An improved auxin-inducible degron system preserves native protein levels and enables rapid and specific protein depletion

1The function of proteins can be studied in cells using RNAi depletion, loss-of-function mutants, temperature-sensitive mutations, small molecule inhibitors, CRISPR interference, or nucleic acid aptamers. The two greatest limitations of these methods are (1) the chronic nature of permanently disrupting function and (2) the limited availability and specificity of inhibitor molecules. Inducible degradation methods, such as auxin-inducible degron (AID) or degradation tag (dTAG) systems, can overcome these limitations. Rapidly regulated systems and inhibitors permit measurements of primary changes in molecular, cellular, and organismal phenotypes and subsequent tracking of the cascade that accompanies immediate protein dysregulation. Gene-editing techniques now permit endogenous tagging of genes, which should preserve the target gene's expression levels and patterns. Endogenous fusion of target genes with rapidly inducible degradation tags have the potential to revolutionize the way we study biological systems by providing specific and temporally defined perturbation techniques for any protein of interest.

2Auxin is a plant hormone that regulates various aspects of plant growth and development. In plants, auxin signaling triggers a rapid switch between transcriptional repression and transcriptional activation (Supplemental Fig. S1;;; ). There are three key components in this signal transduction system: (1) transport inhibitor response 1 (TIR1), a subunit of a ubiquitin ligase complex that binds to the target substrate; (2) the auxin response transcription factors (ARF), which directly regulate gene expression; and (3) auxin/indole-3-acetic acid (Aux/IAA) proteins, which are destabilized in the presence of auxin-mediated ubiquitination. In the absence of auxin, domains III and IV of the Aux/IAA family of proteins form heterodimeric complexes with the PB1 (Phox and Bem1) domain of ARF proteins. Domain I of Aux/IAA interacts with a plant-specific transcriptional corepressor, Topless (TPL). The repressive function of TPL dominates relative to the activation function of ARF transcription factors. In the presence of auxin, TIR1 interacts with domain II of Aux/IAA to facilitate ubiquitination and degradation of Aux/IAA, which liberates ARFs to regulate transcription. This AID system was engineered to function outside plant cells.

3Although the AID system has been widely adopted to degrade tagged proteins of interest, recent studies report auxin-independent depletion of endogenously tagged proteins. For instance, tagging the centromeric histone chaperone HJURP results in depletion of >90% of HJURP protein in the absence of auxin in human cell lines. Chronic depletion has also been reported in chicken cells and yeast cells, suggesting that auxin-independent depletion may be a systemic problem when tagging endogenous genes. The extent of factor-dependent depletion is often impossible to evaluate because many studies do not directly compare protein levels in the progenitor and tagged cell lines. Neither defining a minimal degradation domain of AID nor using an inducible TIR1 system has overcome this deficiency. In yeast, auxin-independent depletion of the AID-tagged factors is dependent on cellular TIR1 concentration, suggesting that depletion is due to ubiquitin-mediated degradation.

4Here, we show that AID tagging of endogenous genes commonly results in auxin-independent and chronic depletion by proteasome-mediated protein degradation. We found that coexpressing the PB1 domain of ARF improved the robustness of the AID system by rescuing auxin-independent degradation and increasing the rate of auxin-induced degradation. Our control experiments also revealed that auxin treatment alone results in the activation of the aryl hydrocarbon receptor (AHR) transcription factor and regulation of AHR target genes. Excluding AHR-responsive genes from downstream differential expression analyses is critical when investigating the activity of AID-tagged transcriptional regulators. Collectively, these improvements enhance the robustness, sensitivity, and specificity of the AID system. We applied ARF-rescued AID-mediated rapid degron depletion to the transcription factor ZNF143 to identify the primary ZNF143 response genes and define a functional role of ZNF143 in transcriptional regulation.

5In nonplant systems, the auxin-mediated degradation system requires the presence of exogenously expressed TIR1. We generated an HEK293T cell line with the TIR1 gene stably integrated into the AAVS1 locus (Supplemental Fig. S2;; ). We independently tagged all copies of either ZNF143, TEAD4, or p53 in these progenitor HEK293T-TIR1 cells (Fig. 1A). In the absence of auxin, tagging the factors resulted in depletion of the proteins to levels that range from between <3% and 15% of endogenous levels, as measured by quantitative Western blots (Fig. 1B–D). To determine whether chronic depletion is unique to HEK293T cells, we generated an MCF-7 cell line with TIR1 incorporated heterozygously into the AAVS1 locus and tagged all copies of ZNF143, which also resulted in auxin-independent ZNF143 depletion (Fig. 1E). Next, we characterized a previously published AID-tagged CENP-I colorectal epithelial DLD-1 cell line. CENP-I is depleted to ∼50% of the abundance of the progenitor DLD-1-TIR1 cell line (Fig. 1F). The limited degree of CENP-I depletion relative to ZNF143, TEAD4, and p53 may be because CENP-I is an essential protein, and a minimal abundance of CENP-I is necessary for cell line viability. Basal depletion of these factors in three distinct cell lines confirms the generality of auxin-independent depletion of endogenously tagged proteins.

6We next sought to investigate the mechanism by which AID-tagged proteins are depleted by focusing on AID-tagged ZNF143. We performed nascent RNA sequencing in the progenitor HEK293T-TIR1 cells and the ZNF143-AID cells. Nascent ZNF143 transcript levels remain unchanged (1.08-fold increase; false discovery ate [FDR] = 0.27) in the AID-tagged ZNF143 cell line (Fig. 2A). Because the auxin-inducible system is dependent on the proteasome and because AID can interact with TIR1 in the absence of auxin, we hypothesized that low protein abundance may be mediated by an auxin-independent interaction with TIR1, ubiquitination, and subsequent proteasome-mediated degradation. We treated ZNF143-AID cells with the proteasome inhibitor MG132 and observed a modest increase in ZNF143 levels after 4.5 h (Supplemental Fig. S3A). TIR1 depletion also results in higher ZNF143-AID and TEAD4-AID levels (Supplemental Fig. S3B,C). These results indicate that chronic auxin-independent depletion of AID-tagged proteins is TIR1-mediated and due to proteasomal degradation (Supplemental Fig. S3D).

7TIR1 and AID proteins can directly interact in the absence of auxin in vitro. However, degradation is tightly and robustly regulated in plants (Supplemental Fig. S1; ). We proposed that expression of an ARF protein, which is an AID interaction partner, may confer protection of the tagged protein from auxin-independent ubiquitination and degradation. To test this hypothesis, we supplemented the engineered AID system by expressing the AID-interacting PB1 domain of ARF. We reconstituted the system by expressing the PB1 domain of Oryza sativa ARF16 and ARF25 based on yeast two-hybrid experiments that quantified interaction of these ARFs with IAA proteins. Moreover, ARF16 and ARF25 harbor conserved charged residues, corresponding to K944, D994, and D998 of ARF16, at critical positions within the ARF/IAA-binding interface. Transfection of either eGFP-ARF25-MR-PB1 or eGFP-ARF16-PB1 stabilizes TEAD4-AID, with ARF16 promoting a higher degree of TEAD4 stability (Supplemental Fig. S4A,B). These results prompted us to generate HEK293T-TIR1-ZNF143-AID cells with stable genetic integration and expression of ARF16-PB1. This strategy restored ZNF143 levels to >50% of untagged levels (Fig. 2B; Supplemental Fig. S4C,D). Similarly, we found that genetic integration and expression of eGFP-ARF16-PB1 stabilized endogenously tagged TEAD4-AID (Fig. 2C). In contrast, stable expression of eGFP-ARF16-PB1 did not alter ZNF143, TEAD4, or p53 protein levels in HEK293T or HEK293T-TIR1 cells (Supplemental Fig. S4E).

8Transcriptional output is a quantitative measure of ARF-mediated functional rescue of ZNF143. We performed nascent RNA transcriptional profiling with the three successive genetically modified HEK293T cells: progenitor TIR1 cells, chronic ZNF143-depleted AID-tagged cells, and ARF-rescued ZNF143-AID cells. Chronic ZNF143 depletion resulted in activation of 1188 genes and repression of 774 genes at a FDR of 0.001 (Fig. 2A). Next, we analyzed the raw changes in expression upon ARF rescue to determine whether rescuing ZNF143 stability can functionally rescue gene expression profiles. Of the 1188 genes activated upon chronic ZNF143 depletion, 899 (76%) decrease their expression upon ARF rescue (Fig. 2D). Of the 774 chronically repressed genes, 561 (72%) increase their expression upon ARF rescue (Fig. 2D). These changes are consistent with a functional rescue of gene expression upon ARF rescue of ZNF143 stability.

9We treated ZNF143-AID and TEAD4-AID cells with 500 µM auxin to determine how the ARF16 rescue affects inducible depletion. Auxin treatment induces degradation of both ZNF143-AID and TEAD4-AID in a time-dependent manner (Fig. 3A,B). Importantly, the rate of degradation of ZNF143 was increased upon ARF16-PB1 rescue when compared with cells not rescued with ARF16-PB1 (Fig. 3C–E). To test whether ARF16-PB1 rescue affects the synthesis rate of ZNF143-AID and thus contributes to the perceived degradation rate, we simultaneously blocked new protein synthesis with cycloheximide at the time of auxin treatment (Supplemental Fig. S5). Upon blocking protein synthesis, the ZNF143-AID protein degraded faster in the presence of ARF16-PB1 (Supplemental Fig. S5). Therefore, ARF expression promotes faster degradation kinetics and does not influence protein synthesis rate.

10To test the specificity of the ARF16-PB1 rescue, we mutated ARF16 residues within the interaction domain interface that are critical for its interaction with IAA17 (Fig. 4A,B;; ). We converted K944, D994, and D998 to alanine in the ARF16-PB1-MT. In Arabidopsis thaliana, the corresponding mutations abolish ARF homodimerization and heterodimerization with IAA17. We tested whether this mutant is capable of rescuing AID-tagged protein stability in HEK293T-TIR1 cells. Chronic depletion of AID-tagged ZNF143 and TEAD4 was rescued with wild-type ARF16-PB1 but not with ARF16-PB1-MT (Fig. 4C,D). However, we note that the ARF16-PB1 mutant protein is not as abundant as the wild type ARF16-PB1 (Fig. 4C,D), presumably because dimerization can stabilize exogenous ARF16-PB1. This result suggests that rescue of stability is mediated by the interaction between ARF16-PB1 and AID. To confirm a physical interaction, we performed a coimmunoprecipitation experiment. We found that NLS-mCherry-AID and eGFP-ARF16-PB1 interact in the absence of TIR1 (Fig. 4E). However, the ARF16 mutations are sufficient to disrupt the ARF16/AID interaction (Fig. 4E). The reciprocal immunoprecipitation confirmed the AID/ARF16-PB1 interaction, and we were unable to detect an interaction of TIR1 with AID (Fig. 4F). These data indicate that an ARF16-PB1/AID interaction mediates AID stability in the improved ARF-AID degron system.

11Rescuing chronic ZNF143 depletion does not result in full recovery of protein levels, and ∼25% of the genes that are differentially expressed in the chronic depletion remain dysregulated upon ARF rescue (Fig. 2D). To further improve the system and preserve native levels of the tagged protein, we constitutively expressed ARF16-PB1 prior to AID tagging. We generated a bicistronic construct with ARF-PB1 and TIR1 and an intervening P2A site, which separates the two proteins during translation. Genetic incorporation of this construct into the AAVS1 locus resulted in two independent clones that express ARF16-PB1 and TIR1 (Fig. 5A). ZNF143 AID tagging of both ARF-PB1/TIR1 HEK293T progenitor cell lines results in ZNF143 protein levels that are comparable with the parental lines (Fig. 5B,C).

12We directly and quantitatively assessed auxin-induced perturbation of ZNF143 by measuring genome-wide transcriptional changes upon auxin treatment. Conventional transcriptional profiling, such as RNA sequencing (RNA-seq), requires mature RNA to accumulate or degrade to detect changes in transcription. Delayed detection poses a challenge when measuring the immediate transcriptional response upon rapid protein depletion. To overcome this limitation, we measured nascent RNA using precision nuclear run-on sequencing (PRO-seq). PRO-seq quantifies the immediate effect that degradation has upon transcribing RNA polymerases. ZNF143 was first characterized as a sequence-specific activator, as measured by reporter assays. In order to identify direct ZNF143 target genes and primary response genes, we performed genome-wide nascent RNA profiling after 90 min of auxin treatment in the chronically depleted and ARF-rescued ZNF143-AID cell lines. We define “primary effect genes” as immediately regulated upon rapid ZNF143 depletion and “direct gene targets” as primary response genes that ZNF143 regulates in cis. Many genes are activated and repressed upon auxin treatment in both backgrounds (Fig. 6A). ZNF143 binding, as measured by ChIP-seq (chromatin immunoprecipitation [ChIP] combined with high-throughput sequencing), is enriched proximal to the repressed gene class in the ARF rescue background (Kolmogorov-Smirnov two-sided P-value = 1.1 × 10−16) and not the activated gene class (P-value = 0.047) (Fig. 6B). In the chronic ZNF143 depletion background, the auxin-repressed gene class is not significantly closer to ZNF143-binding sites (P-value = 0.022), and genes within the activated gene class tend to be further away from ZNF143-binding sites (P-value = 0.0017) (Fig. 6B). This supports recent genomic data, which found that many transcription factors (ER, GR, PPARγ, NFκB, and HSF) specialize to exclusively activate or repress transcription in cis.

13Transcription factors can direct bidirectional transcription at their binding sites. Bidirectional transcription is a feature of gene promoters and enhancers and a measure of regulatory element activity. We tested whether we could detect changes in bidirectional nascent RNA production upon ZNF143 depletion. We implemented a discriminative regulatory element detection method (dREG) to identify promoters and enhancers de novo. We performed differential bidirectional transcription analysis to identify regulatory elements that increase and decrease transcriptional activity upon auxin treatment (Supplemental Fig. S6). Next, we performed de novo motif analysis within the regulatory elements that increase or decrease bidirectional transcription. The canonical ZNF143 motif was found exclusively in the auxin-repressed regulatory elements (Fig. 6C). Taken together with the integrative ChIP-seq/PRO-seq analysis from Figure 6, A and B, we conclude that ZNF143 activates transcription of proximal genes and enhancers. These results serve as direct evidence that we are not only depleting ZNF143 protein levels but functionally perturbing ZNF143 activity using the ARF-AID system.

14Of the 168 genes that are classified as repressed upon auxin treatment in the ARF rescue background, 167 genes have a net negative change in auxin-induced gene expression in the chronic depletion background (Supplemental Fig. S7A). Upon auxin treatment, 87% (146 out of 168) of the genes have a greater magnitude of response in the ARF rescue background compared with chronic ZNF143-depleted cells (Supplemental Fig. S7A). These data show that the ARF rescue is more sensitive to detect auxin-induced changes in ZNF143-dependent transcription compared with the chronically depleted background. Seventy-eight percent (57 out of 73) of the auxin-repressed genes in the chronic ZNF143 depletion background are categorized as repressed in the rescue as well (Supplemental Fig. S7B); 70% (40 out of 57) are repressed to a greater magnitude in the rescue (Supplemental Fig. S7B). This analysis indicates that expressing ZNF143 at near-endogenous levels is necessary to detect a robust transcriptional response upon ZNF143 depletion.

15In order to determine whether this system could be generally applied to study transcription factor function, we performed a control experiment to test whether auxin treatment alone affects transcription of human genes. Few genes are repressed upon auxin treatment (Supplemental Fig. S8A) regardless of FDR thresholds. However, over a range of FDR thresholds, we consistently observed that the activated genes (Supplemental Fig. S8) are enriched in AHR-binding sequences in their promoters and that the most enriched pathway for this gene set is the AHR pathway (q-value = 0.002). We found that AHR binding is enriched proximal to the activated gene class (P-value = 3.6 × 10−11) and not the repressed gene class (P-value = 0.74) (Supplemental Fig. S8B,C). This control experiment highlights the importance of filtering AHR response genes from analyses when using any AID system to study transcriptional response.

16Transcription can be regulated at various steps, including chromatin opening, preinitiation complex formation/stability, and RNA polymerase II (Pol II) recruitment, Pol II initiation, Pol II pausing and release, and elongation. General transcription machinery and cofactors directly catalyze these steps, but these factors are targeted to DNA by sequence-specific transcription factors, such as ZNF143. We sought to determine which transcription step(s) ZNF143 targets by characterizing the change in RNA Pol II profiles after rapid ZNF143 depletion (Supplemental Fig. S9). We found that the repressed gene class, which represents direct ZNF143 targets, shows dramatic changes in the pause region compared with the gene body (Fig. 7A,B; Supplemental Figs. S10, S11).

17We implemented a mathematical modeling approach to better understand the potential mechanisms underlying decreased pause/body densities for repressed genes following ZNF143 depletion. We formulated a two-compartment model with dynamics for a pause region and gene body region. Model parameters included rate constants for transcriptional initiation/Pol II recruitment, premature pause release, pause release into productive transcription elongation, and termination of transcription (Fig. 7C). This model showed that only changes in initiation and premature pause release could account for a large magnitude of Pol II density change in the pause region relative to the gene body (Fig. 7D). These results suggest that decreases in the pause and gene body regions observed following ZNF143 depletion could be accounted for by either decreases in initiation or increases in nonproductive pause release. Moreover, the model predicted that the pause index would not change if initiation rate or premature pause release rate were affected by ZNF143 depletion, which is consistent with our result that the pause index at repressed genes is only modestly changed at genes that are repressed upon ZNF143 depletion (Supplemental Fig. S12). These combined experimental and modeling results suggest that ZNF143 regulates pausing density by facilitating RNA Pol II initiation or preventing premature dissociation of paused RNA Pol II.

18Expression of ARF-PB1 improves the auxin-inducible degradation system. First, ARF-PB1 interacts with the AID-tagged protein to prevent degradation in the absence of auxin; thus, the tagged protein's abundance is more representative of native levels. Second, the rate of auxin-induced depletion is increased in the ARF-AID system. We demonstrated the power of the ARF-AID system by rapidly depleting the transcription factor ZNF143 and quantifying genome-wide changes in RNA polymerase density.

19Protein function can be studied by rapidly inducible degron systems in cases where translational fusion of the degron tag does not disrupt protein function or protein stability. These systems provide advantages in interpretation of protein function because measurements taken immediately after protein dysregulation can be attributed directly to the protein of interest. In contrast, other techniques that are general, such as RNAi and genetic knockout, do not provide opportunities to assay phenotypes immediately after protein depletion due to the gradual or chronic nature of dysregulation. The newly developed dTAG system provides advantages comparable with those of ARF-AID, and exogenous expression of two additional proteins is not required. Endogenous tagging of BRD4 with FKBP12F36V dTAG did not result in dramatic protein depletion in the absence of the inducible degradation molecule dTAG-13. As the dTAG and ARF-AID systems become more widely adopted, we look forward to studies that systematically compare these different degron technologies. We found that auxin treatment alone causes undesired transcriptional changes at AHR target genes, but it is unclear whether dTAG-13 treatment results in off-target changes in cellular phenotypes. These types of side effects can be abrogated by including proper control experiments and depleting target proteins by multiple independent methods.

20ARF transcription factors are a critical component of the plant auxin response system. Here we found that ARF expression is an important component of engineered AID systems. The ARF-AID system confers two distinct advantages. (1) ARF expression limits auxin-independent degradation of target proteins, and (2) ARF expression promotes more rapid auxin-inducible degradation of AID-tagged proteins. ARF and Aux/IAA proteins harbor a conserved PB1 domain that can homodimerize or heterodimerize. Mutations that interfere with ARF/AID interaction fail to rescue chronic auxin-independent AID degradation. TIR1 binds to domain II of Aux/IAA, and ARF binds to domains III and IV; therefore, ARF and TIR1 do not directly compete for the same binding surface of Aux/IAA. We propose that an ARF/AID interaction may cause conformational changes within AID that inhibit its interaction with auxin-unbound TIR1 (Fig. 8).

21Auxin-inducible degradation of ZNF143 is more rapid in the ARF-AID system. There are many plausible and nonmutually exclusive possibilities for the observed difference in degradation kinetics. The simplest explanation is that promoting the stability of the AID-tagged protein increases its concentration, which directly affects degradation kinetics. Alternatively, ARF binding to AID may promote more efficient ubiquitination by exposing target residues to the auxin-bound TIR1 complex. ARF binding may cause allosteric changes in domain II of AID that promote a higher-affinity interaction with auxin-bound TIR1 compared with ARF-unbound AID. Recent work has shown that different Aux/IAA proteins can interact with TIR1 in the absence of auxin to influence the association of auxin with TIR1. Although we did not detect an interaction of TIR1 and IAA17 in Figure 4, we cannot dismiss such an interaction in cells. Therefore, the ARF-AID system may promote faster kinetics because ARF binding affects AID structure, which in turn modulates the affinity of TIR1 and auxin. Future studies and systematic comparisons are needed to determine the mechanisms that contribute to a more rapid auxin-inducible response.

22Despite incredible advances in our understanding of the mechanisms of eukaryotic transcription and developments in systems biology, accurately predicting direct target genes and primary response genes of transcription factors remains a challenge. Proximal binding of a transcription factor to a gene is neither necessary nor sufficient to modulate gene expression. A fundamental question remains: How do transcription factors discriminate between genes in the genome in which to regulate? To begin to address this question, we must first define the set of genes regulated by the transcription factor of interest. Technical limitations preclude experimentally identifying a comprehensive set of primary response genes for the vast majority of transcription factors because we cannot rapidly induce or rapidly repress their activity. Perturbation methodologies that can be universally applied to any gene, such as RNAi, require days to efficiently deplete protein. This time frame of depletion poses a major barrier to understanding transcription factor function because secondary (and beyond) effects dominate conventional depletion/knockout methods. Here, we show that applying the ARF-AID system to study transcription factor function overcomes these challenges.

23The Xenopus laevis homolog of ZNF143, Staf (selenocysteine tRNA gene transcription activating factor), was first cloned and characterized nearly 25 yr ago. This original report characterized the binding site of Staf within a regulatory element of the tRNASec gene and characterized the activator function of Staf using reporter assays. Recent transcriptional profiling upon siRNA-mediated ZNF143 depletion identified many activated and repressed genes. ZNF143 depletion caused twice as many genes to decrease expression relative to the number of genes that increased expression. The investigators concluded that ZNF143 is primarily an activator but note that ZNF143 may be involved in repression. Our results corroborate the activation function of ZNF143 and indicate that although many genes are activated upon immediate ZNF143 depletion, ZNF143 does likely not act in cis to mediate repression. Importantly, we depleted ZNF143 for only 90 min and measured nascent RNA levels, so the repressive role of ZNF143 cannot be attributed to the postprimary response of ZNF143 dysregulation. Alternative mechanisms of immediate indirect repression, such as squelching, may be responsible for the observed repressive role of ZNF143. We further characterized ZNF143's role in activation and found that ZNF143 functions to control paused RNA Pol II density. Mathematical modeling indicates that ZNF143 either positively regulates RNA polymerase initiation or prevents nonproductive dissociation of paused Pol II.

24ZNF143 is also involved in chromatin looping of distal enhancers to promoters. Enhancer–promoter looping frequently and preferentially occurs at promoters containing paused RNA Pol II. Therefore, we propose a model in which ZNF143 directly regulates the amount of paused Pol II on a given promoter, which facilitates enhancer looping.

25Rapidly inducible systems, such as hormone signaling and heat-shock response, have contributed greatly to our understanding of transcriptional regulation. The success of these models in part is because the regulatory processes can be triggered instantaneously and tracked. New technologies and inhibitors that permit rapid and specific protein dysregulation promise to revolutionize the study of complex regulatory mechanisms.

26HEK293T cells were purchased from American Type Culture Collection (ATCC) and were grown in DMEM with 10% FBS, penicillin/streptomycin, and 5% glutamine. MCF7 cells were purchased from ATCC and were grown in DMEM with 10% FBS and penicillin/streptomycin. CENP-I-AID-eGFP DLD1-OsTIR1 cells were generated in Ian Cheeseman's laboratory and grown in RPMI1640 medium with 10% FBS and penicillin/streptomycin.

27OsTIR1 was integrated into AAVS1 locus of the HEK293T cells using the CMV-OsTIR1-PURO plasmid from Masato Kanemaki (pMK232; Addgene, 72834). OsTIR1 was integrated into the genome by CRISPR–Cas9-mediated repair using an sgRNA targeting AAVS1 safe harbor locus cloned into pSpCas9(BB)-2A-GFP from Feng Zhang (PX458; Addgene, 48138). The ARF16-PB1 domain (amino acids 878–1055) and ARF25 (MR and PB1 domain, amino acids 369–889, Os12t0613700-01) genes were codon-optimized for humans and synthesized from Bio Basic, Inc. We inserted ARF25-MR-PB1 into the eGFP-C2 vector. A nuclear localization signal (NLS) was added at the N terminus of the ARF16-PB1 domain, and NLS-ARF16-PB1 was inserted into eGFP-C2 vector digested with XhoI and HindIII by cold fusion cloning (pMGS36; Addgene, 126581). The CMV-OsTIR1-PURO plasmid was digested with AfeI, and we inserted the codon-optimized ARF16-PB1 domain separated by P2A from OsTIR1 to generate the ARF16-PB1-HA-P2A-OsTIR1 construct (pMGS46; Addgene, 126580). Additionally, we generated two GFP-ARF16-P2A-TIR1 plasmids that can be incorporated into a safe harbor locus: GFP-ARF16-PB1-MCS-P2A-OsTIR1(pMGS55; Addgene, 129667) and GFP-ARF16-PB1-P2A-OsTIR1(pMGS56; Addgene, 129668). The ARF16-PB1-HA-P2A-OsTIR1 plasmid was cotransfected with AAVS1 sgRNA (pMGS7; Addgene, 126582), and we selected puromycin-resistant clones to generate the homozygously integrated transgenic cell line. The resulting transgenic HEK293T-TIR1 and ARF16-PB1-HA-P2A-OsTIR1 cell lines were used to tag transcription factors with the AID tag using the CRISPR–Cas9 system. The NLS-mCherry-AID plasmid was constructed by digesting pCDNA5 vector with PmeI enzyme and inserting the NLS-mCherry-AID fragment using cold fusion cloning (System Biosciences).

28Endogenously AID-tagged TEAD4, ZNF143, and p53 cells were generated using CRISPR-mediated gene editing. sgRNAs that target the 3′ end of the respective coding sequences were cloned into hSpCas9 plasmid (PX458; Addgene, plasmid 48138). The linear donor was generated by PCR and gel-purified from a plasmid harboring a synthetic AID-P2A-hygromycin insert (pMGS54; Addgene, 126583). Note that this plasmid contains the full-length AID, and the mini-AID tag is unlikely to function in the ARF-AID system due to mini-AID's lack of domains III and IV. We amplified the AID-P2A-hygromycin insertion using primers that contain 50-nucleotide homology tails. The primers contained 5′ phosphorothioate modifications to increase PCR product stability in the cell. The primers used for making PCR donor fragments are reported in Supplemental Table S1. HEK293T cells were cotransfected with 1 µg of CRISPR/Cas9-sgRNA plasmid and 400 ng of linear donor PCR product using Lipofectamine 3000 in a six-well plate. Cells were expanded into 10-cm plates 2 d after transfection. The knock-in cells were selected by treating with 200 µg/mL hygromycin B 3 d after transfection. Individual clones were selected and confirmed by Western blotting and Sanger sequencing of PCR amplicons. The sgRNAs from Supplemental Table S2 were used for targeting the 3′ end (C terminus of the protein) of the indicated genes.

29We transfected plasmids expressing NLS-ARF16-PB1 fused with eGFP at the C terminus or the eGFP-NLS-ARF16-PB1K944A, D994A, D998A mutant into each of the following: HEK293T, HEK293T-TIR1, and HEK293T-TIR1 cells in which either TEAD4 or ZNF143 were AID-tagged. Cells were expanded for 1 wk and GFP-sorted iteratively (three times) until we obtained a stable population of GFP-expressing cells.

30HEK293T or HEK293T-TIR1 cells were cotransfected with mCherry-NLS-AID and eGFP-ARF16-PB1 or eGFP-ARF16-PB1 mutant plasmids. The mutant eGFP-ARF16-PB1 plasmid was cotransfected at a concentration three times higher than the wild type to get comparable expression of the mutant protein. The coimmunoprecipitation data from Figure 4 were generated by separately transfecting mCherry-AID into HEK293T, HEK293T-eGFP-ARF16-PB1, and HEK293T-TIR1-eGFP-ARF16-PB1 cells. Cells were lysed 24 h after transfection in a buffer containing 50 mM Tris (pH 7.5), 150 mM NaCl, 0.5% NP-40, 1 mM EDTA, and protease and phosphatase inhibitors for 30 min on ice and then sonicated 30 sec on and 30 sec off for 15 cycles. Lysate was clarified by centrifugation at 12,000g for 15 min in 4°C. Anti-GFP antibody-conjugated magnetic beads (Chromotek, gtma-10) or anti-mCherry-conjugated affinity gel (Biolegend, 689502) were blocked with 1% BSA for 15 min. The beads were incubated with clarified lysate for 1.5 h at 4°C. The immunoprecipitate was recovered with DynaMag racks or by centrifuging at 3000g for 1 min. Beads were washed three times for 5 min in the lysis buffer, and 60 µL of 2× Laemmli buffer was added directly to the beads. The complex was heat-denatured for 5 min at 95°C. We used the following antibodies for the Western blots: anti-GFP (gift from Daniel Foltz, Northwestern University), anti-mCherry (rabbit, 1:5000; Abcam, ab183628), ZNF143 (1:5000; H00007702-MO1, Abnova), TEAD4 (1:1000; Santa Cruz Biotechnology, sc-101184), p53 (1:1000; Santa Cruz Biotechnology, DO1), anti-TIR1 (1: 10,000; gift from Masato Kanemaki, Osaka University), β-Actin (1:5000; Sigma, A1978), β-Tubulin (AA2; gift from Todd Stukenberg, University of Virginia), and CENP-I (rabbit; gift from Todd Stukenberg, University of Virginia). Regression lines for kinetic data were fit using GraphPad Prism.

31Cells were treated with 10 µM MG132 for 4.5 h to test whether AID-tagged proteins were degraded through the proteasome pathway. A stock of 50 mM auxin was diluted to a final concentration of 500 µM auxin in the culture medium. The 50 mM stock was solubilized in DMSO for the 3-h auxin treatment PRO-seq experiments using the progenitor line. For all other experiments, 50 mM auxin was solubilized in water. A degradation rate upon auxin treatment was measured by treating cells with 10 µg/mL cycloheximide and 500 µM auxin and collecting samples at every 15 min for 4 h.

32Cell permeabilization was performed as described previously. Cells were collected in 10 mL of ice-cold PBS after trypsinization and then collected and washed in 5 mL of buffer W (10 mM Tris-HCl at pH 7.5, 10 mM KCl, 150 mM sucrose, 5 mM MgCl2, 0.5 mM CaCl2, 0.5 mM DTT, 0.004 U/mL SUPERaseIN RNase inhibitor [Invitrogen], protease inhibitors [cOmplete, Roche]). The washed cells were then permeabilized with buffer P (10 mM Tris-HCl at pH 7.5, 10 mM KCl, 250 mM sucrose, 5 mM MgCl2, 1 mM EGTA, 0.05% Tween-20, 0.1% NP-40, 0.5 mM DTT, 0.004 U/mL SUPERaseIN RNase inhibitor [Invitrogen], protease inhibitors [cOmplete, Roche]) for 3 min. Cells were washed again with 10 mL of buffer W before transferring into 1.5-mL tubes using wide-bore pipette tips. Finally, cells were resuspended in 500 µL of buffer F (50 mM Tris-HCl at pH 8, 5 mM MgCl2, 0.1 mM EDTA, 50% glycerol, 0.5 mM DTT). After counting the nuclei, we generated 50-µL aliquots with ∼3 × 105 to 5 × 105 cells that were snap-frozen in liquid nitrogen and stored at −80°C. All centrifugations were done at 500g for 10-mL conical tubes and 2000g for 1.5-mL tubes at 4°C, and all buffers were maintained on ice. PRO-seq libraries were prepared as described previously with the following modifications. The libraries were amplified by PCR for a total of 10 cycles. We performed 5′ decapping using RppH, 5′ hydroxyl repair, 5′ adapter ligation, and reverse transcription, while the 3′ RNA biotin moiety was bound to magnetic streptavidin beads. We added an 8-base random unique molecular identifier (UMI) to the 5′ end of the adapter that was ligated to the 3′ end of the nascent RNA. We did not perform any size selection because we were willing to tolerate excessive adapter/adapter ligation products to ensure that our libraries were not biased against short nascent RNA insertions.

33We removed adapters from the paired end 1 or single-end reads using CutAdapt. Each 3′ adapter harbored an 8-base UMI. We removed PCR duplicates based on the UMIs using fqdedup (https://github.com/guertinlab/fqdedup), trimmed UMIs with fastx_trimmer (https://github.com/agordon/fastx_toolkit), and implemented fastx_reverse_complement to generate the reverse complement sequence (https://github.com/agordon/fastx_toolkit). We aligned reads to hg38 with Bowtie2, sorted aligned BAM files using SAMtools, and used seqOutBias to generate bigWig files. We used the bigWig R package (https://github.com/andrelmartins/bigWig) and University of California at Santa Cruz (UCSC) Genome Browser utilities to query bigWig files within genomic coordinates. Bedtools was used to parse genomic coordinate files and query for overlapping regions. Differential nascent transcript abundance was measured by DESeq2. Bidirectional transcription was identified using dREG. MEME was used for de novo motif discovery within dREG-identified regulatory elements that change upon ZNF143 depletion; TOMTOM matched the ZNF143 motif to a database that is curated by HOMER. All of the analysis details and codes are available at https://github.com/mjg54/znf143_pro_seq_analysis. Raw sequencing files and processed bigWig files are available from Gene Expression Omnibus accession record GSE126919.

34The dynamics for the concentrations or densities of RNA polymerases at pausing regions and gene bodies, defined as p and b, are described as follows: and where kinit is the rate of transcription initiation, kpre is the rate constant for premature paused Pol II release, krel is the rate constant for the release of a paused Pol II as transcription proceeds into productive elongation in the gene body, and kterm is the rate constant for transcription termination. The term lp/lb is a ratio of the relative DNA segment lengths that is applied to adjust the gene body concentration based on the larger amount of DNA in the gene body as compared with the pause region. The term kinit implicitly accounts for the product of the unbound Pol II concentration and the rate constant for initiation. Concentration is considered as the number of RNA polymerases per length of DNA. Thus, this concentration is referred to as a density. We consider this model as though the units are dimensionless for the analysis of how specific rates influence the relative Pol II quantities at pause sites and within the gene bodies. The steady-state levels (pss,bss) were found by setting dp/dt = db/dt = 0. Because understanding the effects of the relative pause region length was not our focus, we set r = lp/lb for the parameter sensitivity analyses: and The pause index is the relative density of reads in the pause region compared with the gene body (Pi = pss/bss). The pause index is only dependent on kterm and krel, the rate constants for the termination of transcription and release into productive gene body elongation:

35To examine the effects of each parameter on the pause and body concentrations, we considered high and low parameter values k(hi) and k(lo) and computed the changes in pss and bss. The following notation documents the change in pss when k is changed from a relatively high to a relatively low value: Δp(k) = pss[k(hi)] − pss[k(lo)]. First, we evaluated the effects of the transcription initiation rate (kinit) on the steady state pause region and gene body Pol II concentrations: and Note that k(hi)> k(lo) by definition, so that Δp(kinit) > 0 and Δb(kinit) > 0. Therefore, increasing the rate of transcription initiation will result in both pause region and gene body increases. We present results of single-parameter changes that can also be considered using a standard sensitivity analysis. For this example, the sensitivities were computed as follows: and In general, Δp(k) ∼ (δp/δk)[k(hi)− k(lo)] for small changes in k, and sign[Δp(k)] = sign(δp/δk). Next, we considered the effects of varying the rate constant for premature pause release (kpre): and These results show that increasing the rate constant for premature pause release will decrease both pause region and gene body Pol II concentrations [Δp(kpre) < 0 and Δb(kpre) < 0]. We next demonstrated that an increase in the rate constant for the release of Pol II into the gene body (krel) decreases pss and increases bss: and It is interesting to note that if there is no premature pause release (i.e., kpre = 0), the model predicts that a change in the rate of pause release into transcriptional elongation will not affect the concentration of RNA polymerases in the gene body if all other factors are identical [i.e., Δb(krel) = 0 for kpre = 0]. Finally, we evaluated the effect of modifying the rate constant for the termination of transcription: and These results show that an increase in the rate constant for transcriptional termination does not affect the steady-state level of paused Pol II but will decrease the gene body concentration of RNA polymerases [Δp(kterm) = 0 and Δb(kterm) < 0]. The results from our parameter sensitivity analyses are illustrated in Table 1, where the entries for pss and bss indicate the effect of increasing each parameter.

36Our preceding analyses show that consistent changes in the pause region and gene body (i.e., both either increase or decrease) are observed only following changes in the rate of transcriptional initiation or the rate constant for premature pause release (kinit,krel). Our experimental data show that repressed genes show decreases in both the pause region and the gene body region. These decreases in Pol II density were greater in the pause region in comparison with the gene body. Next, we documented the conditions under which the effects of varying the rates of initiation and premature release of the paused Pol II are greater for the pause region as compared with the gene body [e.g., Δp(kinit) > Δb(kinit)]. Changes in transcriptional initiation are set as:

37This condition is satisfied for lb/lp > krel/kterm (recall r = lp/lb). In general, lb/lp is a large value because the pause region (<100 base pairs [bp]) is much smaller than the gene body (approximately >10 kb). So, krel/kterm < lb/lp ∼ 100 must be obtained. For changes in premature release, the condition is as follows:

38This relation leads to the same constraint observed for kinit:krel/kterm < lb/lp. Recall that the pause index is defined as Pi = kterm/rkrel = (lb/lp)(kterm/krel). Therefore, Pi > 1 for the same condition that constrains Δp(kpre) > Δb(kpre) for changes in kinit and kpre: krel/kterm < lb/lp This demonstrates that for an arbitrary change in either the initiation rate or the premature pause release rate constant, the effect of the change in the pause region will be greater than the change in the gene body region whenever the pause region Pol II concentration is greater than that at the gene body. Furthermore, the effect ratio is identical for changes in kinit and krel and is equal to the value of the pause index: Δp/Δb = Pi.

39We aimed to generate plots in which steady-state levels of pause region and gene body concentration were imposed upon the peak and flat regions of a profile that is characterized by an exponential approach toward the peak followed by an exponential decay toward a stable plateau. We used a sum of exponential functions for the waveform: where bp is the independent variable, and τ is the exponential decay constant. The parameter pk is set to the root of the derivative as shown below so that pkpause determines the peak of the waveform. The parameter pkbody is set such that the asymptotic gene body region decays to a desired level. The derivative of this waveform is and the root of the derivative is given by the value of bp at the peak of the waveform [max(density)]: For bp ≫ τ, the gene body level asymptotically approaches pkbodypkpause/pk. We implicitly determined a value for pkbody that will give a gene body level of a desired level. We selected a value of pkbody that will give a plateau pkbodypkpause/pk of choice for a given setting of the pause peak pkpause. After implicitly finding a value for pkbody that produces the gene body level of choice, the waveform is produced.