Design of synthetic promoters for controlled expression of therapeutic genes in retinal pigment epithelial cells

Age‐related macular degeneration (AMD) associated with dysfunction of retinal pigment epithelial (RPE) cells is the most common cause of untreatable blindness. To advance gene therapy as a viable treatment for AMD there is a need for technologies that enable controlled, RPE‐specific expression of therapeutic genes. Here we describe design, construction and testing of compact synthetic promoters with a pre‐defined transcriptional activity and RPE cell specificity. Initial comparative informatic analyses of RPE and photoreceptor (PR) cell transcriptomic data identified conserved and overrepresented transcription factor regulatory elements (TFREs, 8–19 bp) specifically associated with transcriptionally active RPE genes. Both RPE‐specific TFREs and those derived from the generically active cytomegalovirus‐immediate early (CMV‐IE) promoter were then screened in vitro to identify sequence elements able to control recombinant gene transcription in model induced pluripotent stem (iPS)‐derived and primary human RPE cells. Two libraries of heterotypic synthetic promoters varying in predicted RPE specificity and transcriptional activity were designed de novo using combinations of up to 20 discrete TFREs in series (323–602 bp) and their transcriptional activity in model RPE cells was compared to that of the endogenous BEST1 promoter (661 bp, plus an engineered derivative) and the highly active generic CMV‐IE promoter (650 bp). Synthetic promoters with a highpredicted specificity, comprised predominantly of endogenous TFREs exhibited a range of activities up to 8‐fold that of the RPE‐specific BEST1 gene promoter. Moreover, albeit at a lower predicted specificity, synthetic promoter transcriptional activity in model RPE cells was enhanced beyond that of the CMV‐IE promoter when viral elements were utilized in combination with endogenous RPE‐specific TFREs, with a reduction in promoter size of 15%. Taken together, while our data reveal an inverse relationship between synthetic promoter activity and cell‐type specificity, cell context‐specific control of recombinant gene transcriptional activity may be achievable.


| INTRODUCTION
The retinal pigment epithelium (RPE) is a multifunctional monolayer of neuroepithelium-derived cells, flanked by photoreceptor (PR) cells and the choroid complex. The significance of the RPE in the ocular system is exemplified by the major association of these pigmented cells in genetically determined retinal diseases such as age-related macular degeneration and retinitis pigmentosa. Accordingly, various in vitro human RPE models have been established as a convenient platform to study RPE functions, where the two most commonly used models are primary human fetal RPE cells and immortalized cell lines such as ARPE-19 and hRPE7. Nonetheless, studies of their morphologic and functional characteristics have produced contradictory results (Ablonczy et al., 2011) demonstrating the limitation of immortalized cells in studying native human RPE function. The more recent discovery of induced pluripotent stem (iPS) cells has also yielded iPS-derived RPE cells that closely mimic the gene expression, polarity, and physiology of native human RPE cells (Kokkinaki et al., 2011;Liao et al., 2010). However, despite a significant amount of research on global gene expression profiling of stem-cell-derived/ primary RPE cells and native tissue, there are no methods to systematically link transcriptomic data sets with genomic data to identify cis-regulatory elements that would provide inherently RPEspecific expression of recombinant genes.
The recent approval of voretigene neparvovec (Luxturna®) to treat retinal degeneration highlights that gene therapies to diseasecausing genetic mutations are possible. This achievement is partly made possible by the use of adeno-associated viral (AAV) vectors that can transduce and maintain therapeutic gene expression in nondividing cells (including the retina) with minimal immune responses (Naso et al., 2017). Ubiquitous promoters such as that derived from human cytomegalovirus (CMV) are often used to drive high transgene expression despite potentially undesirable attributes such as promoter silencing and lack of cell-type specificity. Endogenous promoters, on the other hand, often have lower activity compared to viral-derived promoters and are a relatively large size, thus limiting their use in viral vectors. For example, the~1.6 kb-long RPE65 promoter displayed only 10% of CMV activity when used to induce targeted expression of the RPE65 gene in RPE65-deficient canines, and was inactive in older animals (Le Meur et al., 2007). The latter illustrates a further possible constraint in the use of endogenous cellspecific promoters in which their expression may be downregulated for target tissues that are already in a state of disease. Similarly, Komáromy et al. (2010) also reported (endogenous) promoter length and age dependency, where the long version of the red cone opsin promoter (~2.1 kb) in younger animals led to a more stable therapeutic effect for achromatopsia. In this context, bespoke synthetic promoters are an attractive alternative as they can be customdesigned to control recombinant gene expression predictably in a specific cellular context.
A number of studies report engineering of natural promoters for improved activity in a therapeutic context. Examples include the creation of hybrid promoters to selectively kill cancerous tissues via incorporation of the prostate-specific probasin promoter into the retroviral LTR to target prostate cancer cells (Logg et al., 2002) or by coupling the endothelin enhancer element with the Cdc6 promoter to target dividing tumor endothelial cells (Szymanski et al., 2006).
More advanced attempts to create promoters with increased tissuespecificity are exemplified via de novo design of synthetic promoters that could specifically mediate gene expression in muscle cells (Li et al., 1999), colorectal cancer cells (Roberts et al., 2017) or liver cells with responsiveness to glucose (Han et al., 2011). However, these studies involved screening hundreds to thousands of synthetic promoters, which is unfeasible for primary and iPS-derived cells, such as RPE, with a limited capacity for expansion and which exhibit a particular differentiated morphological state. Further, there remains no information on how it is possible to utilize RPE model cells in vitro to characterize the function of individual genetic components to eliminate uncontrollable and functionally ill-defined parts of endogenous promoter assemblies.
In this study, we test the hypothesis that RPE genomic information can be mined to identify active transcription factor regulatory elements (TFREs) that could be utilized to design compact, space-efficient synthetic promoter assemblies that exhibit both a high degree of cell type specificity and transcriptional activity.
Through systematic bioinformatic analysis of 'omic data streams coupled with in vitro screening we identified endogenous human TFRE sequences that are potentially active in the different eye components (i.e., RPE vs. PR) as well as those that function as transcriptional repressors. We further identified highly active TFREs present in the human CMV promoter that enable active and spaceefficient synthetic promoter constructs that recruit RPE cell intrinsic transcriptional capacity. Based on these data, we designed RPEactive promoter/TFRE assemblies de novo with different objective functions, either high RPE specificity (no, or low transcriptional activity in PR cells) and/or high RPE transcriptional activity. We also compared the de novo (bottom-up) synthetic promoter design strategy to (top-down) targeted re-engineering of the RPE-specific bestrophin-1 (BEST1) promoter for improved activity in RPE cells.
While this study demonstrates effective construction of promoters for RPE cells, similar approaches could be used to design promoters for applications requiring specific and/or high expression of recombinant genes in other cell types.

| In silico analysis of TFREs
RPE and PR microarray data were obtained from the literature (Booij et al., 2009(Booij et al., , 2010  used to analyze the region -1000 to +200 relative to the TSS (or up to the start codon) to find putative TFREs. Overrepresented TFREs were identified by analyzing the promoters against Genomatixdefined human promoter background followed by selection of the TFREs with Z-score > 2.5, whereas common TFREs were identified using core similarity of 1.0 and optimized matrix similarity of +0.01 followed by selection of the TFREs with p > .2 against Genomatixdefined randomly drawn promoter samples. Identification of TFREs in hCMV-IE promoter using MatInspector was performed using core similarity of 0.75 and optimized matrix similarity.

| TFRE-reporter vector construction
pmaxGFP vector (Lonza) was utilized as a backbone. The CMV promoter and chimeric intron of pmaxGFP were deleted by digestion with BsrGI and KpnI, and replaced with a short DNA fragment containing a HindIII site. A minimal CMV core promoter from the human CMV was synthesized (Eurofins Genomics), PCR amplified (Q5 high-fidelity 2× master mix; NEB), and purified (QIAquick PCR Purification Kit; Qiagen). The PCR products were then digested with HindIII and KpnI enzymes (NEB), gel extracted (QIAquick Gel Extraction Kit; Qiagen) and inserted directly upstream of the green fluorescent protein (GFP) open reading frame (ORF) of the promoterless pmaxGFP vector. The CMV core promoter sequence used was as follows: 5′-AGGTCTAT ATAAGCAGAGCTCGTTTAGTGAACCGTCAGATCGCCTGGAGACGC CATCCACGCTGTTTTGACCTCCATAGAAGAC-3′. To create TFRE reporter plasmids, synthetic oligonucleotides containing 7× repeat copies of the TFRE sequences in Tables 1 and 2 were synthesized, PCR amplified, and inserted into BsrGI and HindIII sites upstream of the CMV core promoter. Clonally derived plasmids were purified using a QIAGEN Plasmid Plus kit (Qiagen). The sequence of all plasmid constructs was confirmed by DNA sequencing.

| Synthetic promoter vector construction
To create synthetic promoters, synthetic genes containing combinations of specific TFREs were designed in silico (Tables 3 and 4).
The positions of the TFRE blocks within the promoters were randomly arranged using R software in forward orientation of 5′ DNA strand. Synthetic genes were synthesized (Eurofins Genomics) and inserted into BsrGI and HindIII sites upstream of the CMV core promoter. A full length CMV promoter containing the same CMV core (-600 to +50 relative to the TSS) and endogenous BEST1 promoter (-585 to +76 relative to the TSS) were also synthesized and inserted separately into BsrGI and KpnI sites upstream of the GFP ORF. The sequence of all plasmid constructs was confirmed by DNA sequencing. The estimated RPE/PR specificity ratio of a promoter was calculated as follows:  Figure 4. A BEST1 promoter sequence map displaying the TFREs is shown in Figure S1A TFRE In BEST1 Endogenous sequence Consensus sequence Overrepresented and common in at least 25% 3 | RESULTS

| Bioinformatic identification of TFREs in endogenous RPE gene promoters
We previously devised a methodical, comprehensive approach for bioprocess-directed design of synthetic promoters based on genomic sequence information (Johari et al., 2019). The work flow enables identification of TFREs associated with endogenous gene promoters with specific characteristics, thus allowing de novo design of synthetic promoters with relevant functional features. In this study, we hypothesized that it was feasible to construct RPE-specific promoters using assemblies of the corresponding RPE-specific TFREs (summarized in Figure 1). To profile endogenous gene expression in RPE and PR cells, we utilized transcriptomic datasets from Booij et al. (2009Booij et al. ( , 2010. A subgroup of 35 highly expressed RPE-specific genes was created by selecting genes with a microarray expression level > 6,000 units in RPE and at least four-fold higher expression levels in RPE than in PR cells (as well as choroids). As expected, the BEST1 (VMD2) gene was highly and preferentially expressed in RPE cells, confirming the potential use of its promoter to drive RPEspecific expression in vitro (Esumi et al., 2004) and in vivo (Guziewicz et al., 2013;Kachi et al., 2006). A corresponding PR-specific group was created by selecting 35 genes with expression level < 6,000 units in RPE and RPE/PR expression fold change of < 0.5, whereas a nonspecific group was created by selecting 35 genes with expression level > 6,000 units and RPE/PR expression fold change of between 0.97 and 1.03. Table S1 lists the selected genes in each group.  Figure S1A) were also included. We note that in our analysis MITF was identified only in BEST1 promoter and the primary site overlapped HELT (Esumi et al., 2004(Esumi et al., , 2007. Table 1 lists the final set of 29 TFREs incorporated into the functional screen and their endogenous and consensus sequences (derived from the endogenous promoters and Genomatix software, respectively). The TF matrix/family, frequency and Z-score/p-value of the selected TFREs are detailed in Table S3.
To further analyze the "transcriptional landscape" of RPE cells and identify active TFs contained within, we surveyed putative TFREs in hCMV-IE1. Using Genomatix MatInspector tool, 70 discrete TF families were identified in the CMV promoter. A subset of 10 TFs T A B L E 2 Transcription factor regulatory elements (TFREs) identified by bioinformatic survey of cytomegalovirus (CMV) promoter and their overrepresentation in the retinal pigment epithelium (RPE)-specific, photoreceptor (PR)-specific, and nonspecific groups. Known transcriptional repressors (YY1 and Gfi1) are excluded from analysis. Measurement of the TFRE relative ability to activate transcription of recombinant photoreceptor (GFP) genes in RPE cells is shown in Figure 4b. A CMV promoter sequence map displaying the TFREs is shown in Figure S1B TFRE No. of copies Viral sequence In BEST1 In RPEspecific In PRspecific In nonspecific Abbreviations: AhR/ARNT, aryl hydrocarbon receptor and nuclear translocator heterodimer; AP-1, activator protein 1; C/EBPε, CCAAT/enhancer binding protein epsilon; CREB, cAMP-responsive element binding protein; NF1, nuclear factor 1; NF-κB, nuclear factor-κB; RAR, retinoic acid receptor; Sp1, stimulating protein 1; SRF, serum response factor; TLX1, T-cell leukemia homeobox 1.
that are known to be positive regulators of CMV activity in different cell types (e.g., Brown et al., 2015;Ghazal et al., 1992;Lashmit et al., 2009) were selected for screening ( Figure S1B). To further minimize this pool (design space) as well as false positives, we selected TFRE sequences with the highest Genomatix matrix similarity from each TF family as summarized in Table 2. The TF matrix/family, frequency and matrix similarity of the selected TFREs are detailed in Table S4. We note that viral-derived NF-κB sequence is identical to the consensus sequence (Table 1).

| Determination of TFRE activity in RPE cells
Previous studies showed that iPS-derived RPE cells exhibit membrane potential, ion transport, polarized vascular endothelial growth factor secretion, and gene expression profile that closely resemble to those of native human RPE (Kokkinaki et al., 2011;Liao et al., 2010).   Table S1) were surveyed for the presence of discrete TFREs using Genomatix Gene Regulation software using (a) overrepresented TFRE method by analyzing the promoters against a pre-defined human promoter background, and (b) common TFRE method by selecting TFREs that present in at least 25% (9/35) of the genes. The region −1000 to +200 relative to putative transcriptional start site (TSS) was analyzed against a human promoter background to find overrepresented TFREs in each group, or for common TFREs that present in at least 25% (9/35) of the promoters. To identify potentially active TFREs in the high activity group, TFREs that also occurred in the PR and/or nonspecific groups were excluded and (c) the remaining TFREs from both methods were narrowed down to 25 as described in text. DNA sequences of the 29 selected TFREs (including 4 from the literature) are listed in Table 1 active, attaining up to 36% of CMV activity. Consensus NF-κB (p50), MafF, and PAX6 also displayed substantial activities (2.4%-8.5% CMV, p < 0.001) although we suspect the former was contributed by a weak binding of NF-κB (V$NFKAPPAB.02 matrix similarity 0.868).
Contrarily, an alteration of six nucleotides flanking the E-box

| First generation RPE promoters enable identification of repressor elements in RPE cells
Given the above finding that HELT/MITF could bind to suboptimal E-box sequences (leading to lower activities), we utilized the four positive endogenous TFREs identified in the screening exercise to evaluate their compatibility in a heterotypic construct (promoter 1/01). Importantly, given that a TF can act as a transcriptional activator or repressor (or both), we designed a set of promoters (promoter 1/02-1/10) containing specific combinations of the endogenous TFREs selected in the bioinformatic analysis to assess elements that primarily function as repressors in RPE cells. For each promoter, 6 copies of HELT/MITF were included to provide a promoter basal expression, and the other 27 specific TFREs (4 copies each) were randomly distributed using R software with the following design rules: (i) each specific TFRE occurred twice in different promoters, (ii) no two specific TFREs re-occurred in another promoter, and (iii) the relative order of constituent TFREs was random, separated by minimal spacers (Brown et al., 2017;Johari et al., 2019).
Additionally, to test whether the selected endogenous elements from the bioinformatics analysis could generally act as positive effectors in heterotypic constructs, we constructed a promoter (1/11) containing two repeat copies of each element that present in upstream of BEST1 promoter (-900 to -1 relative to the TSS). The synthetic promoter constructs were chemically synthesized and inserted upstream of the minimal CMV core promoter in GFP reporter plasmids.
Further bioinformatic examination on the endogenous NF-κB sequence (V$NFKAPPAB.02, matrix similarity 0.869) indicated that this particular sequence overlapped with transcriptional repressor RBP-Jκ binding sequence (V$RBPJK.02 matrix similarity 0.958; Figure S1A). This constitutive binding by RBP-Jκ (Lee et al., 2000) expounds the repression observed in promoters 1/03 and 1/04 ( Figure 5) as well as the inactivity of endogenous NF-κB sequence F I G U R E 4 Screening discrete transcription factor regulatory element (TFRE) activity in retinal pigment epithelial (RPE) cell models. Seven copies of each TFRE (as described in Tables 1 and 2) comprising (a) endogenous sequences and (b) consensus/viral sequences were cloned in series upstream of a minimal cytomegalovirus (CMV) core promoter in reporter vectors encoding green fluorescent protein (GFP). 5 × 10 4 iPS-derived RPE cells were plated and 0.4 μg of plasmid was transfected into the cells by lipofection at Day 3 post-plating. 0.72 μg of plasmid was transfected into 3 × 10 5 primary RPE cells by nucleofection followed by plating. GFP fluorescence level was measured in the fully differentiated iPS-derived and primary RPE cells at Day 28 and Day 14 post-plating, respectively. Data are expressed as a percentage with respect to the GFP expression of a vector containing CMV-IE promoter. Data shown are the mean value ± standard deviation of three independent experiments each performed in triplicate compared to the consensus sequence ( Figure 4). In contrast, PAX6 has been reported to act as a transactivator in RPE cells (Raviv et al., 2014) and its consensus sequence displayed significant activity ( Figure 4b). Thus, PAX6 may be excluded as a repressor in which the repression observed in promoters 1/04 and 1/09 could be attributed to RBP-Jκ and NF-κB (p50) elements respectively. Further, as promoter 1/02 exhibited lower GFP expression in primary cells compared to promoters 1/01 and 1/05 and did not contain any of above repressor elements, we deduce that the AREB6 secondary binding sequence employed repressed transcription as reported by Ikeda and Kawakami (1995) with E-box motif in HELT/MITF sequence acted as the main binding site of multi-domain AREB6 protein (see Figure S1A). Promoter 1/11, which contained two copies of selected elements from the BEST1 promoter demonstrated~3-4-fold higher expression than the BEST1 promoter itself, confirming the utility of the RPE cell synthetic promoter design approach.

| Second generation RPE promoters exhibit an inverse correlation between activity and specificity
Based on the observations from the first library, we created a second library of functional synthetic promoters (Table 4) by omitting probable repressor elements with two objectives: (i) high RPE cell-specificity, attainable by limiting the constructs to endogenous elements (promoters 2/01-2/07), and (ii) high RPE transcriptional activity, attainable by including active viral-derived elements (promoters 2/08-2/13). The former is highly desirable for targeted AAVmediated transduction in vivo, whereas the latter can be efficiently applied for recombinant protein expression in vitro to investigate the effects of protein overexpression in retinal diseases (e.g., HtrA1 enrichment; Melo et al., 2018) as well as to reprogram RPE cells to an altered lineage (e.g., neuronal cells; Hu et al., 2014). Additionally, we created an engineered BEST1 promoter by mutating a total of 25 nucleotides to remove repressors and/or introduce active TFREs ( Figure S3).

Measurement of GFP expression after transfection into RPE
cells is shown in Figure 6. These data show that promoter activities vary by~30-fold, where the most active promoters exceeded the transcriptional activity of CMV in primary cells. The engineered BEST1 promoter exhibited a 30%-32% improvement in expression compared to its native counterpart, although this was largely insignificant (p = 0.10-0.14). Similarly, promoter 2/03, devoid of repressor elements, displayed a 2.1-fold increase in expression compared to promoter 1/11 in primary cells, although biasing the elements towards active elements (promoter 2/04) appeared to have a negative effectlikely due to suboptimal TFRE stoichiometry (Martinelli & De Simone, 2005). Anticipated to be RPE-specific, the in vitro expression levels obtained from promoters 2/01-2/05 were up to~8fold higher than that of BEST promoter but significantly lower compared to CMV (≤17%). While NF-κB in promoters 2/06 and 2/07 significantly improved expression with the latter attaining 62% CMV activity in primary cells, we conjecture the use of high copy number of NF-κB would increase promoter activity in PR (and other) cells, where the cognate TF is also present (Table S2). Expectedly, the data in Figure 6 also shows that strong RPE promoters can be constructed by incorporating viral-derived CREB, Sp1 and AP-1, and biasing the TFRE copy number towards highly active elements resulted in F I G U R E 5 Measurement of first-generation synthetic promoter activity in retinal pigment epithelium (RPE) cells. (a) 11 synthetic promoters comprising Library 1 (transcription factor regulatory element [TFRE] compositions described in Table 3) were transfected into induced pluripotent stem (iPS)-derived and primary RPE cells. Intracellular green fluorescent protein (GFP) level was analyzed in differentiated RPE cells. Data are expressed as a percentage with respect to the expression level exhibited by the control cytomegalovirus (CMV) promoter. GFP expression driven by the BEST1 promoter was also tested as RPE-specific promoter control. Data shown are the mean ± standard deviation of three independent experiments each performed in triplicate. (b) The GFP levels exhibited by the synthetic promoters in (a) were plotted against the promoters' specific TFRE components to identify potential repressor elements (marked by an asterisk [*]) by setting the minimum expression threshold to 2% CMV and further analyzed as described in the text substantial increase in promoter strength up to 109% of CMV in primary cells. However, reporter expression in iPS-derived cells was relatively lower compared to that observed for the primary cells (achieving only~50% of CMV activity), indicative of differences in transcriptional landscape. We deduce that iPS-derived cells, reprogramed from somatic (skin or blood) cells, contained untested TFs that conferred relatively higher CMV expression.
To further illustrate the promoters' (predictive) capability in conferring specific and exclusive cellular tropism for RPE gene therapy, we calculated an "estimated RPE/PR specificity ratio" for each promoter as a function of (i) the transcriptional activity of a specific TFRE (Figure 4), (ii) the copy number of a specific TFRE within the promoter (Table 4), and (iii) the cognate TF mRNA expression fold-change in RPE over PR (Table S2; Equation 1). As shown in Figure 6, this analysis indicates that promoters designed with endogenous elements would drive targeted transgene expression to the RPE cells in vivo, whereas promoters with viral-derived elements would have significantly reduced specificity. While not directly relevant, our data from a separate study showed that these promoters exhibited no or very low activity (≤20% CMV) in a human kidney cell line ( Figure S4), illustrating that cell-specific control of recombinant gene transcriptional activity is feasible. Furthermore, one major limitation of AAVs as vectors is that AAV packaging capacity is fundamentally restricted to 5 kb where packaged vector genomes derived from plasmid-encoded vectors exceeding 5 kb are truncated on the 5′ end and heterogeneous in length, as well as result in a considerable reduction in viral production yields (Wu et al., 2010). The synthetic promoters in this study were relatively small in size compared the control BEST1 and CMV promoters (  (Vaquerizas et al., 2009), only a small range of core TFs are responsible in programming the gene expression that define individual cell identity (D'Alessio et al., 2015). This is evident, for example, where our study indicated that SRF (serum response factor) element had no transcriptional activity in RPE cells (Figure 4b) yet the identical sequence formed the primary building block of synthetic promoters that conferred muscle-specific expression (Li et al., 1999).
We further identified suboptimal TF binding sequences and endogenous TFREs (25%) that acted as transcriptional repressors.
These were not entirely unexpected as the promoters in RPE have evolved to function in a complex spatiotemporal gene regulation of the retina including pigment biogenesis, ion transport, and growth factor secretion (Booij et al., 2010). However, it is highly unlikely that they will be optimal for use in a more specific context such as AAV-  Table 4) were transfected into induced pluripotent stem (iPS)-derived and primary RPE cells. An engineered BEST1 promoter (E.BEST1) was constructed by mutating a total of 25 nucleotides to remove and/or introduce specific TFREs ( Figure S3). Intracellular green fluorescent protein (GFP) level was analyzed in differentiated RPE cells. Data are expressed as a percentage with respect to the expression level exhibited by the control cytomegalovirus (CMV) promoter. Predictive RPE/photoreceptor (PR) specificity ratio of each promoter was calculated using Equation 1 based on the transcriptional activity of a specific TFRE, the copy number of a specific TFRE in the promoter, and the cognate transcription factor (TF) mRNA expression fold-change in RPE/PR. Data are expressed as a percentage with respect to the specificity ratio of the control bestrophin-1 (BEST1) promoter elements, as well as optimization of active TF binding sequences, would permit engineering of endogenous promoters for enhanced performance. With regard to the latter, tens of TFRE motif sequence variants can be characterized simultaneously through in vitro use of parallel high-throughput screening techniques, allowing determination of their optimal binding affinity. With regard to the former, the functional impact of repressor elements can be identified and accurately quantitated using the TFRE-specific decoy technology previously developed in this laboratory (Brown et al., 2013(Brown et al., , 2015. Even though we have not tested our promoters against PR (due to lack of reliable model cells, see below), and therefore cannot definitively claim that they will exhibit restricted gene expression in RPE cells, the methodology presented allows creation of promoters using binding sequences that are exclusive to RPE-specific promoters and correspond to relatively high expression of their cognate TFs in RPE cells, thus enabling confident prediction of their functionality.
On the other hand, TATA box (present in the minimal core promoter) is a known modular component in that the strength of the TATA-RNA polymerase complex and the ensuing transcription that it mediates has very little noise to promoter activity-it simply acts as a linear amplifier without influencing specificity of gene expression controlled by upstream cis-regulatory elements (Mogno et al., 2010).
Indeed, muscle (Li et al., 1999), deregulated β-catenin (Lipinski et al., 2004), liver (Han et al., 2011), and colorectal cancer cellspecific (Roberts et al., 2017) synthetic promoters all contained a TATA box. Furthermore, our bioinformatic analysis indicated that Sp1 is highly prevalent in all three groups (RPE, PR, and nonspecific) analyzed-in agreement with the notion that Sp1 is essential for maintaining basal transcription of genes and protection of CpG islands from de novo methylation (Samson & Wong, 2002). Accordingly, we conjecture that Sp1 does not influence the specificity of a promoter that is designed to mediate cell type specific expression.
We further acknowledge that the RPE cell models we employed may contain small subsets cells in a variably differentiated state that have a transcriptional landscape deviating from that of the fully differ- Day 45-60 (Mellough et al., 2012). On the other hand, 661W PR cell line, derived from a mouse retinal tumor, expresses several markers of cone PR cells but not of rod cells (Tan et al., 2004) and was also reported to exhibit the properties of retinal ganglion cells (Sayyad et al., 2017). Nevertheless, where reliable model cells are not available, we demonstrated that it may possible to design cell-specific promoters in silico for in vivo applications. As the quality and volume of 'omics data continues to increase, and, given the progressive development of TFRE database and informatic tools (e.g., Wu et al., 2019), we envisage that synthetic promoters will facilitate advancement of the current revolution in AAV-mediated gene therapy.

ACKNOWLEDGMENTS
This study was supported by REGENXBIO, USA. The authors thank Stephen Jaffé and William Morgan-Evans (University of Sheffield) for assistance in cell culture and fluorescence microscopy, respectively, and Laura Moran (REGENXBIO) for project management.