from sg minus-strand templates also in toro- and in roniviruses. Here, the conserved sequence elements (TPs) preceding the 3′-proximal genes might serve dual roles as signals for premature termination of minus-strand synthesis and as promoters for plus-strand production. The torovirus S gene, expressed from mRNA 2, lacks a TP. Apparently, transcription-competent minus-strand sg RNAs are produced by inclusion of a complementary copy of the 5′-terminal genomic TP via a similarity-assisted RNA recombination process analogous to that seen in corona- and arteriviruses.

5.3.2 Transient Analysis

The transient analysis of deterministic differential models aims at determining the time-dependent values of the variables over a given time window.

Consider for instance the mRNA synthesis and degradation model given by Eq. (5.21). If we want to compute the value of the variable mRNA(t) for any t ≥ 0, we can directly solve the differential equation to find the following analytical closed-form expression for mRNA(t):

(5.26)mRNA(t)=k1k2+mRNA(0)−k1k2e−k2⋅t.

This function can be plotted for different values of the initial abundance of mRNA. Fig. 5.4 shows curves of the transient behavior of mRNA(t) for three different initial conditions when the rate constants k1 and k2 are set to 10.0 and 5.0, respectively. The plots in Fig. 5.4 allow us to see the convergence of the transient values to the steady-state solution that we determined in the previous section, and that is given by k1/k2 = 2.0. Notice that setting the initial abundance of the mRNA to the equilibrium value of 2.0 immediately nullifies the derivative in Eq. (5.26) and therefore provides a transient behavior that stays aligned with the steady-state value (curve for mRNA(0) = 0 in Fig. 5.4).

The possibility of finding explicit solutions to systems of ordinary differential equations rapidly diminishes with the increase in model complexity. Therefore, the approach we described above, which allows us to determine the transient behavior of the modeled species by finding analytical closed forms for the continuous variables, cannot be generalized. Again, the transient analysis of models is usually done via numerical approximated methods, which starting from the initial known value of the variables (the value at time t = 0) compute an approximation for the values of the variables at time δt. This step-ahead computation is iteratively repeated to determine a polygonal curve approximating the real-valued variables over the time interval of interest in the analysis. Methods belonging to this family are usually (and improperly) referred to as numerical integration methods for ordinary differential equations.

A variety of approaches exist which differ on the amount of information used in computing the approximated value of the variable at time t + δt given its previously computed value at time t. The simplest one, which provides the essence of this numerical computation, is Euler’s method, which computes the value of the variable at time t + δt from the slope of the tangent line at time point t, by considering that the real unknown curve will not be significantly different from the tangent inside the interval (t − δt,t + δt) if a sufficiently small value of δt is used in the approximation step.

The approximation can be made arbitrarily precise by reducing δt, which, however, increases the number of evaluations required to determine the approximating polygonal in the interval of interest. A natural variation of Euler’s method is therefore one that considers an adaptive step, based on an estimation of the speed with which the distance of the tangent from the real value of the variable increases. Estimation of this distance requires the evaluation of higher-order derivatives for the variables. A commonly applied method of this type is the fourth-order Runge–Kutta method.

Numerical integration of systems of deterministic differential models provides a quick and effective means for transient analysis. Consider again the model described in Fig. 5.3 and the following differential equation model that describes the variation of the abundance of protein P over time:

(5.27)ddtP(t)=k1KnP(t)n+Kn−k2⋅P(t).

The first term on the right-hand side of Eq. (5.27) provides an abstract modeling of the repression effect exerted by the dimeric form of protein P on the synthesis of new molecules of protein P. Because the repression is realized by a cooperative effect (via the dimerized form of P), we can model the reduction of the synthesis rate by the negative Hill function, which provides a decreased synthesis rate as the abundance of P increases. Obviously, with such a complex form of the differential equation terms, it is not possible to find an explicit analytical solution for P(t). Resorting to numerical integration provides the time-dependent behavior shown in Fig. 5.5, where we plot the transient evolution of P(t) for three different values of the parameter K of the Hill function. Because K defines the amount of P that determines the half strength of the repression effect, increasing values of K result in less pronounced reductions in the synthesis rate, which results in increasing levels of protein P. In all cases, we assumed that the initial amount of protein P is zero — that is, P(0) = 0.

Finally, we mention a special family of numerically integrators that have been developed to cope with stiffness, a common characteristic of differential equation models of biological systems. A model is stiff when it represents different phenomena which occur on very different timescales. In the model of a biological system, this is an easily encountered situation if we are representing events that proceed at very diverse speeds. For instance, consider the enzymatic reaction described in Eq. (5.13). The association and the dissociation of the enzyme and substrate is typically much faster than the catalysis step, and this is the main rationale that justifies our abstracting the two reactions to obtain the Michaelis–Menten form of the rate shown in Eq. (5.14). This means that inside the same model, some variables will be changing very fast and some will be changing much more slowly, which in turns calls for the necessity of our using very small integration steps to compute sufficiently accurate results. To solve stiff systems of ordinary differential equations, specific algorithms have been proposed, such as LSODA [212], which are implemented by most computational tools (see, eg, [208]) and are included in well-known numerical libraries such as ODEPACK [213].

2.5 Kinetics measurements by spectrofluorometry

Real-time monitoring of mRNA and protein synthesis was carried out on a fluorescence spectrophotometer (Cary Eclipse, Varian). The PUREfrex reaction solution was transferred to a 15-μL cuvette (Hellma) positioned in a Peltier thermostated four-cell holder maintained at 37 °C. Spinach and YFP fluorescence signals were detected every 30 s in the high-voltage mode at (excitation/emission wavelengths) 460/502 and 515/528 nm, respectively, to minimize fluorescence overlap.

For each conditions tested, at least three independent (no pooling of the PURE system reaction solution) fluorescence curves have been generated and representative kinetics are displayed (Fig. 4).

The cuvettes need to be thoroughly cleaned immediately after each experiment to maximize data reproducibility. Washing was done by successively filling the cuvette with Hellmanex 2%, KOH 1 M, nuclease-free water, EtOH 100% for 1 min in a bath sonicator, applying three washes with nuclease-free water in between the different reagent treatments.

43.5 DNA-Dependent RNA Transcription

From the previous sections, it becomes clear that the chromatin structure imposes profound and ubiquitous effects on almost all DNA-related metabolic processes, including transcription, recombination, and DNA repair and replication. In this section, the main focus is on the steps in transcription regulation, with emphasis on chromatin structure. The regulatory steps are similar to both mRNA and microRNAs (miRNAs) until 3′ end processing, since most of the protein-coding genes and miRNAs undergo RNA Pol II–mediated transcription.

The typical Pol II transcription cycle for a naked DNA molecule begins with the binding of activators upstream of the core promoter, including the TATA box and transcription start site (Table 43.1). This event leads to recruitment of adaptor complexes, such as SAGA, with multiple enzymatic activities [25], which leads to the binding of the TATA binding protein (TBP) to the TATA box, facilitating the assembly of TFIID upstream of the core promoter. After TFIID binds to the TATA box via the TBP, five more TFs and RNA polymerase combine around the TATA box in stages to form a closed preinitiation complex (PIC). TFIIH then melts 11–15 bp of DNA to position the single-strand template in the Pol II cleft (open complex), which initiates RNA synthesis. The carboxy-terminal domain (CTD) of Pol II is phosphorylated by the TFIIH subunit during the first 30 bp of transcription and loses its contacts with the general TFs before the elongation stage.

Table 43.1. Characteristics of Common Promoter Elements

Promoter ElementDistance from TSSConsensusFunctionTATA box28–34 bp upstreamTATAABinding of PIC complexInitiator (Inr)at +1 TSS siteYYANWYYRecruitment of PIC complexDownstream promoter element (DPE)28–32 bp upstreamRGWYYDirect PIC to TSSTFIIB recognition element (BRE)Upstream of TATA boxSSRCGCCIncrease or decrease in transcriptionCpG islandTSS, intragenic, or intronic(CG)nTranscription of ubiquitous genes

However, the PIC alone drives only the basal rate of transcription. Other proteins known as activators and repressors, along with any associated co-activators or co-repressors, are responsible for modulating transcription rate. The phosphorylated CTD begins to recruit the factors that are important for productive elongation and mRNA processing [26]. While TFs can share targets and bind to sequence-specific binding sites in the context of free DNA, they have to adopt a different strategy to achieve binding to sites buried in chromatin. Early biochemical experiments suggested that TFs can bind to nucleosomal DNA in a cooperative manner before nucleosome disassembly. However, later studies showed that chromatin-remodeling complexes help in recruiting TFs to chromatin sites [27]. Additionally, it has been demonstrated that promoters have a lower density of nucleosomes than the coding regions [28].

Considering the amount of DNA directly contacted by Pol II and general TFs, the structure of the nucleosome seems to pose a significant obstacle in PIC formation [29]. A number of studies have proposed that, upon gene activation, nucleosomes are dissembled and reassembled as the transcription is completed. In other cases, nucleosomes are not displaced; rather, promoters assemble partial PICs, excluding Pol II and TFIIH. This implies that when template DNA engages into the Pol II active site, DNA–histone contacts may be broken (Figure 43.4(a,b)).

Transcription elongation begins when Pol II releases from general TFs and travels into the coding region. This event signals the recruitment of the elongation machinery, which includes the factors involved in polymerization, mRNA processing, mRNA export, and chromatin function. The main difference between the factors that aid initiation and those that aid elongation is that the elongation factors require direct or indirect binding to the CTD of Pol II [26]. The Pol II CTD undergoes two major phosphorylation changes during elongation: Serine5 is phosphorylated by TFIIH at the 5′ end of the open reading frame (ORF), and Serine2 is phosphorylated by the CTK kinase, as Pol II transits toward the 3′ end. This controls and couples elongation with alterations in chromatin structure. For example, an evolutionarily conserved multi-subunit complex called PAF is loaded onto the Serine5-phosphorylated CTD and regulates the binding of other CTD-associated chromatin regulators, such as the histone H3K4 methyltransferase Set1 complex for elongating Pol II [30]. In contrast, the histone methyltransferase Set2 targets primarily Serine2-phosphorylated CTD while Pol II travels toward the 3′ end of the ORF [31].

The first change that a nascent pre-mRNA transcript undergoes is 5′ capping. When a transcript reaches about 20–30 nt in length, a 7-methylguanosine cap is added to the 5′ end, which protects the nascent pre-mRNA from degradation by the cap-binding complex (CBC) [32] (Figure 43.4(c)). Pre-mRNA capping is followed by splicing. Although the exons are relatively small and embedded in large intron sequences, the splicing machinery recognizes the exons, removes the introns from the pre-mRNA molecule, and ligates the exons to form a mature mRNA. The main features of intron in splicing include (1) the exon–intron junction at the 5′ and 3′ ends of introns (5′ ss (single strand ) and 3 ss); (2) the branch site sequence located upstream of the 3′ ss; and (3) the polypyrimidine tract located between the 3′ ss and the branch site. All types of pre-mRNA splicing take place within the spliceosome—a large complex composed of five small nuclear RNA (snRNA) molecules (U1, U2, U4, U5, and U6)—and as many as 150 proteins. Each of the five snRNAs assembles with proteins to form small nuclear ribonuclear protein complexes (snRNPs) [33].

Once splicing is completed, the exon-junction complex (EJC) proteins are deposited at the site of exon fusion (Figure 43.4(d)). Capping and splicing are both important for the recruitment of the transcription-export (TREX) complex. This is primed by the 3′ end processing of transcripts, aided by transcription termination. This step is crucial not only to release Pol II from its template but also to ensure that a pool of polymerases is available for reinitiation or new transcription. Termination can also prevent formation of antisense RNAs, which can interfere with normal pre-mRNA production, thereby preventing aberrant gene expression. Transcription termination can occur anywhere from a few base pairs to several kb downstream from the 3′ end of the mature mRNA. This process is coupled to 3′-end processing of the pre-mRNA, namely polyadenylation, which refers to the addition of a poly(A) tail.

In mammals, cleavage and polyadenylation occur 10–30 nts downstream from a conserved hexanucleotide, AAUAAA, and 30 nt upstream of a less conserved U- or GU-rich region. This signal is recognized by the cleavage and polyadenylation specificity factor (CPSF). CPSF binds other pre-mRNA binding protein components such as poly(A) polymerase and poly-A binding protein (PABP) (Figure 43.4(e)). PABP protects the pre-mRNA from exonuclease degradation by binding the poly(A) tail, and it is required for correct and efficient poly(A) tail synthesis [34]. Several other factors, including the CTD of polymerase help link 3′-end processing to transcription. Another interesting observation is that H3K36 methylation often decreases near the poly(A) site during or prior to Pol II release, suggesting that H3K36 methylation plays a role in transcription termination [35].

The final step in mature mRNA synthesis is the nuclear export of mRNA, now bound with protein components and called mRNP. Following completion of proper nuclear processing and the recruitment of an export receptor, an mRNP is considered to be export-competent. The TREX complex is highly conserved and is essential for mRNP export (Figure 43.4(f)). This export-competent mRNP is targeted to the nuclear pore complex (NPC) via its export receptor. For some transcribed genes, the positioning of the respective chromatin region near the NPC might facilitate export by physically linking the processes. The export receptor docks at the NPC via special proteins called FG-Nups that mediate the movement of the mRNP through the NPC by some type of facilitated diffusion. When the mRNP reaches the cytoplasmic site of NPC, specific proteins from mRNPs are lost by ATP hydrolysis and recycled for the next round of mRNA binding. The binding of cytoplasmic factors occurs immediately with the entry of the 5′ end of the transcript into the cytoplasm; the CBC is replaced with eIF4e, ribosomes bind, and translation begins before the entire mRNP has been extruded from the NPC (Figure 43.4(g–j)).

Section Key Points

RNA Polymerase II transcribes mRNA, which undergoes nuclear processing, including 5′ capping, splicing, and 3′ polyadenylation before being exported into the cytoplasm to be translated.

RIFAMPICIN

Regulation of Gene Expression

There is a perfect coordination of mRNA synthesis and protein accumulation during leaf development and chloroplast maturation. Much of the control has been found at the level of gene transcription. Light has a central role in regulating the biogenesis of the photosynthetic apparatus during chloroplast development, including switching on the expression of genes encoding enzymes of C3 photosynthesis. These processes appear to be complex, involving many regulatory proteins and cross-talk between the different photoreceptor signaling pathways.

The expression of C3 photosynthesis genes in leaves is also modulated by environmental and metabolic signals. High levels of glucose or sucrose reduce levels of a number of Calvin cycle mRNAs, including those encoding the small subunit of rubisco, SBPase, and FBPase. This feedback mechanism, acting at the level of transcription, might be important for source–sink regulation in the plant. Photosynthesis genes respond remarkably to nutrient status, particularly nitrogen and phosphorus levels, and this is influenced again by the carbohydrate status. Thus, an interaction between carbohydrate and nutrient status may be a long-term strategy to control carbon metabolism.

Why can the mRNA strand made during transcription be thought of as a mirror image of the DNA strand from which is was made?

What happens when mRNA is transcribing DNA?

Why does mRNA have to be copied in transcription?

What direction is the mRNA transcript manufactured?