Team:Heidelberg/Project Synthetic promoters

Synthetic Promoters

The central question of the synthetic promoter project is: Are we able to make specific promoters for use in mammalian cells synthetically?

Or, going even further: Are we able to develop a standard method for creating promoters of

• Defined strength
• Defined response
• Defined pathway integration

How we got this to work... Read on!

Abstract

Promoters are the key regulators of gene expression. Possessing promoters which are active under a desired condition, at a desired strength and in a specified tissue is of great value for Plant Biotechnology, Gene Therapy and fundamental research in Bioscience. Therefore, it has become a desire to synthetically construct promoters responsive to a variety of pathways. We explore two ways to the synthesis of promoters: On one hand, we have developed a bioinformatical model and database (HEARTBEAT) describing the structure of promoters responsive to user-defined inputs. On the other hand, we have developed a biochemical method for the synthesis of randomized promoter libraries. Using this method, we have created a library of constitutive promoters of varying strength. Also, we have created libraries of promoters putatively responsive to a variety of pathways. We have screened these libraries for functional, pathway responsive promoters and present a detailed characterization of a NF-κB responsive promoter of our making. We finally discuss ways to combine randomized biochemical synthesis and bioinformatical modeling to propose a method towards the generation of promoters of complex regulation (i.e. by multiple pathways).

Introduction

Controlling gene expression in a condition and tissue specific manner is a desirable tool for a wide variety of applications ranging from fundamental bioscience research to gene therapy. Having this idea in mind it is inevitable to consider the starting point of gene expression: the promoter. Most efforts of obtaining any promoters focus on cloning them from nature. In eukaryotes, this approach leads to several problems. The one major issue is the complexity of eukaryotic systems resulting in cross-talk between pathways. In particular, promoters tend to respond to several transcription factors and are thus activated by more than one pathway. That way, one cannot determine which signal originally induced the promoter. Alongside this specificity, induction under a desired combination of conditions and at defined strengths would be valuable implementations.

Therefore, efforts have emerged to synthetically construct promoters. Two concepts of synthetic promoters in mammalian cells co-exist independently from each other. One is the concept of "genetic switches" (see [2] for a recent review) - promoters which can specifically be induced by a stimulus mammalian cells are usually insensitive to, e.g. tetracycline [3]). Much fewer efforts have been put into developing promoters sensitive to endogenous signals (referred to as "synthetic promoters" in the rest of this article). Such promoters are of very high value for a broad variety of applications. Three examples should demonstrate this: First, in virotherapy for cancer and other diseases, it has become a desire to express toxic genes only in affected cells (reviewd in [4]). For example, breast cancer cells are characterized by high levels of estrogen receptor. Constructing a promoter which is active only at high estrogen receptor levels (plus, maybe, only in cells which are irradiated, as ER can be very active in other tissues of the female reproductive tract[17]) might therefore help developing novel breast cancer therapies. Second, biologists studying pathway interactions are in need for transcriptional assays, that is promoters which are specifically activated by a single transcription factor. Third, the concept can be transferred to plants, where synthetic promoters can be very valuable, as plant biotechnology is always in need for novel tissue- or development-specific promoters.

Three approaches exist to construct synthetic promoters responsive to endogenous factors. First, the structure of promoters is modeled by generating large data sets describing the relative spacing and coincidence of transcription factors (reviewed in [5]). To our knowledge, such predictions have not been tested in vivo. Second, promoters are generated by randomly or repeatedly cloning response elements upstream of a core promoter. In our understanding, repeated cloning of response elements works well [6] and is frequently applied, but no suggestions exist on how to apply this strategy to the generation of more complexly regulated promoters. The random creation of promoters works well to generate constitutive promoters [7] and was even applied to broadly identify activating elements [8], but no promoters of specific regulation have been described for this approach. A third approach is the randomization of spacer elements between transcription factor binding sites, which is applied to generate libraries of promoters of varying strength [9], [10].

In order to be able to design synthetic promoters, an understanding of natural promoters is required. Mammalian promoters can be subdivided into several "domains". The core promoter is the binding site of the basal transcription machinery, i.e. RNA polymerase and associated factors. Core promoters differ in composition, but are more or less similar for most genes (reviewed in [11]). The main regulatory domain is the proximal promoter, which is where regulatory elements bind. It can be very large (4kb), meaning that some transcription factors regulate transcription despite being very far away from the RNA polymerase. This is mainly possible because of the three-dimensional structure the DNA adopts. In addition to this, there are even more distal elements that are referred to as "enhancers" and "silencers". A further challenge is that some transcription factors are not able to initiate transcription on their own, but rather they require other transcription factors for their activity [5].

Results

RA-PCR, a method for the generation of randomized promoter libraries

Figure 1: The method of RA-PCR

We have developed a standard method (termed "Random Assembly PCR / RA-PCR") for the construction of randomized promoter libraries. We modified Assembly PCR [12] to create promoters with randomized spacing and frequency of Transcription Factor Binding Sites. It relies on using different oligos containing a transcription factor binding site (or random DNA) plus two annealing sequences (see Figure 1 for a comprehensive explanation of the method). We use two sets of oligos, one for the top strand, one for the bottom strand. The oligos for each strand have the same annealing sequences (which are complementary to the annealing sequences of the other strand). If these oligos are pooled, they will randomly anneal to each other, thus generating randomized repeats of the transcription factor binding sites of interest at varying spacing. In order to be able to clone the construct, we also add two stop oligos (termed stop 5' and top 3') which contain only one annealing sequence, plus a cutsite (SpeI 5', HindIII 3'). Double-stranded DNA is created by running a seven-cycle PCR, and amplified by a 25-cycle PCR. Then, the resulting (proximal) promoter is cloned 5' of a core promoter (we used the core promoter of JeT [10]) by inserting it into pSMB_MEASURE ([http://partsregistry.org/wiki/index.php?title=Part:BBa_K203100 Part:BBa_K203100]), the promoter measurement plasmid we developed (from there, it can be excised like any standard biological part in a submission plasmid). Thus, a mixture of different promoters in the same plasmid backbone is generated. These can then be transformed into bacteria. Each colony represents a single putative promoter, which can be transfected into mammalian cell. 24 hours after transfection cells are exposed to the conditions of interest and control conditions, serving as a reference. Promoters active under the desired conditions, but not under control conditions, are selected for further characterization.

Please see a detailled protocol for RA-PCR below.

Generation of a library of constitutive promoters

As a first application of RA-PCR, we have created a library of constitutive promoters. We performed RA-PCR on oligos containing binding sites for some well known generally activating transcription factors (Sp1, Ap1,CREB, NF-Y) which we identified from literature search [8],[7],[10]. We also added NF-κB responsive oligos as NF-κB has non-specific activity and is therefore used by a variety of viral constitutive promoters, e.g. the HIV promoter [13]. We picked 24 colonies, two of which we dismissed after a test digest (not shown). Figure 3 shows the sequence analysis of some randomly selected clones and demonstrates that RA-PCR is able to generate randomized repeats of oligos. We then measured the activity of the clones we picked by applying the Concept of Relative Expression Units (REU) we developed. Figure 2 shows that we have been able to create a library of promoters of varying strength, some of which have an expression strength higher than JeT (which was not accomplished by JeT's developers, although attempted [10]). Such a library is of great value for fine-tuning gene expression levels.

Figure 2: A library of constitutive promoters created by RA-PCR Promoters were analyzed by the standards developed in the Measurement part of our project in HeLa cells.

Figure 3: RA-PCR generates randomized repeats of transcription factor binding sites. Sequence analysis of clones of constitutive promoters generated by RA-PCR. Transcription factor binding sites are marked in color, random sequences in light grey. Annealing sequences are shown in bold and bold italic.

Generation and screening of a library of promoters putatively responsive to NF-κB

RA-PCR was conducted with oligos containing a NF-κB binding site, plus a small number of "general activators" (NF-Y, Sp1, Ap1, CREB) . Box 1 demonstrates how the oligos were designed from a frequency matrix. 33 clones were picked, miniprepped and transfected. NF-κB was then induced by the addition of TNF-α (2.5µM) for 10 hours, and left uninduced as a control. The plate was then scanned by TECAN, an automated fluorescence plate reader. TECAN is very imprecise on eukaryotic cells, and the arbitrary fluorescence we measured is not proportional to REU or another precise measure of mammalian promoter activity, but it can serve as a rough indicator of promoter induction. The result (Fig.4) shows that most clones appear not to be induced by NF-κB / TNF-α, whereas others are induced at varying levels of strength. Considering the sequence analysis of some randomly selected clones (Fig.5), this result is not intuitive, as most sequences contain a NF-κB binding site, but it demonstrates that simply cloning repeats of a Transcription Factor Binding Site in front of a core promoter will not necessarily work.

We picked clone 31 for further characterization in REU.

Figure 4: A library of putative NF-κB responsive promoters created by RA-PCR Promoters were induced by TNF-α in U2OS cells and screened by TECAN (automated plate reader).

Figure 5: Sequence analysis of putative NF-κB responsive promoters

Box 1: RA-PCR allows for synthesis of promoters responsive to imprecisely described transcription factors. Considering the graphical representation of NF-κB's frequency matrix shown above (source: [http://jaspar.cgb.ki.se/ JASPAR]), the oligo can be designed in order to represent this matrix, instead of a static NF-κ B binding site. A sensible representation of this matrix would be GGGRHTTYCC (for the IUPAC nucleotide code, refer to [http://www.bioinformatics.org/sms/iupac.html bioinformatics.org]). Most oligonucleotide manufactures provide the option to synthesize such mixtures of individual oliogs without further cost. As our method is PCR-based (unlike other methods such as [6] and [7]), we are able to synthesize even promoters responsive to badly-described transcription factors

Figure 6: Characterization of NF-κB-responsive clone 31. NF-κB was induced by TNF-α 2,5µM for 10 hours or left uninduced, REU for 3 independent experiments, three replicates each.

Characterization of a NF-κB responsive promoter

We characterized clone 31 by Flow cytometry (3 independent experiments, each experiment having 3 replicates) and found it to be upregulated by approx. 100% (Fig. 6). We also did a timecourse measurement of induction (Fig. 7). To verify our assumption that promoters created by RA-PCR are specific to the pathway of interest, we tried to induce clone 31 by a wide variety of conditions (full medium, full medium lacking growth factors / serum, minimal medium / starvation and pPARγ induction by Thiazolidinedione and found no change in indcution levels (Fig. 8) Also, we show that the promoter is active only if p65, a component of NF-κB, localizes to the nucleus (Fig. 9 and 10).

Figure 7: Timecourse Characterization of NF-κB-responsive clone 31. NF-κB was induced by TNF-α 2,5µM, REU were measured after 1, 7 and 10 h.	Figure 8: NF-κB responsive clone 31 responds only to TNF-α and is not induced by other conditions. U2-OS Cells transfected with clone NF-κB 31 were subjected to a variety of conditions (full medium DMEM+++, full medium lacking growth factors / serum DMEM++, starvation by minimal medium and pPARγ induction by Thiazolidinedione) and did not show any significant induction or repression.
Figure 9: NF-κB-responsive-promoter is not active if NF-κB is localized in the cytoplasm. A p65-mcherry fusion protein (kind gift of nathan brady) was co-transfected to cells with clone 31 (expresses GFP under the control of our NF-κB regulated promoter)	Figure 10: NF-κB-responsive-promoter is active if NF-κB is localized in the nucleus. A p65-mcherry fusion protein (kind gift of nathan brady) was co-transfected to cells with clone 31 (expresses GFP under the control of our NF-κB regulated promoter), cells were then induced by TNF-α as described.

RA-PCR can generate promoters responsive to a variety of pathways

We performed RA-PCR to construct promoters putatively responsive to transcription factors as diverse as p53 (DNA damage sensor), pPARγ (metabolism & diabetis), SREBP (Sterol nutrition), HIF (hypoxia) and Estrogen receptor . While screening these promoters we found the following:

Figure 11: pPARγ responsive promoters induced by Thiazolidinedione in U2OS cells.

• For pPARγ, we identified by screening two clones which appear to be responsive to the anti-diabetis drug 2,4-Thiazolidinedione. We roughly characterized these promoters by a triple TECAN read relative to JeT (Fig. 11)
• For p53, induction of the pathway by the Topoisomerase inhibitor Camptothecin (an anti-cancer drug) turned out to be difficult as is severly harms the cells and makes promoter indcution levels difficult. We therefore attempted to normalize screening conditions to number of living cells by Hoechst-Staining. We found that some promoters appeared to be strongly dowregulated by Camptothecin and therefore experimented with a variety of conditions inducing by p53 by different pathways, at different phosphorylation sites, but where unable to obtain a conclusive picture.
• For HIF, we failed to induced the condtions sufficiently to achieve promoter activation. We discuss below how screening can be improved.
• For SREBP and Estrogen, we encoutered technical problems during promoter synthesis (probably damaged HinDIII enzyme) and therefore were unable to produce enough clones for sufficient screening. For SREBP, we therefore cloned two natural, SREBP-upregulated promoters we had at hand and submitted them to the registry (Main Article: Natural promoter subproject).

HEARTBEAT, a model describing promoter structure

Main article: HEARTBEAT

Figure 12: Probability density function for the distribution of VDR-Binding sites along an ideal promoter as modelled by HERATBEAT

In figure 13 the spatial distribution of VDR (Vitamin D receptor) binding sites within 140 natural promoter sequences is shown as an example. The size of each bin equals the number of VDR-TFBS within a range of 20 bps. The solid line represents the probability density function (pdf). Here, the maximum of the pdf is located 54 bps upstream to the TSS indicated by the vertical line. Natural promoter sequences usually exhibit multiple TFBS which implies dependencies between different TFs according to their binding behaviour to the DNA. Figure 11 shows the frequency distribution of coincidental appearing TFBSs if VDR is present. The highest peak represents VDR itself. The next three highest peaks are Kid3 (inhibitory), WT1 and AP-2 (stimulatory). In total, together with VDR, there are over 300 different TFBS coincidentally present. Both plots represent data deduced from the Heartbeat-database which enable a well-defined synthesis of promoter sequences.

Figure 13: Frequency of other transcription factors occuring together with VDR. 680 transcription factors were examined, of which the displayed 340 show coincidence at least once.

Discussion

The results shown above demonstrate the potential of RA-PCR towards the synthesis of any promoter. Even by analyzing modest amounts of clones for each individual pathway, we were able to obtain promoters of a wide variety of strength and inducibility. Also, we were able to obtain constitutive promoters of greater strength than JeT, which has not been possible before[10].

Many insights about promoter regulation are possible by analyzing different promoters created by RA-PCR. For example, NF-κB clone 3 and clone 11 (see figure 4 and 5) differ only in the positioning of the single response element (RE), but still, induction strength differs threefold. This gives hints about Nf-κB's binding preference. A systematic study of promoters generated by RA-PCR and their strength could therefore be used to develop a comprehensive model of transcriptional regulation. Nao, kann man hier dein model verlinken und mit einem oder zwei sätzen elegant beschrieben?

Improving RA-PCR

Screening conditions and induction strength:

As noted above, we experienced difficulties inducing some of the pathways (namely, HIF and p53). From our cell culture work, we learned that finding the ideal timepoint of induction for a certain pathway and the ideal conditions is very difficult even with literature at hand. Also, one would expect a much higher induction than the one observed for the NF-κB responsive clone we describe. Our induction levels might be low because NF-κB has a high constant actvity [16], especially if the cells encounter "rough" cell culture conditions. This also explaines the (small) induction of uninduced NF-κB responsive promoters we observe in Fig. 7. Therefore we suggest that for future screening, a library of siRNAs for the transcription factors of interest should be compiled. Also, a library of transcription factors mutated to be constantly active is required. With these libraries at hand, individual transcription factors can be knocked down, and activated specifically at 100% efficiency. This will greaty facilitate screening and parts characterization.

Generation of down-regulated promoters

As shown, we were able to generate a set of promoters upregulated by certain factors. For several applications, promoters of a high constant strength, which become down-regulated by a signal, are required. We think it might be possible to construct such promoters by performing a RA-PCR with oligos containing weak binding sites for generally activating transcription factors (that is, binding sites which deviate from the consensus sequence) and to add some oligos containing very strong binding sites for the transcription factor of interest (say, NF-κB). If this factor is not active, the general activators will be able to bind to the DNA and activate transcription. Upon factor activation, the general activator will be replaced. If the binding site is then in a position where it does not initiate transcription (as for some of the clones (32 etc.) shown in Figure 4 and 5), the promoter will be downregulated, instead of upregulated. This hypothesis remains to be tested.

M-RA-PCR, a model-guided biochemical method for synthesis of complex promoters

Figure 14: RA-PCR can be modified to reflect probability densitiy curves in vitro

We believe that this technique, termed Model-guided Random Assembly PCR, or M-RA-PCR, is the way forward to constructing the promoters of complex regulation described in the Introduction.

Final remarks

We have developed two independent methods for the generation of truly synthetic promoter for use in mammalian cells and discussed possibilites for their combination and improvement. We are looking forward to continuing this work and generating promoters which can be used in medical or biotechnological applications, such as transcriptional targeting in virotherapy or a reporter cell line.

Methods

RA-PCR protocol

• All Oligos we used can be found in Material and Methods
• Obtain density curves about the distribution of your TF of interest from our model HEARTBEAT GUI. If this densitiy curve shows a decisive peak at distance >250Bp from the Transcriptional Start Site (TSS), consider performing M-RA-PCR . If a peak is present close to the TSS, or if data is insufficient, continue here.
  How our model was developed is described in detail on the model page.
  
• Check HEARTBEAT GUI for transcription factors coinciding with your transcription factor of interest
  

• Design two annealing sites, each 15-18 base pairs long. Annealing sites should be void of transcription factor binding sites. Calculate the reverse complement of both sequences. The following sequences MAY be used:

• Design a 5' stop oligos containing a cutsite (SpeI) and AS1_F.
• Design a 3' stop oligos containing a cutsite (HindIII) and AS1_RC.
• Design forward and reverse Oligos for each transcription factor of your interest. Forward oligos contain AS2_F, the transcription factor binding site and AS1_F. Reverse oligos contain AS2_RC, the TFBS and AS1_RC. TFBS should be designed to represent the matrix describing the factor's binding preferences (Box 1).
• Design forward and/or reverse oligos for coinceding transcription factors identified in step 2 in the same way as described in step 6.
  
• Design forward and/or reverse oligos for general activators.
• Design forward and reverse spacer oligos, which contain 10-15*dNTPs (random nucleotides) instead of a TFBS.
• Order oligos at 100µM. Pool the oligos. As a general rule, use 0,8µL oft Stop5' and Stop 3'; ~4µL of the transcription factor (forward), ~4µL of the transcription factor (reverse), 1-2µL each of the forward and reverse spacer oligo, ~1µL of coinceding transcription factors and a total of 0,5µL of general activators. For the oligo concentrations we used, please refer to Material & Methods. Define Vo as the total Volume of oligos pooled.
• Add Oligos (volume Vo) to water in total volume Vw = (10 * Vo – Vo). Add volume Vo of diluted Oligos Vw to water in volume Vf = (10 * Vo – Vo). Add volume Vo of final oligo dilution Vf to a PCR reaction. Total reaction volume SHOULD be 50 µL. We used Phusion MasterMix 2x (Finnzymes) as PCR reagent. Carry out this PCR reaction in two replicates in order to achieve greater heterogenety.
  
• Run the PCR, 7-10 cycles, with the following setup:
  
  • 1 cycle Initial dentaturing, 5 minute 95°C
    
  • 7-10 cycles assembly: 30 seconds 95°C, 45 seconds 58°C, 45 seconds 72°C
    
  • Terminal hold, 4°C, forever
    
• Remove oligonucleotides by performing a PCR purification using PCR purification kit (QIAGEN) or a gel extraction using Gel extraction kit (QIAGEN)
  
• Add PCR reagent (Phusion MasterMix 2x) again. Add 5' stop oligo and 3' Stop oligo, 25pmol (1µL of 1:4 diluted stock).
  
• Run the PCR, 25 cycles, with the following setup:
  
  • 1 cycle Initial dentaturing, 5 minute 95°C
    
  • 25 cycles amplification 30 seconds 95°C, 45 seconds 68°C, 60 seconds 72°C
    
  • Terminal hold, 4°C, forever
    
• Gel purify PCR products to exclude everything <200Bp. Use a 1% agarose gel, 50V for at least 2h to achieve a good resolution
  
• Digest with HindIII and SpeI (or whatever cutsites were included in step 4 and 5). Digest a reporter plasmid containing a core promoter and a reporter gene with the same enzymes. We used [http://partsregistry.org/Part:BBa_K203100 BBa_K203100] for this task. Make sure to perform a thorough digest; in addition, digest the plasmid with shrimp alkaline phosphatase or calf intestine phosphatase afterwards. Gel purify the plasmid backbone, PCR purify the digested PCR products.
  
• Ligate. Perform a thorough ligation to increase transformation efficiency. We used Fermentas T4 DNA Ligase for 5h, 19°C or overnight, 16°C.
  
• Transform into comptetent E. coli cells and plate out. Pick no more than 20 colonies per individual PCR reaction. If more putative promoters are desired, set up several PCR reactions
  
• Isolate plasmid DNA from the selected colonies. We used a QIAGEN Miniprep kit for this tasked.
  
• Recommended step: Test-digest miniprep DNA with the same enzymes used in step 17 to make sure you get plasmid with synthesized promoters of varying length. Length of the inserts (that is, synthetic promoters) should be between 100 and 600 basepairs. If this is not the case, vary stop oligo concentration in step 10, improve gel purification setup in step 16 or alter PCR conditions in step 12 and 15.
  
• Perfom a screening to select functioning clones. For example, transfect clones in triplicates into eukaryotic cells on a 96 well plate by using transfection agents such as EFFECTENE or Lipofectamine. Then, induce the conditions of interest in one replicate, shut them off in a second replicate, and leave control medium on the third replicate. When the pathway is fully active, read flourescence (or luminescence, if a luciferase reporter is used) by a plate reader (TECAN) or other automated methods. For the conditions we used for pathway induction, please refer to our Methods

References

[1] Alberts, B., Johnson A., Walter P., Lewis J. Molecular Biology of the Cell, 5th edition, 2008. Garland Science, p. 432-453

[2] Fussenegger, M., Weber, W. Engineering of Synthetic Mammalian Gene Networks. Chemistry and Biology 16, 287-297 (2009).

[3] Gossen, M., Bujard, L. Tight control of gene expression in mammalian cells by tetracycline-responsive promoters. PNAS 89, 5547-5551 (1992).

[4] Dorer, D. E., Nettelbeck, D. Targeting cancer by transcriptional control in cancer gene therapy and viral oncolysis. Advanced Drug Delivery Reviews 61, 554-557 (2009).

[5] Venter, M. Synthetic promoters: genetic control through cis engineering. Trends in Plant Science 12, 118-124 (2007). (and the references cited therein)

[6] Rushton P. J., Reinstädler A., Lipka V., Lippok B., Somssich I. E. Synthetic plant promoters containing defined regulatory elements provide novel insights into pathogen- and wound-induced signalling. Plant Cell 14, 749–762 (2002).

[7] Ogawa, R. Construction of strong mammalian promoters by random cis-acting element elongation. Biotechniques 42, 628-632 (2007).

[8] Edelman G. M., Meech R., Owens G. C., Jones F. S. Synthetic promoter elements obtained by nucleotide sequence variation and selection for activity. PNAS 97, 3038-43 (2000).

[9] Ellis T., Wang X., Collins J. J. Diversity-based, model-guided construction of synthetic gene networks with predicted functions. Nature Biotechnology 27, 465-471 (2009).

[10] Tornoe, J. Generation of a synthetic mammalian promoter library by modification of sequences spacing transcription factor binding sites. Gene 297, 21-32 (2002).

[11] Heintzman N. D., Ren B. The gateway to transcription: identifying, characterizing and understanding promoters in the eukaryotic genome. Cellular and Molecular Life Science 64, 386-400 (2007).

[12] Stemmer W. P., Crameri A., Ha K. D., Brennan T. M., Heyneker H. L. Single-step assembly of a gene and entire plasmid from large numbers of oligodeoxyribonucleotides. Gene 164, 49-53 (1995).

[13] Rattner, A. NF-κ B activates the HIV promoter in neurons. EMBO 12, 4261–4267 (1993).

[14] Yokoyama D. K., Ohler U., Wray G. A. Measuring spatial preferences at fine-scale resolution identifies known and novel cis-regulatory element candidates and functional motif-pair relationships. Nuc Acids Res, 1-21 (2009).

[15] del Val C., Pelz O., Glatting K-H, Barta E., Hotz-Wagenblatt A. PromoterSweep: a tool for identification of transcription factor binding site. Theor Chem Acc (in press)

[16] Ghosh S., May M. J., Kopp E. B. Rel/NF-κB and IKB proteins: an overview. Seminars in Cancer Biology, 8, 63-73 (1997).

[17] Moyano P.& Rotwein P. Mini-review: estrogen action in the uterus and insulin-like growth factor-I. Growth Hormone & IGF Research, 14, 431-435 (2009).