Promoter studies using CRISPR/dCas9 and LC-MS/MS based proteomics
Hidde R. Zuidhof, BSc
Research project II
March – July 2016
Biomolecular Sciences program, Molecular Biology and Biotechnology, University of Groningen
Experimental Hematology group, University Medical Centre Groningen
Coordinated by dr. M.H.K. Linskens and prof. dr. J.J. Schuringa Supervision by dr. A.T.J. Wierenga
Contents Page number
Abstract 3
Introduction 4
Materials and methods 6
Results 8
Construction of K562 cell lines expressing dCas9 and gRNA 8
The chromatin around the targeted PDK1 and SLC2A1 promoters can be enriched for in pull outs 9
Proteomics analysis of dCas9 guided pull outs yield unique peptides 11
Discussion 14
Acknowledgements 16
References 17
Supplementary figures 19
Supplementary tables 21
Abstract
Many different types of gene regulation studies exist. However, most of these either look at DNA binding of a protein, or study the downstream effects of e.g. overexpression of a certain factor. So far it has been very challenging to uncover the complete set of proteins that bind to an endogenous promoter and govern its activation. This study aimed to do just that: develop a system based on CRISPR/dCas9 in order to specifically pull down promoters. Potentially, subsequent mass spectrometry analysis could identify associated proteins. A restriction deficient Cas9 (dCas9) fused to an avi-tag was transduced into human chronic myelogenous leukemia (K562) cells along with a BirA biotin ligase.
Furthermore, two sets of lentiviral vectors were created, expressing ten different gRNAs each. Both sets of gRNAs spanned a region of 200bp from either the PDK1 or SLC2A1 promoter, two genes critically involved in glucose metabolism control. After crosslinking using a Chromatin Immunoprecipitation (ChIP) protocol, avi-tagged dCas9 protein was precipitated using streptavidin beads for subsequent LC-MS/MS analysis to identify proteins that were co-precipitated. Using this protocol, we show that targeted endogenous promoters can efficiently be precipitated providing a
means to identify sets of unique proteins that might be specific regulators of the targeted genes.
Introduction
Mammalian cells have elaborate mechanisms in place to regulate their protein expression under different conditions or at different time points. This regulation can take place at several stages in the protein synthesis process and thereafter: transcription regulation (e.g. by activators, repressors or chromatin remodeling), mRNA processing, translation regulation and protein activity control (including post-translational modifications and degradation control) are all well-known mechanisms to control protein activity. 1-4 Here the focus will lie on the beginning of the cascade: transcription regulation.
Classical transcription regulation through the binding of transcription factors can either enhance or repress transcription at a certain site. Several techniques, including chromatin immunoprecipitation (ChIP) followed by sequencing (ChIP-seq), are available for studying chromatin binding behavior of proteins to specific endogenous loci in the genome. These studies can provide valuable information on how certain genes are regulated. This approach however has certain shortcomings: a candidate protein must a priori be selected, ChIP-grade antibodies must be available for this protein and it does not allow for a non-biased assessment of the repertoire of proteins present at that specific endogenous locus in a certain cellular state. Furthermore, the finding that a protein regulates transcription of a certain gene does not mean that this is happening under all circumstances. Regulation can be dependent on environmental factors or presence of co-repressors/activators. The interaction with other proteins might provide an explanation as to why proteins with the same core DNA binding sequence can regulate a distinct set of genes. The hypoxia inducible factor (HIF) protein family serves as an example; HIF1α and HIF2α (sharing an identical DNA binding sequence) regulate a distinct set of genes in addition to common targets regulated by both HIF1α and HIF2α.5 This study was aimed at dissecting the mechanisms that regulate gene expression on a per-locus basis. It is likely that in some cases the effect a transcription factor has on gene expression is dependent on other factors binding either directly to it or to the local chromatin. In essence, the aim was to identify regulatory proteins from a specific genetic locus by combining the knowledge we already have about interaction studies with loci targeting.
Combining interaction studies with site specificity
Methods for protein-protein interaction studies are as abundant as they are diverse.6 For practically every interaction experiment conceivable, multiple different approaches can be taken. Techniques vary in their efficiency, thoroughness, sensitivity, selectivity and amount of input material needed, to name a few.6 Also for studies of promoter regulation several systems exist. If certain regulators are already known and antibodies against these proteins are available, pull-out experiments could be performed.7 The benefit of such a system is that the native locus is tampered with as little as possible.
Normal cellular regulation can take place up until the point of harvesting, and the foreign antibody is only added later. A downside of this approach is that the antibody is raised against a certain protein, not a genetic locus. This makes it difficult to pinpoint a detected protein to a locus. In other words:
although the DNA binding sites of a protein can be mapped with ChIP-seq, pull out experiments will never yield site specific information on interaction partners. Targeting specific genetic loci has been described by multiple groups, and different approaches such as CRISPR/Cas, TALE-technology and photo crosslinking are taken.8-13 Several of these studies use proximity based tagging of a protein of interest by e.g. an expressed biotin ligase.11,12 One of the advantages of this approach is the high affinity of biotin for streptavidin. This allows for relatively easy purification of tagged proteins by pull downs using immobilized streptavidin.14 In practice this means expressing fusion proteins of both the proteins suspected to interact; one tagged with the biotin accepting tag (e.g. an avitag), and one with the biotin ligase. The issue with techniques like this however is that a priori certain targets need to be known, unless one uses a mutated version of biotin ligase to be able to detect any protein in the vicinity.15 Inherent disadvantage of the latter approach being of course that tagging becomes based on proximity instead of real binding, likely increasing the number of false positives. Attempts have been made recently to visualize specific loci (promoters, genes) in human cells using fluorescently
tagged transcription activator like effectors (TALEs), or CIRSPR/dCas technology. On repetitive stretches in telomeres imaging works quite well for both techniques. Even when targeting non repetitive regions an acceptable signal to noise ratio was reached with CRISPR/dCas.9,13 One of the challenges here is that for a large part of the cell cycle, only two copies of each allele exist. For imaging therefore, multiple stretches of DNA in the same exon were targeted.9 With the upcoming of these technologies it has become possible to monitor a specific genetic locus in time. TALE-based imaging technology uses a fluorophore covalently linked to a DNA-binding domain that can be varied, allowing for different loci to be targeted.13 A downside of this technology however, is that for each DNA sequence to be studied the entire fluorophore conjugation needs to be cloned and subsequently expressed in cells. Especially because these vectors tend to be large, this can become difficult. The CRISPR/dCas system is relatively flexible compared to TALE technology. As explained below a single effector protein can be expressed in cells, and subsequently only expression of a small guide RNA is needed to provide sequence specificity. As the fluorophore and DNA targeting domain are on different vectors, vector sizes are decreased and a higher degree of flexibility is achieved. Because of these apparent advantages, an approach based on the CRISPR/dCas9 system was used in this study.
Using CRISPR/Cas to target specific loci
CRISPR/Cas9 (Clustered Regularly Interspaced Short Palindromic Repeat / CRISPR associated protein 9) systems to edit genomes have been applied in a variety of organisms across all kingdoms.16 CRISPR/Cas is a defense system first discovered in Staphylococcus pyogenes (but also present in other staphylococci) based on the Cas9 endonuclease (figure 1A).16 Upon phage infection, DNA from the invader is incorporated adjacent to CRISPR repeat sequences that are present throughout the genome.
Transcripts of this sequence (deemed CRISPR-RNA, or crRNA) can subsequently pair with a tracrRNA that is transcribed from a different locus, promoting interaction with the Cas9 nuclease. For research applications a conserved Cas9 binding stretch of RNA (tracrRNA) is coupled to a CRISPR-RNA: a Figure 1. Adaptation of CRISPR/Cas to perform genome editing. A) Clustered regularly interspaced short palindromic repeats (CRISPRs) can be found throughout the genome of S. pyogenes. Foreign DNA is often incorporated into such a cluster. Transcripts of these CRISPRs hybridize with a tracrRNA, through which they can interact with the Cas9 endonuclease. Subsequent hybridization of the crRNA with target DNA precedes cleavage of the DNA. B) Through cloning a fusion of tracrRNA and a designed crRNA can already be made, named guide RNA (or gRNA). Further mechanism of action remains the same, allowing controlled targeting of a DNA site for cleavage. In principle any DNA sequence can be targeted provided it ends with 5’-NGG (PAM-sequence), needed by Cas9 to make a double strand break.
sequence directed against the target genome sequence. This fusion RNA is called a guide RNA (gRNA) that can subsequently bind to its target. Upon binding to the DNA the Cas9 endonuclease can induce a double strand break if the target sequence is next to a proto-spacer adjacent motif (PAM) (figure 1B). The canonical PAM-sequence is 5’-NGG, although activity on other PAM-sequences has also been described.17 The editing of genomes by inducing double strand breaks is however not the only application of the CRISPR/Cas system. The endonuclease activity can be removed by two point mutations, yielding nuclease-deficient Cas9 (dCas9). Fusion proteins of dCas9 and an effector domain can target specific genomic loci and thereby up- or down regulate eukaryotic gene expression of a desired gene.18
In this study a new technique involving dCas9 was tested. The goal was to use the sequence specificity of the gRNAs to investigate certain promoter regions. The K562 chronic myelogenous leukemia cell line was chosen to test the new system, because extensive experience with this cell line was present in the lab. K562 is a ‘classical’ CML cell line, expressing the BCR-ABL fusion protein from the t(9;22) translocation, also known as the Philadelphia chromosome.19 The SLC2A1 and PDK1 genes were chosen as primary targets. Solute carrier family 2 member 1 (SLC2A1 or GLUT1) encodes a glucose uptake transporter that is upregulated in CML and other types of cancer. 20,21 Pyruvate dehydrogenase kinase 1 (PDK1) is an enzyme indirectly controlling pyruvate entry into the TCA cycle by phosphorylating pyruvate dehydrogenase and thereby rendering it inactive, preventing TCA entry.22 This protein was also found to be upregulated in CML, and has been suggested to be a potential therapeutic target.23 To study the components regulating these promoters, a promoter pull-out was set up. An avi-tag was cloned at the C-terminus of dCas9 which can be biotinylated and subsequently purified by means of streptavidin beads. Proteins were crosslinked followed by DNA sheering prior to affinity purification, and it was assumed that factors binding close to dCas9 on the chromatin would subsequently be co- purified allowing for their identification by mass spectrometry. Since the dCas9 binds to the promoter of the gene of interest, factors binding near it may be expected to have some function in the regulation of the gene. Multiple (10) gRNAs were designed per promoter. These gRNAs could all target the same promoter region, but at slightly different positions, thereby reducing the risk of false-positives through any off-target binding of the individual gRNA. The ultimate goal of this study was to develop a method for examining gene regulation under specific conditions using a dCas9-based approach.
Materials and methods
Cell culture and counting
Cells were grown in either RPMI 1640 medium with HEPES and L-glutamine (Lonza, Verviers, Belgium) with 10% Fetal Cow Serum (FCS) and 1:100 penicillin/streptomycin (Gibco, 10,000 U/ml, ThermoFischer scientific, Waltham, Massachusetts, USA) for K562 cell line or derivatives thereof and DMEM with L-glutamine and glucose (Lonza) with 10% FCS and 1:100 penicillin/streptomycin for HEK- 293T-cells or derivatives thereof. Incubation was at 37ºC, 5% CO2. For the adherent HEK-293T cell line trypsin treatment was needed prior to handling: cells were washed twice in PBS with aspiration of supernatant. In T75 flask, 0.5ml trypsin + EDTA was added and flask was incubated at room temperature till cells detached. Subsequently cells were resuspended in medium and spun down to remove trypsin. Cell counting was performed by harvesting 200 µl of homogenized cell culture in 10 mL Isoton II. After addition of 2 drops ZAP-Oglobin II Lytic Reagent and brief shaking, cells were counted on Coulter Counter (Beckman Coulter, Brea, California, USA).
CRISPR sgRNA design
Sequences around exon 1 from SLC2A1 and PDK1 were retrieved from NCBI (www.ncbi.nih.gov/gene/6513 and www.ncbi.nih.gov/gene/5163 respectively) using GRCh38.p7 assembly. Potential sgRNA designs were evaluated using an algorithm from Doench et al. (2014).1 sgRNA’s were ordered as separate forward and reverse primers with linkers (Fw: CACCG-3’, Rv: AAAC- 3’) to clone into alkaline phosphatase treated BbsI cut pLsBlue vector, see supplementary table 1 for
primer sequences. Primers were annealed, phosphorylated and ligated as described by Ran et al.
(2013).2
Virus production and transduction of K526 cells
HEK-293T cells (20% confluent in T75) were transfected with pCMV-VSV-G (0.7 µg), psPAX2 (3 µg) and pRRL-S-dCas9:Avi-iGFP:BirA or pRRL-S-MCS:Avi-iGFP:BirA (3 ug) with FuGENE HD (Promega, Madison, Wisconsin, USA). Viral supernatant was collected and filtered through a 0.45 micron filter (Merck Millipore). K562 cells were transduced with the vector by adding 100 µL viral supernatant to 1mL medium containing 250.000 cells and 8 µl/mL Polybrene (Sigma-Aldrich, St. Louis, Missouri, USA). After 24 hours, cells were washed three times in 10 ml PBS + 5% FCS to wash away virus. The same protocol was followed for subsequent transfection of dCas9 expressing cell line with sgRNA constructs mentioned above and transfection of GFP cell line with scrambled control.
Flow cytometry analysis and cell sorting
MACSQuant Analyzer 10 (Miltenyibiotec, Bergisch Gladbach, Germany) was used for fluorescent marker analysis. For virus titer testing (in 96-well plate) samples were fixed in 1% para-formaldehyde for 10 minutes at room temperature prior to analysis. For cell sorting, cells were resuspended in 500 µl medium (maximally 10 x 106 cells/ml) and sorted on MoFlo XDP (Beckman Coulter) for GPF expression, Blueberry expression, or both. Collection was always in 500 µl pure FCS. Data for both types of experiment was analysed using WinList 3D, version 8.0 (Verity Software House, Topsham, Maine, USA).
qPCR
Quantitative real-time polymerase chain reaction (qPCR) was performed to determine mRNA levels of SCL2A1 and PDK1. RNA was isolated from 1 x 105 cells using RNeasy kit, final elution in 20 µl nuclease free water (Qiagen Venlo, The Netherlands). iScript reverse transcriptase was used to obtain cDNA (iScript reverse Transcription kit, Biorad, Veenendaal, The Netherlands) according to the manufacturer’s protocol. cDNA was diluted 15 times in nuclease free water and 5 µl was subsequently used in combination with 10 µl SSo Advanced Universal SYBR Green Supermix (Biorad), 0.16 µl primer mix (forward and reverse, at 25mM each) and 4.84 µl nuclease free water. Primer sequences for PCR can be found in supplementary table 1. Solution was mixed, briefly spun down and analysed using CFX connect Thermocycler (Biorad). RLP27 was used as housekeeping gene.
Western blotting
Samples were harvested and boiled for 10 minutes in 1x SDS-sample buffer (2% SDS, 10% glycerol, 2%
β-mercaptoethanol, 60 mM Tris-HCL pH 6.8 and bromophenol blue). Proteins were separated by 7.5%
SDS-PAGE (4% stacking) and transferred to Immobilon FL membranes (Merck Millipore, Darmstadt, Germany). Transfer was semi-dry using an electroblotter (Biorad). Membranes were blocked in 1x Odyssey Blocking Buffer (PBS) (LI-COR Biosciences, Lincoln, Nebraska, USA) for 30 minutes at room temperature. Primary and secondary antibody staining was either for 2 hours at room temperature or overnight at 4ºC (under agitation) in 1/10 diluted blocking buffer (in PBS). IRDye streptavidin 680 was used for biotin visualization at 1:5000. Actin staining was 1:2000 with mouse anti-ß-actin (Sigma- Aldrich), followed by 1:2000 IRDye 800 CW goat anti mouse (LI-COR Biosciences).
‘ChAP’-analysis
Chromatin affinity-precipitation was performed on 4 x 106 cells. Cells were crosslinked with 1%
formaldehyde for 10 minutes at room temperature on the rotator. Crosslinking was quenched by addition of 125 mM glycine (final) and rotation for 5 minutes at room temperature. Cells were spun down (8 min, 900 x g, 4ºC) and washed in ice cold PBS. Next cells were resuspended in 200 µl IP-buffer (SDS-lysis buffer and Trition dilution buffer mixed 2:1) and sonicated for 15 cycles of 30s on/off on Bioruptor pico (Diagenode, Liege, Belgium) followed by 2x 30s intensity 14 tip sonication on Soniprep 150 (MSE, London, UK) after adjusting volume to 1mL with IP-buffer.
Samples were rotated overnight at 4ºC after addition of 50 µl IP-buffer equilibrated Dynabeads M280- Streptavidin beads (ThermoFischer scientific). Next day beads were harvested (magnet) and washed with ice cold Mixed Micelle Buffer (3x 1 ml), buffer 500 (2x 1ml), LiCl-buffer (2x 1 ml) and TE (1x 1ml).
Complexes were eluted from beads at 65ºC for 90 minutes in 250 µl 1% SDS, 0.1M NaHCO. After removal of beads samples were left at 65ºC overnight. Samples were treated with 1:100 RNAseA for 2 hours at 37ºC and 1:100 Proteinase K (Roche, Basel, Switzerland) for 2 hours at 55ºC. DNA was purified using a PCR clean up kit (Qiagen). For buffer compositions see supplementary table 4.
Mass spectrometry
Protein gels were run at 100V till gel front ran off, or for 5 minutes and stained with mass-spec grade Coomassie for 30 minutes at room temperature. Gels were destained in demi water and each lane was cut in 11 pieces for LC-MS/MS analysis on LTQ Orbitrap (ThermoFischer scientific), on FT-ICR/ORBITRAP MS scan mode (gels that were ran for 5 minutes were processed as a single slice). Excised gel pieces were washed 3 times (1. 300µl 70% 100 mM NH4HCO3 in water, 30% acetonitrile. 2. Same substances at 50%:50%. 3. 100 % acetonitrile) by shaking at room temperature for 30 minutes. Acetonitrile wash was for 5 minutes. Gel pieces were dried at 37°C for 30 minutes. Samples were reduced and alkylated by incubating in 20 µl 10 mM DTT in 100 mM NH4HCO3 for 30 minutes at 55°C. 10 µl 55 mM iodoacetamide in 100 mM NH4HCO3 was added prior to incubation at room temperature in the dark for 30 minutes.300 µl acetonitrile was added and samples were shaken for 30 minutes. Gel pieces were again dried at 55°C for 15 minutes. 20 µl 10ng/µl trypsin (sequencing grade modified trypsin, Promega) was added prior to overnight incubation at 37°C. Liquid was collected next day and mixed with extraction liquid after shaking 20 minutes in 5% formic acid. Proteins were identified using software with the following settings: search engine: initially PEAKS, v8.0 (Bioinformatics Solutions Inc., Waterloo, Canada) 10.0 ppm parent mass error tolerance, 0.03 Da fragment mass error tolerance, monoisotopic precursor mass search type. All based on trypsin digestion with 3 max missed cleavages and 1 non-specific cleavage. PMT’s: oxidation and carbamidomethylation. Maximum number of PMT’s per peptide: 3. For later searches against transcription factor lists, Morpheus_Thermo was used (revision 255).3 Same settings as mentioned above were used, except FDR was 5% and phosphorylation was also allowed as PMT.
Results
Construction of K562 cell lines expressing dCas9 and gRNA
In order to study dCas9 potential to precipitate a specific locus first cell lines stably expressing the desired constructs were made. dCas9 was cloned into a pRRL expression vector, under control of a spleen focus forming virus (SFFV) promoter. In addition, a C-terminal avitag was added to the protein.
An internal ribosome entry site (IRES) sequence was placed adjacent to the tagged protein, followed by an EGFP-BirA fusion (figure 2A). EGFP fluorescence facilitates cell sorting while BirA (a biotin ligase) can transfer the biotin moiety to acceptor peptides, among which are avitags. K562 cells were subsequently transduced with lentivirus containing the avi-tagged dCas9 or empty vector. Since also the empty vector will prompt EGFP expression, both transductions could be sorted based on EGFP expression (figure 2B). Western blot analysis using a streptavidin conjugated fluorophore on nuclei of transduced cells confirmed dCas9 expression at the protein level (figure 2C). Having established a parental cell line stably expressing dCas9, different gRNA constructs were designed against the first exons of PDK1 and SLC2A1, near the translation start site (for gRNA sequences see supplementary table I). The guide sequences were cloned into pLs expression vectors and their correctness confirmed by sequencing. In these constructs the gRNA was placed under control of a human U6 promoter, while expression of the fluorophore Blueberry was governed by a human PGK promoter (figure 3A).
Subsequently lentiviruses containing the pLsgRNA plasmid were made from 293T cells. Virus was harvested and tested on K562 cells, using Blueberry expression as a marker for transduction efficiency (figure 3B). Although the titer experiment showed the highest transduction efficiency upon the use of
Figure 2. dCas9 expression could be confirmed in K562 cells upon transduction with lentivirus. A) Schematic representation of essential part of vector transduced into K562. dCas9-avi was under control of an SFFV promoter, with EGFP-BirA being expressed from the same mRNA through an IRES sequence.
B) FACS-plots depicting either forward scatter (FSC) vs. side scatter (SSC) or GFP intensity vs PE (non- present fluorophore). dCas9 transduction efficiency was 6.6% (top three graphs) whereas the empty vector reaches 79.6% efficiency. Black lines in graph indicate gates for cell sorting. C) Western blot using streptavidin conjugated fluorophore to visualize Avi-tagged dCas9. Bottom picture is β-actin loading control. SFFV: Spleen focus-forming virus, IRES: Internal ribosome entry site, EGFP: Enhanced green fluorescent protein
undiluted virus, a five times dilution was used for larger scale transduction of dCas9 expressing cells (or GFP-control). Underlying idea was to achieve a higher probability of having a single
genomic integration per cell. As a single integration per cell is sufficient for the types of experiments desired here, multiple hits will only result in an increased background (as the desired binding site is already occupied by a gRNA) in addition to stress to the cell.
Having established that the produced viruses were functionally active, a larger scale transduction was set up. One million dCas9 cells per construct were transduced and subsequently sorted for EGFP and Blueberry expression (figure 3C and supplementary figure I). Overall transduction efficiency appeared similar in all constructs, with an average of 39% double positive cells. As there are only few Blueberry negative cells, this transduction appears highly efficient. Of note however is that the population of EGFP positive cells has decreased to about 40%. After the initial sort this was 100%, indicating at least a population of cells does not tolerate dCas9-Avi expression very well. No further decrease in the percentage of EGFP positivity was observed when culturing these dCas9 cells for another week (data not shown). Following these findings, cell lines expressing dCas9 and gRNA were grown up and aliquots were frozen as soon as possible to keep the population pure. When expanding cells, a difference in growth rate was observed between cells expressing different gRNA’s. However, a subsequent growth experiment showed no major differences in growth rate over three days (supplementary figure 2A).
Gene expression levels of the targeted genes also did not show a clear change upon gRNA expression as determined by qPCR (supplementary figure 2B).
The chromatin around the targeted SLC2A1 and PDK1 promoters can be enriched for in pull outs Before looking at the protein level, the ability to enrich for the desired stretch of chromatin had to be determined for the constructs. This was done by means of a chromatin immunoprecipitation (ChIP), using streptavidin coated beads to pull out the biotinylated avi-tagged dCas9. Of each of the 10
constructs per gene, 4 x 106 cells were crosslinked with formaldehyde. After washing and nuclear isolation, equal volumes of all samples were mixed and biotinylated proteins were precipitated overnight following sonication. Subsequent washing and DNA purification yielded the input for a qPCR using primers against the specific genomic regions (for primer sequences see supplementary table I).
As anticipated, the qPCR showed enrichment for the desired genomic locus in both the gene sets (figure 4). Both genes were probed with 2 primer sets, with the SLC2A1 performing equally well in both sets. For PDK1 there did appear to be a difference between the primer pairs. A potential explanation for that could be the position of the primer pairs relative to the guide binding sequence, as will be discussed later on. Since there was no enrichment for PDK1 chromatin in the SLC2A1 PCRs and vice-versa, it is likely the constructs precipitate not just any chromatin but specifically the region targeted.
Figure 3. Construction of cell lines expressing both dCas9 and gRNAs. A) Schematic representation of pLsgRNA.
The guide RNA fused to the scaffold section was expressed from a human U6 promoter, while the Blueberry expression originated from a separate phosphoglycerate kinase (PGK) promoter. B) Percentage of double positive cells (both GFP and Blueberry) upon transfection of K562 with gRNA targeting either the SLC2A1 or PDK1 locus, or scrambled control. For each construct pure virus was tested along with a 5x or 25x dilution. C) Representative images of sorted dCas9 cells expressing gRNA constructs. GFP and Blueberry expression are plotted and percentages of double positive cells are indicated. Re-analysis of sorted cells (S1-R and P6-R plots) indicate sorting is at least 96.5% accurate. For images of more cell lines see supplementary figure 1.
Figure 4. Pooled samples for SLC2A1 and PDK1 constructs enrich for the targeted chromatin, as analysed by two primer sets. qPCR analysis on DNA purified from streptavidin pull-down of combined PDK1 (or SLC2A1) constructs or control reveals enrichment of target DNA relative to input. For SLC2A1 both primer sets yield similar results. For PDK1 only the second primer pair gives distinct values for the different samples. As expected, there is no enrichment for SLC2A1 chromatin in the P-constructs and vice-versa. All samples measured in triplo.
Having determined an enrichment for the desired chromatin was obtained upon ChIP, samples were subsequently analysed individually. As it was clear from previous results that only one of the two PDK1 primer pairs identified the enrichment, the other one was excluded from further analyses. Although nearly all samples showed enrichment compared to input, differences between certain constructs were quit large (figure 5). The efficiency of different guides appeared not to be primarily dependent on the relative position of their binding site to the primers. This is exemplified by looking at the S- constructs; although S8 is both closest to the primers and performs the best, it is unlikely that the difference in efficiency between S8 and S2, S3 and S6 can be explained solely by their position. Moving further away from the primers there is also no decrease in signal with an increase in distance. From the PCR using PDK1 primers it was clear that S8 does not simply bind more DNA in general (supplementary figure 3A). For the guides designed against the PDK1 promoter a similar result could be seen. Here the lack of signal for the first set of primers could be caused by the large distance between the primer pair and the target sequence of the guide. Also in this case the P2 and P3 constructs (performing the best) did not simply bind more DNA (supplementary figure 3B).
Proteomics analyses of dCas9 guided pull outs yield unique peptides
Having determined that an enrichment for the desired chromatin was obtained using this approach, the attention turned to proteomics analysis. In order to obtain a large signal to noise ratio the gRNAs that performed the best in ChIP analysis were chosen for subsequent experiments, based on combined analysis of enrichment of the desired locus versus not above average enrichment of a non-targeted locus (combine figure 4,5 and supplementary figure 3A and B). gRNAs P2, P3 and S8 were chosen for subsequent experiments. Having established that sonication efficiency was maintained at higher cell densities than before (data not shown), a large scale pull-out was set up. The eluted proteins were ran on gel till the dye front ran off and subsequently processed for mass spectrometry. In addition, a duplo of the 4 conditions was ran shortly on gel. This was done to get a feel for the amount of slicing, and therefore measuring time, needed to obtain results (see supplementary figure 4 for gel pictures).
Lanes were sliced up and analysed using LC-FT-ICR/Orbitrap mass spectrometry following trypsin digestion and protein extraction. Mass spectrometry results were matched against the full human proteome database (HUMAN_UniProt_2016_06_17), 1% FDR (see Materials and Methods for full search parameters). Resulting protein hits were listed and made non redundant: as the database search is directed against multiple databases, a single protein might yield two hits (from two distinct databases). Identical protein names were filtered to obtain a single hit per protein name. These lists were subsequently converted to official gene symbols. The different samples led to different amount
Figure 5. ChIP efficiency of a gRNA is not solely dependent on its position relative to the qPCR primers. A) Binding positions of the ten SLC2A1 guides are indicated along with the regions amplified by the primers used. Blue arrow indicates translation start site. On the right the enrichment for this locus is measured after a ChIP targeting biotinylated proteins. As can be seen from the combination of both graphs, the ChIP efficiency is not solely dependent on how far away from the primers the guide binds. S3 and S8 are quite different, and there appears no drop in efficiency when moving away from the primers (compare S5 and S7).
B) Same as in A), but now for the PDK1 guides.
of proteins being identified (figure 6A). Proteins found in the scrambled control were subtracted from the S8 or combined P2-P3 lists. Identified proteins were subsequently ranked based on the amount of corresponding peptides found in the analysis (figure 6B). These lists were analysed for gene ontology (GO) terms, specifically biological process (BP). For the top 10 most abundant proteins per set (as measured by number of peptides by which they were identified) GO terms found were for example RNA splicing, protein complex assembly and cell cycle regulators (figure 6C). In all samples proteins corresponding to translation or translation regulation could be found (supplementary table 3). The lists of unique proteins per sample were also scanned for suspected regulators/binders based on
Figure 6. Unique proteins are identified in promoter targeted pull downs. A). Venn diagrams showing the number of proteins found in the different samples that were analysed against the full human proteome.
Single slice indicates the samples only ran shortly on gel, 11 slices indicates results from full lane analysis.
B). Graph showing the number of peptides unique proteins were identified with. For P2-P3 only the proteins found in both lists and not in the scrambled control were used. C). Gene ontology (GO) analysis on biological process (BP) of the 10 highest scoring proteins based on the number of peptides they were identified with.
previous ChIP-seq data. One of the hits popping up in the S8 list for example was the transcription factor STAT5B. In order to get a better overview of the potential transcription factors involved, a separate analysis was performed on the raw data. Spectra were matched against a list of transcription factors and other proteins involved in transcription regulation, now also allowing phosphorylation as a post translational modification (5% FDR, for the list see supplementary table 2). For each protein identified, a q-value was determined. This serves as an indicator of confidence for assigning a measured spectrum to an expected set of peaks based on protein sequence and potential post translational modifications. A value of 0 indicates all peptide spectrum matches used to identify a protein performed better than the highest scoring decoy.24 Unique proteins in either the S8 or combined P2-P3 lists with a q-value of 0 were identified (figure 7). The total list consisted of 1208 proteins, of which roughly one tenth was found back. These results indicate unique transcription regulators can be identified.
Figure 7. Unique transcription factors can be found in promoter targeted pull downs. Venn diagrams showing the number of identified proteins from provided list (see supplementary information 2). Proteins that were unique in the experimental fractions and had a Q-value of 0 are identified. The P3 fraction had no significant hits on its own.
Discussion
Gene expression regulation is one of the important mechanisms a cell has to ensure target protein levels are maintained within a sometimes small margin of error. Drastic changes in expression levels of certain genes cause an imbalance that cells must somehow overcome. It is therefore not surprising that mutations in transcription factors, controlling expression of a set of genes, often lead to malignancies such as cancer.25 In this study an attempt was made to elucidate which proteins are regulating expression of a specific gene, on a specific locus. The CRISPR/dCas9 system was used as a starting point because of its flexibility and its potential for targeting endogenous loci, which has been previously described.9,26
Construction of the K562 cell line stably expressing EGFP-tagged dCas9 was relatively straight forward.
As was the subsequent lentiviral transduction of these cells with vectors expressing various gRNAs (with IRES:Blueberry for fluorescence based sorting) targeting either the SLC2A1 or PDK1 promoter. Of note however was the apparent drop in EGFP expression of the dCas9 expressing cells. This indicated at least a population of cells did not tolerate dCas9:EGFP expression well. The cells were sorted following initial transduction with the dCas9 vector, but when sorting after transduction with the gRNA constructs, only 40% of the population was still EGFP positive (regardless of Blueberry expression).
Although this could be harmful to the experiment, the problem was alleviated by sorting EGFP and Blueberry double positive cells after transduction with a specific gRNA. These cells were subsequently expanded and either used immediately or frozen to prevent dropping of dCas9 expression throughout the population. For subsequent use of systems like this it might be useful to not just randomly integrate the dCas9 in the genome as was done here, but integrate it in a specific genomic region of which it is known that the integration process does no harm to the cells. In this way the question can be addressed whether the expression of dCas9 itself, or the integration into an unfavourable locus causes the apparent drop in EGFP expression for a part of the population. Although the selection of double positive cells allows for circumvention of this question in the experiments described here, it would still be interesting to see whether the cells can tolerate constitutive expression of such a large (foreign) fusion protein.
Furthermore, although no major changes in either growth rate or target gene expression were observed in cells expressing either PDK1 or SLC2A1 targeting guides, the question remains what the
effects of dCas9 binding to a locus are. It seems likely that binding of a relatively large protein to a locus disturbs the set of proteins normally present. This appears a downside to the system which can however be partially circumvented by the use of guides placed slightly further up- or downstream on the genome. Whatever the effect on local protein binding may be, the principle of a CRISPR/dCas9 mediated ChIP appears to be working. Analysis of the individual guides revealed that their ChIP efficiency is not only dependent on position relative to the primers used.
Furthermore, we observed strong differences between similar guides or guides that target the same sequence but the complementary strand. This indicates local chromatin structure, or occupation by other proteins, plays a role in on target guide efficiency. The relatively high background signal appears to be intrinsic to the approach taken, as the enrichments for non-targeted loci are similar to the ones for the scrambled control. Given the high enrichment for the target this does not seem to be a problem, as a reasonable signal to noise ratio is till obtained. Retrieving the chromatin targeted with CRISPR/dCas9 was however only the first step. The ultimate goal was to be able to analyse the proteins pulled along.
Mass spectrometry analysis of the pull out samples yielded lists of proteins unique in the sample fractions. Looking at the number of identified proteins, it appeared that they differ quite a lot between samples, even within the single slice or full lane groups (single slice means the entire lane was processed as a single slice in a single LC-MS/MS run, full lane means the lane was divided into 11 slices which were separately run through the LC-MS/MS). There can be multiple causes for this finding and the most likely seems to be a difference in material processed. The samples that appeared to contain less protein on the gel also gave the shortest list of (unique) peptides (P3 in both cases). A shorter list was also obtained if samples were processed in a single slice vs. a full lane. The increase in detected proteins is however not extreme considering full lane analysis takes significantly more effort, time to run and database searching. The question remains however if the extra depth of data gathering is needed/necessary for obtaining biologically useful results. Slicing of the lane in less than 11 pieces could very well be considered in future experiments, especially if lots of runs are to be made. In the top hits of unique proteins, GO terms related to gene regulation were unfortunately not found at 1%
FDR. Matching the single slice data with a separate list of transcription factors at a less stringent FDR (5%) and allowing for phosphorylations (which were by mistake not allowed in previous searches) did yield unique factors for both promoters. This is a promising finding. Due to time restrictions several analyses remain incomplete however. It would for example be very interesting to run all measured samples (including full lane samples) against the list of transcriptional regulators at both 1% FDR and 5%, when including phosphorylation as a PTM. Also, a scan against the full human proteome at 1% FDR including phosphorylations might provide useful insights. At the very least a background list (based on the scrambled samples) can be made for future experiments.
After obtaining results from these lists, the first step would be to confirm binding of a factor to the targeted locus by ChIP, for example for DNMT3B. Enriching for the locus will provide the proof that the identified factor is actually present at the locus, and not a measuring/analytical artefact. In addition to the ideas for future research already given, labelling of the samples for means of quantification should also be considered. The lists that were obtained from mass spectrometry provide more qualitative information as opposed to quantitative. By means of for example SILAC labelling one can run a sample along with the control and so determine fold changes of protein presence. This might pick up certain factors that are now missed because they are in the scrambled list already, whereas they might be much more abundant in the fractions derived from targeted pull outs.
In short this study has shown it is possible to enrich for a specific genetic locus using a pull out set up.
The fact that proteins could be identified using mass spectrometry indicates that the protocol used (including the formaldehyde crosslinking) does not damage proteins beyond recognition. This is an important finding. Although several technical challenges need to be overcome, the technique is over all promising. Awaiting confirmation of the identified factors, this tool might provide an opportunity to study gene regulatory elements on a specific endogenous locus under specific circumstances.
Acknowledgements
I would like to thank Bart-Jan Wierenga for daily supervision and discussions, I really learned a lot from you. Maarten Linskens for being my supervisor from the study track. Jan Jacob Schuringa for having me in the lab. Henk Moes for assistance in fluorescence based cell sorting. Esméé Joosten for help in preparing mass spectrometry samples, and Marcel de Vries for mass spectrometry data analysis.
References
1. Lee, T. I. & Young, R. A. Transcriptional regulation and its misregulation in disease. Cell 152, 1237–1251 (2013).
2. Wickramasinghe, V. O. & Laskey, R. A. Control of mammalian gene expression by selective mRNA export. Nat. Rev. Mol. Cell Biol. 16, 431–442 (2015).
3. Van Der Kelen, K., Beyaert, R., Inzé, D. & De Veylder, L. Translational control of eukaryotic gene expression. Crit. Rev. Biochem. Mol. Biol. 44, 143–168 (2009).
4. Prabakaran, S., Lippens, G., Steen, H. & Gunawardena, J. Post-translational modification:
nature's escape from genetic imprisonment and the basis for dynamic information encoding.
Wiley Interdiscip Rev Syst Biol Med 4, 565–583 (2012).
5. Wierenga, A. T. J., Vellenga, E. & Schuringa, J. J. Convergence of hypoxia and TGFβ pathways on cell cycle regulation in human hematopoietic stem/progenitor cells. PLoS ONE 9, e93494 (2014).
6. Zhou, M., Li, Q. & Wang, R. Current Experimental Methods for Characterizing Protein-Protein Interactions. ChemMedChem n/a–n/a (2016). doi:10.1002/cmdc.201500495
7. Falk, R. et al. Targeted protein pullout from human tissue samples using competitive elution.
Biotechnol J 6, 28–37 (2011).
8. Yan, P. et al. A targeted releasable affinity probe (TRAP) for in vivo photocrosslinking.
Chembiochem 10, 1507–1518 (2009).
9. Chen, B. et al. Dynamic imaging of genomic loci in living human cells by an optimized CRISPR/Cas system. Cell 155, 1479–1491 (2013).
10. Miyanari, Y., Ziegler-Birling, C. & Torres-Padilla, M.-E. Live visualization of chromatin dynamics with fluorescent TALEs. Nat. Struct. Mol. Biol. 20, 1321–1324 (2013).
11. Shoaib, M. et al. PUB-NChIP--‘in vivo biotinylation’ approach to study chromatin in proximity to a protein of interest. Genome Res. 23, 331–340 (2013).
12. Roux, K. J. Marked by association: techniques for proximity-dependent labeling of proteins in eukaryotic cells. Cell. Mol. Life Sci. 70, 3657–3664 (2013).
13. Miyanari, Y., Ziegler-Birling, C. & Torres-Padilla, M.-E. Live visualization of chromatin dynamics with fluorescent TALEs. Nat. Struct. Mol. Biol. 20, 1321–1324 (2013).
14. Hofmann, K., Wood, S. W., Brinton, C. C., Montibeller, J. A. & Finn, F. M. Iminobiotin affinity columns and their application to retrieval of streptavidin. PNAS 77, 4666–4668 (1980).
15. Roux, K. J., Kim, D. I. & Burke, B. BioID: a screen for protein-protein interactions. Curr Protoc Protein Sci 74, Unit 19.23.–19.23.14 (2013).
16. Sander, J. D. & Joung, J. K. CRISPR-Cas systems for editing, regulating and targeting genomes.
Nat. Biotechnol. 32, 347–355 (2014).
17. Pattanayak, V. et al. High-throughput profiling of off-target DNA cleavage reveals RNA- programmed Cas9 nuclease specificity. Nat. Biotechnol. 31, 839–843 (2013).
18. Gilbert, L. A. et al. CRISPR-mediated modular RNA-guided regulation of transcription in eukaryotes. Cell 154, 442–451 (2013).
19. Rumjanek, V. M., Vidal, R. S. & Maia, R. C. Multidrug resistance in chronic myeloid leukaemia:
how much can we learn from MDR-CML cell lines? Biosci. Rep. 33, 875–888 (2013).
20. Kharas, M. G. & Fruman, D. A. ABL oncogenes and phosphoinositide 3-kinase: mechanism of activation and downstream effectors. Cancer Res. 65, 2047–2053 (2005).
21. Fatrai, S., Wierenga, A. T. J., Daenen, S. M. G. J., Vellenga, E. & Schuringa, J. J. Identification of HIF2alpha as an important STAT5 target gene in human hematopoietic stem cells. Blood 117, 3320–3330 (2011).
22. Saunier, E., Benelli, C. & Bortoli, S. The pyruvate dehydrogenase complex in cancer: An old metabolic gatekeeper regulated by new pathways and pharmacological agents. Int. J. Cancer 138, 809–817 (2016).
23. Sutendra, G. & Michelakis, E. D. Pyruvate dehydrogenase kinase as a novel therapeutic target
in oncology. Front Oncol 3, 38 (2013).
24. Wenger, C. D. & Coon, J. J. A proteomics search algorithm specifically designed for high- resolution tandem mass spectra. J. Proteome Res. 12, 1377–1386 (2013).
25. Darnell, J. E. Transcription factors as targets for cancer therapy. Nat. Rev. Cancer 2, 740–749 (2002).
26. Gilbert, L. A. et al. CRISPR-mediated modular RNA-guided regulation of transcription in eukaryotes. Cell 154, 442–451 (2013).
Supplementary figures
Supplementary figure 1. Construction of K562 cell lines expressing both dCas9 and a guide RNA. Sorting images are shown, displaying the GFP vs. Blueberry expression in cells serially transduced with both constructs. On average 39% of cells is double positive for dCas9 expressing cells. –R: reanalysis of sorted cells. SCR: scrambled. By mistake no sorting images were saved for constructs P7 and P8, graphs did however look similar to the ones shown.
Supplementary figure 2. K562 cells expressing both dCas9 and a gRNA targeting either SLC2A1 or PDK1 show no major growth defects or changes in target gene expression. A) Growth experiment on 150.000 cells showing the fold change in cell number after 65 hours of culture. B) qPCR analysis of SLC2A1 or PDK1 expression. RPL27 was used as housekeeping gene.
Supplementary figure 3. Different gRNA constructs have different background levels of chromatin precipitation. A) Samples from SLC2A1 constructs ChIP assayed with PDK1 primer pairs. Although the amount of background chromatin precipitated is similar between the primer pairs for the same sample, it is variable between samples. Note the S8 construct (that performed best in the SLC2A1 PCRs), has a medium background compared to other constructs. B) Samples from PDK1 constructs ChIP assayed with SLC2A1 primer pair. Also here the top performing constructs on the target region (P2 & P3) do not have an abnormally high background.
Supplementary figure 4. Gel pictures showing what the input for mass spectrometry was derived from.
Samples were either ran till the dye front ran off (top), or only shortly (bottom). Top gel was subsequently cut in 11 slices, and the bottom gel was processed as a single slice per sample. Differences in the amount of protein loaded can clearly be seen. P3 in both cases appears to contain relatively little protein.
Supplementary table 1
All primer sequences are written 5’ to 3’, oligo’s were ordered with standard desalting.
sgRNA assembly primers:
SLC2A1 Assembly 1 FWD CACCGCCTCGCTCAGGCTGCCGAT SLC2A1 Assembly 1 REV aaacATCGGCAGCCTGAGCGAGGC SLC2A1 Assembly 2 FWD CACCGCGTCGCCAGCCAATGGCCG SLC2A1 Assembly 2 REV aaacCGGCCATTGGCTGGCGACGC SLC2A1 Assembly 3 FWD CACCGCCGGGGTCCTATAAACGCTA SLC2A1 Assembly 3 REV aaacTAGCGTTTATAGGACCCCGGC SLC2A1 Assembly 4 FWD CACCGTCCTGGGCGGCGCTCTGCG SLC2A1 Assembly 4 REV aaacCGCAGAGCGCCGCCCAGGAC SLC2A1 Assembly 5 FWD CACCGTGTGGGCAGGACCTCGGGG SLC2A1 Assembly 5 REV aaacCCCCGAGGTCCTGCCCACAC SLC2A1 Assembly 6 FWD CACCGATAGGACCCCGGCCATTGGC SLC2A1 Assembly 6 REV aaacGCCAATGGCCGGGGTCCTATC SLC2A1 Assembly 7 FWD CACCGCGGCAGCCTGAGCGAGGCAG SLC2A1 Assembly 7 REV aaacCTGCCTCGCTCAGGCTGCCGC SLC2A1 Assembly 8 FWD CACCGCCGTAGCGTTTATAGGACCC SLC2A1 Assembly 8 REV aaacGGGTCCTATAAACGCTACGGC SLC2A1 Assembly 9 FWD CACCGAGCGAGGCAGTGGTTAGGGG SLC2A1 Assembly 9 REV aaacCCCCTAACCACTGCCTCGCTC SLC2A1 Assembly 10 FWD CACCGGGCCCCAGGCCCCGCCCCG SLC2A1 Assembly 10 REV aaacCGGGGCGGGGCCTGGGGCCC PDK1 Assembly 1 FWD CACCGCTTCAGCCGCAGCTTCAGCT PDK1 Assembly 1 REV aaacAGCTGAAGCTGCGGCTGAAGC PDK1 Assembly 2 FWD CACCGACTCGGCATGAGGCTGGCG PDK1 Assembly 2 REV aaacCGCCAGCCTCATGCCGAGTC PDK1 Assembly 3 FWD CACCGAGGCTGGCGCGGCTGCTTCG PDK1 Assembly 3 REV aaacCGAAGCAGCCGCGCCAGCCTC PDK1 Assembly 4 FWD CACCGAGGAACTGCTTCATGGAGAG PDK1 Assembly 4 REV aaacCTCTCCATGAAGCAGTTCCTC PDK1 Assembly 5 FWD CACCGCGCGGCGTTCCGGGCCAGG PDK1 Assembly 5 REV aaacCCTGGCCCGGAACGCCGCGC PDK1 Assembly 6 FWD CACCGAACTGCTTCATGGAGAGCG PDK1 Assembly 6 REV aaacCGCTCTCCATGAAGCAGTTC PDK1 Assembly 7 FWD CACCGCCGAGTCCGAGCTGAAGCTG PDK1 Assembly 7 REV aaacCAGCTTCAGCTCGGACTCGGC PDK1 Assembly 8 FWD CACCGAGCGCGCGTAGAAGTCCACC PDK1 Assembly 8 REV aaacGGTGGACTTCTACGCGCGCTC PDK1 Assembly 9 FWD CACCGAAGCTGCGGCTGAAGCCGG PDK1 Assembly 9 REV aaacCCGGCTTCAGCCGCAGCTTC PDK1 Assembly 10 FWD CACCGCTGAAGCTGCGGCTGAAGC PDK1 Assembly 10 REV aaacGCTTCAGCCGCAGCTTCAGC SCR-sequence (already cloned
into pLs) GCAACAAGATGAAGAGCACC
ChIP primers to confirm pulling out chromatin near translation start site.
SLC2A1-TSS-F1 atggccggggtcctataaacg SLC2A1-TSS-R1 acgctcgctgttgctacct SLC2A1-TSS-F2 ggcaagaggcaagaggtagc
SLC2A1-TSS-R2 ctcccactgcgactctgact PDK1-TSS-F1 tccataccgcctcgcctcttag PDK1-TSS-R1 tacacgtcgggtgatgggactg PDK1-TSS-F2 acgtccctcacgtaccactc PDK1-TSS-R2 cagccagtacgccaggtttc
Sequencing primers
dC9seq_845 cccagatcggtgaccaatac
dC9seq_1688 tgaagcagctgaaggaggac
dC9seq_2495 tcaacaggctctcagactac
dC9seq_3250 accgtgcgcaaagtgctttc
pJetseq for gcctgaacaccatatccatcc
pJetseq rev tacagcctgaaaatcttgagag
Supplementary table 2
ACTB BATF CNOT1 DNMT1 ETS2 GABPB1 HEY1 HOXC5 KDM5A MEF2A NANOG NR2F6 POLR2K RERE SLC26A3 SUPT4H1 TEAD1 UHRF1 ZIC1 ZNF226 ZNF396 ZNF570 ZNF761 ZSCAN31
ADAR BATF2 CNOT6 DNMT3A ETV1 GAS7 HEY2 HOXC6 KDM5B MEF2B NCOR1 NR3C1 POLR2L REST SLC2A4RG SUPT6H TEAD2 UHRF2 ZIC2 ZNF227 ZNF397 ZNF571 ZNF770 ZSCAN32
AEBP1 BATF3 CNOT7 DNMT3B ETV2 GATA1 HEYL HOXC8 KLF1 MEF2C NCOR2 NR3C2 POU1F1 REXO4 SLC30A9 SUV39H1 TEAD3 UPF1 ZIC3 ZNF229 ZNF41 ZNF573 ZNF772 ZSCAN4
AEBP2 BAZ1B CNOT8 DNMT3L ETV4 GATA2 HIC1 HOXD10 KLF10 MEF2D NEUROD1 NR5A1 POU2F1 RFX2 SLC50A1 SUZ12 TEAD4 USF2 ZIK1 ZNF23 ZNF415 ZNF577 ZNF773 ZSCAN5A
AFF1 BAZ2A CREB1 DOT1L ETV5 GATA3 HIF1A HOXD3 KLF11 MEIS1 NEUROD2 NR5A2 POU2F2 RFX3 SMAD1 TADA2A TEF VAV1 ZIM2 ZNF232 ZNF418 ZNF583 ZNF780A ZSCAN5B
AFF4 BCDIN3D CREB3 DPY30 ETV6 GATA4 HIF3A HOXD4 KLF12 MEIS2 NEUROG1 NR6A1 POU2F3 RFX5 SMAD2 TADA2B TERT VAX2 ZIM3 ZNF234 ZNF419 ZNF584 ZNF780B ZSCAN9
AGO1 BCL3 CREB3L1 DRAP1 EVX1 GATA5 HILS1 HOXD8 KLF17 MEOX1 NFAT5 NRL POU3F1 RFX7 SMAD3 TADA3 TET1 VDR ZKSCAN1 ZNF235 ZNF420 ZNF585A ZNF782 ZXDA
AGO2 BCL6 CREB3L2 DROSHA EXOSC10 GATA6 HINFP HR KLF2 MEOX2 NFATC1 OTX2 POU3F2 RFX8 SMAD4 TAF10 TFAM VSX1 ZKSCAN2 ZNF239 ZNF429 ZNF585B ZNF789 ZXDC
AGO3 BCOR CREB3L4 DYDC1 EZH2 GATAD1 HIRA HSF1 KLF3 MESP1 NFATC2 PA2G4 POU3F3 RFXANK SMAD5 TAF12 TFAP2A WNT5A ZKSCAN3 ZNF24 ZNF43 ZNF586 ZNF79
AGO4 BCORL1 CREB5 DYDC2 FEV GATAD2A HIST1H2AA HSF2 KLF4 METTL3 NFATC3 PADI2 POU3F4 RFXAP SMAD6 TAF13 TFAP2B WT1 ZKSCAN4 ZNF250 ZNF432 ZNF587B ZNF790
AHCTF1 BEND3 CREBBP E2F1 FLI1 GATAD2B HIST1H2AB HSF4 KLF6 MGA NFATC4 PARK2 POU4F3 RHOXF1 SMAD7 TAF1A TFAP2C XBP1 ZKSCAN5 ZNF251 ZNF436 ZNF595 ZNF792
AHR BHLHE40 CREBL2 E2F2 FMR1 GBX2 HIST1H2AC HSF5 KLF7 MIER1 NFE2 PAX1 POU5F1 RING1 SMAD9 TAF1B TFCP2 YBX1 ZKSCAN7 ZNF254 ZNF438 ZNF600 ZNF793
