Improved tool performance by guide RNA designTo improve the utility of the SNAP-ADAR approach, we revisited the guide RNA (gRNA) design, starting from the prior design12 that comprises a guide RNA of 22 nt (19 nt antisense + 3 nt non-binding loop), chemically stabilized by 2′-O-methyl ribose modification outside the central base triplet, and which is equipped with a 5′-terminal, single O6-benzylguanine (BG) moiety for covalent SNAP-ADAR recruitment12 (Fig. 1a). In an initial screen, we identified three key guide RNA properties to improve editing: (1) increasing the length of the guide RNA to ca. 25 nt (22 nt antisense + 3 nt non-binding loop), (2) the inclusion of up to four locked nucleic acid (LNA) building blocks, and (3) the application of a bivalent linker for recruitment of two SNAP-ADAR proteins per guide RNA (BisBG), see Fig. 1a. We have shown the notable effects of these measures in direct comparison to the prior design (BG 22 nt) on the editing of the phospho-tyrosine site 701 on endogenous STAT1 in engineered 293 HEK cells expressing SNAP-ADAR1Q (Flp-In T-REx 293 – SA1Q), see Fig. 1b. While the prior design required the transfection of 20 pmol/96-well for its maximal editing yield of 80%12, the same was already achieved with 1 pmol/96-well with the improved design. The highest potency was achieved with the combination of a specific terminal LNA pattern, the BisBG recruiting moiety, and the extension of the guideRNA length to 25 nt, which gave high editing efficiency (65%) with only 0.1 pmol/96-well and detectable editing even at 0.01 pmol/96-well. The Flp-In T-REx 293 – SA1Q cell line allows to fine-tune the SA1Q expression level by the duration of doxycycline induction12. In contrast to the prior design, where at least 5 pmol guide RNA (per 96-well) were required to induce moderate editing yield at full SA1Q induction (48 h doxycycline, Fig. 1c), the best design achieved considerably higher editing yields (60-80%) at one tenth of the dose (0.5 pmol/well) and under weaker SA1Q induction (down to 4 h doxycycline). We used Western blot to monitor the amount and formation of guideRNA- mediated covalent SA1Q dimers (Fig. 1d) and found that 0.5 pmol/96-well guide RNA engaged almost all available SA1Q for conjugation upon weak SA1Q induction (8 h doxycycline). Furthermore, the application of the BisBG moiety helped to boost editing of notoriously difficult to edit codons like 5′-GAN (N = G, A, U, C)12 in the 22 nt guide RNA design (Fig. 1e). Finally, the improved guide RNA design enabled better usage of the wildtype SNAP-ADAR enzymes, SA1 and SA2, which are more precise regarding global off-target effects, but which have been roughly tenfold less potent in combination with the prior guide RNA design compared to their hyperactive analogs SA1Q and SA2Q12. Notably, the improved guide RNA design allows to obtain good editing yields (up to 70%) also with the wildtype SA1 tool when applying guide RNA amounts (1 pmol/96-well), that gave no detectable editing with the prior guide RNA design (Fig. 1f), highlighting the benefits of the improved guide RNA design.Fig. 1: Improved guide RNA design and characterization in engineered 293 Flp-In T-REx cells expressing one of the four SNAP-ADAR effectors (see key).a The improved guide RNA (gRNA) design is made longer with four LNA s at specific positions (light pink) and can recruit two SNAP-ADAR proteins (BisBG linker in place of mono BG linker as in the prior design). The chemically unmodified nucleotides in the central base triplet are indicated in green, where C denotes the cytidine opposite the targeted adenosine. b Editing of 5′ -UAU site in the endogenous STAT1 transcript (Y701) comparing different amounts of different guide RNA designs tested (see key) in Flp-In T-REx 293 – SA1Q cells induced with 10 ng/ml of Doxycycline for 48 h. c Editing potency of improved guide RNA designs at ten times lower amount compared to old design under varied expression levels of the SA1Q effector by varying doxycycline (Dox) induction times as indicated. d Western blot showing unconjugated SNAP-ADAR (SNAP-ADAR monomer) and BisBG-guide RNA-mediated, intracellular covalent dimerization of SA1Q (see band shift, SNAP-ADAR dimer) at low (8 h Dox) and high (48 h Dox) expression levels of the editing enzyme in Flp-In T-Rex 293 – SA1Q cells; Biological replicate blots of high SA1Q expression levels are available in the Supplementary Fig. 1. e Editing of 5′-GAN codons in the ORF of endogenous GAPDH transcripts (targets used in Vogel et al.12) with 5 pmol of short 22 nt guide RNAs with linkers that could recruit either one (BG) or two (BisBG) SA1Q effectors in Flp-In T-Rex 293 – SA1Q cells (48 h Dox). f the same as panel b, but using wildtype SA1-expessing Flp-In T-REx 293 cells. Data in panel b, c, and f is shown as the mean ± s.d. of N = 3 independent experiments, and in panel e, is shown as the mean ± s.d. of N = 2 (for most cases and N = 3 in some cases) individual data points are shown as dots. All guide RNA amounts denote pmol/96-well with 150 µl total volume. Source data and full western blots are provided as a Source Data file.The SNAP-ADAR system is highly versatile in cell culture applicationsDue to the high transfection and knockdown efficiencies, chemically synthesized siRNAs still dominate most cell culture applications over genetically encoded shRNAs. Given the small size of the SNAP-ADAR guide RNAs, their transfection into challenging cells including primary cells might be a strong advantage over genetically encoded editing approaches which may require viral delivery of the guide RNA component. To extend the scope and improve the versatility of the SNAP-ADAR approach, we demonstrate the delivery of the SNAP-ADAR transgene by various means into various cell types and provide proof of successful editing of the endogenous STAT1 transcript by subsequent guide RNA transfection. Specifically, we applied the PiggyBac transposase system37 and lentiviral delivery38,39 for stable integration, and adenoviral delivery40 for transient expression in HeLa and A549 cells (Fig. 2a), of which the latter cell line is hard to transfect with plasmids. We were able to easily express the transgene in HeLa and A549 cells and to induce maximal editing yields (between 50% and 90%) on the regulatory STAT1 Y701 phospho site with very good guide RNA potency using the improved guide RNA design (Fig. 2b, c). Since PiggyBac integration is particularly simple and efficient, we studied the potency of the improved guide RNA design more deeply in A549 and HeLa cells stably expressing either wildtype SA1 or hyperactive SA1Q (Fig. 2d, e). In HeLa cells, very high editing yields ≥80% were obtained with SA1Q and guide RNA amounts ≥0.5 pmol (per 96-well). The HeLa cell line expressing the wildtype SA1 effector achieved slightly reduced editing levels (ca. ≥60%) and required slightly more guide RNA (≥ 1 pmol/96-well) for this. In A549 cells, similar trends were observed, however, with slightly reduced maximal editing yield and guide RNA potency. Interestingly, we found a certain reduction in editing yield for the SA1 effector in A549 cells at high guide RNA amounts. We speculate that the intracellular guide RNA levels at very high doses might use up the entire SNAP-ADAR protein which could reduce the frequency of recruiting two SNAP-ADAR effectors per guide RNA. We used a BG-FITC labeling assay to determine the amount of free, unconjugated SNAP-ADAR protein depending on the guide RNA dose applied (Fig. 2f). Indeed, we found that all effector was used up when ≥5 pmol/96-well BisBG-guide RNAs were applied.Fig. 2: Efficiency of the improved SNAP-ADAR tool in immortalized cell lines and human primary cells.a Characteristics of the transgene expression systems used to express SNAP-ADAR effector in different cell types. b, c Comparative editing of 5′ -UAU site in the endogenous STAT1 transcript (Tyr701) with different amounts of BisBG – 25nt+LNA guide RNA (BisBG 180) in HeLa (b) and A549 (c) cells expressing SA1Q or SA1 by different delivery options. d, e Dose-dependent editing of the STAT1 transcript (Y701 site) with BisBG 180 in HeLa-PB-SA1Q/SA1 (d) and A549-PB-SA1Q/SA1 (e) cells expressing SA1Q or SA1 after full induction (48 h) with 1 µg/ml doxycycline. f BG-FITC-protein/gRNA conjugation assay to determine unconjugated SA1Q levels in Hela-PB-SA1Q/A549-PB-SA1Q cells induced with Dox for 48 h and transfected with different amounts of gRNA BisBG 180 for 24 h. For biological replicates (N = 2) and more details, see Supplementary Fig. 2. g Time-dependent editing yield after single (T1), double (T2, 48 h post T1) and triple transfections (T3, 48 h post T2) with 2 pmol of BisBG guide RNA (BisBG 180). h Editing in human primary cells, after adenoviral delivery of SA1Q (25-75 MOI), with different guide RNA amounts (0.2–5 pmol of BisBG 180). MOI multiplicity of infection, RPE retinal pigment epithelium cells, NHA normal human astrocytes. All guide RNA amounts denote pmol/96-well with 150 µl total volume. Data in panel b-e, g and h is shown as the mean ± s.d. of N = 3 independent experiments, individual data points are shown as dots. Source data and full western blots are provided as a Source Data file.Depending on the turnover rate of a given protein, a recoding editing event might take some time to fully establish a phenotype. Thus, we tested the time course of the editing yield with the improved guide RNA design in SA1Q PiggyBac HeLa cells (HeLa-PB-SA1Q). In quickly dividing HeLa cells, the editing yield started to drop after three days (Fig. 2g). To generate a window large enough to establish an editing phenotype at the protein level, we developed a protocol with a second transfection 48 h after the first transfection. This double transfection protocol enables to characterize the edited protein on days 4-7 after the first transfection and should satisfy many applications. If required, a third transfection would also be possible (Fig. 2g).Finally, we explored the editing of the endogenous STAT1 transcript in human primary cells by adenoviral delivery of SA1Q (25 to 75 MOI) and transfection of the guide RNA (0.2 – 5 pmol/96-well), see Fig. 2h. In retinal pigment epithelium cells (RPE), moderate editing levels (50-60%) were already achieved at low guide RNA dose (0.2 pmol/96-well). In normal human astrocytes (NHA), the editing levels were generally lower, and higher titer of adenovirus helped to increase the editing yield. Overall, the data highlights the versatility of the SNAP-ADAR tool. Following simple protocols, high editing levels are regularly achieved with low-dose guide RNA transfections. The effector can either be stably expressed in immortalized cell lines by PiggyBac integration, but also transiently and readily in primary cells using adenovirus even without FACS sorting.PTM interference: using RNA base editing to perturb protein functionEven though RNA base editing is currently limited to A-to-I (and C-to-U) changes, there are various applications conceivable beyond the repair of disease-causing G-to-A (and T-to-C) point mutations. RNA editing has the potential to modulate native protein function6,19,20. This becomes particularly clear when looking into the scope of amino acid changes, which include the removal of (regulatory) phospho-tyrosine (Y > C), -serine (S > G), and -threonine (T > A), as well as the removal of (regulatory) lysine residues (K > R, or K > E). The (reversible) posttranslational modification (PTM) of proteins is a hallmark of regulating signaling cues, metabolism, transcription, epigenetics, protein degradation, and many other processes6,19,20. RNA base editing could be employed to interfere with posttranslational regulation of protein function in a highly rational and programmable manner. In analogy to RNA interference (RNAi)41, we call this broader concept PTM interference (PTMi). Applying RNA base editing for PTMi might be an attractive way to study basic biology and to create clinically desirable phenotypes (Fig. 3a). To get an idea about the scope of RNA base editing for PTMi, we set-up an unbiased screen of >70 different PTM sites on various endogenous signaling transcripts (Fig. 3b). For all >70 sites, we transfected guide RNAs of the improved standard design into HeLa-PB-SA1Q cells, which comprised of a 22 nt antisense part (5′−9-C-12 nt, with C = the cytidine opposite the target adenosine), plus a 3 nt 5′-overhang, including 4 LNA bases, and a 5′-terminal BisBG moiety for recruiting two SA1Q effectors per guide RNA. With the improved standard design, we observed editing yields ranging from 0% to 90%, with roughly 40% of the sites being edited with yields ≥50%, highlighting that most of the codons relevant for PTMi are well editable in principle. To better understand the factors that affect editing yields, we analyzed the data further. There is a well-known codon preference for the editing of any 5′-NAN codon (N = A, U, G, C), particularly preferring U > A > C > G for the 5′-neighboring nucleotide. Indeed, we obtained the highest editing yields for 5′-UAG, 5′-UAU and 5′-UAC codons, medium editing levels for 5′-AAN, while 5′-CAN and 5′-GAN (N = A, U, G, C) were clearly more difficult to edit. Interestingly, for each specific codon we found examples with high editing yield but also with rather low editing yield (Fig. 3c). This large spreading in editing efficiency indicates that further key factors play important roles, which could be RNA secondary structure (target or guide RNA), the target gene expression level, or the half-life of the target transcript. To assess structural determinants, we plotted all editing yields against the G/C-content of the guide RNA/target mRNA substrate duplex (Fig. 3d) and against the free energy of guide RNA hairpin folding (Fig. 3e). Indeed, editing yields were higher for guide RNAs with less G/C-content and with little propensity for hairpin folding. We then plotted the editing yields against the expression levels of the target genes (Fig. 3f), using the TPM (Transcript per million) values of the target transcripts determined in the HeLa-PB-SA1Q cell line. We also plotted the editing yields against the average half-lives of the transcripts (Fig. 3g), which have been determined by others in HeLa cells before42. Neither the expression level nor the target half-life seems to have a major influence on the editing outcome. Thus, key parameters regarding editing efficiency are the codon preference, the target structure and the guide RNA hairpin folding propensity.Fig. 3: Unbiased PTM interference screen with RNA base editing.a Various signaling cues related to essential biological processes represent attractive targets for doseable and transient manipulation by RNA base editing. b 70 + PTM sites (Y > C, K > R, S > G, T > A, etc.) on various endogenous signaling transcripts have been targeted in HeLa-PB-SA1Q cells using 2 pmol (per 96-well with 150 µl total volume) of the improved standard guide RNA design (5′−3 + 9 + C + 12 nt, 4 LNAs, BisBG) or variation thereof as indicated. c–g Analysis of effects of various parameters on RNA base editing efficiency. Mean editing yields (N = 2) of 70+ targets plotted against target codon (c), GC content of the guide RNA calculated with Oligo Calc44 (d), minimum free energy of guide RNA hairpin folding secondary structure calculated with NUPACK45 (e), relative gene expression level (mean TPM values of genes in Hela PB SA1Q cells with 48 h dox induced SA1Q expression (f). Half-life of the transcript, taken from42. In panels c-g, the color code indicates the nucleotide 5′ to the edited adenosine. Data in panel b is shown as the mean of N = 2 independent experiments in most cases and N = 3 or 4 in some cases, individual data points are given. In panel c, the box represents the interquartile range showing the middle 50% of the data points. The ends of upper or lower T-shaped whiskers extend to the maximum or minimum data point which is still within 1.5 times the interquartile range. Source data are provided as a Source Data file.Conveniently, these key parameters can be improved by means of optimizing guide RNA sequence and chemistry. While Tyr>Cys (5′-UAY) and Lys>Arg (5′-AAR) editing gave already satisfying yields, other PTMi targets including Ser>Gly and Thr>Ala editing often suffer from low yields, for example, when 5′-CAN codons were addressed (Fig. 3c). From a recent structural analysis of ADAR2 binding to a dsRNA substrate it was discovered that a clash of the exocyclic amino group of the guanine base with the backbone of glycine residue 489 is responsible for loss of editing efficiency at these un-preferred codons43. Thus, we aimed for reducing the space demand of the 5′G:C base pair in the minor grove by applying the nucleoside inosine in the central base triplet for pairing the cytosine base 5′ to the targeted adenosine in 5′-CAN codons, to exemplify: change the 5′-NCG by a 5′-NCI sequence in the guide RNA. We tested the concept on nine endogenous targets, covering all four 5′-CAN codons (N = A, U, G, C). In all nine cases, a deoxyinosine placed at the respective site gave improved editing yields (Fig. 4a). For all four codons, editing yields could be boosted up to twofold. Thus, we suggest to regularly apply the deoxyinosine substitution when targeting 5′-CAN codons. Editing yields of the respective improved guide RNA designs were included into Fig. 3b, with a different color (orange).Fig. 4: Optimization of sequence and chemistry of guide RNAs for effiecient PTMi.Comparison of different guide designs for improved editing of different signaling targets in HeLa-PB-SA1Q cells using 2 pmol (per 96-well with 150 µl total volume) of the respective BisBG-gRNA (BB-gRNA). For all gRNA sequences and modifications, see target list in the Supplementary Data 1. a Placement of deoxyinosine opposite the orphan cytidine in 5′-CAN (N = A, U, G, C) target codons did clearly foster editing efficiency, as shown for nine examples covering all four 5′-CAN codons. b LNA nucleotides were essential to achieve high editing efficiency in a structured mRNA substrate (S166 in MDM2); arrows indicate the target adenosine in the structured MDM2 mRNA. c Relaxing guide RNA hairpin structure by sequence optimization was essential for editing Y705 > C in STAT3. d Placement of a 2′-F modification opposite the 3′-terminal adenosine base was required to suppress bystander editing in the adenosine-rich 5′-AAA codon. In this particular target, LNA nucleotides were only well accepted at the 3′-terminal half of the guide RNA. e Multiplexing RNA base editing by co-transfection of two guide RNAs against up to three different sites on the JAK2 transcript (Y570, Y1007, Y1008). Two adjacent sites, e.g., Y1007/Y1008, can be targeted by two individual guide RNAs without compromising editing efficiency. Data in panel a–c and e is shown as the mean of N = 2 independent experiments, and in panel d, is shown as the mean of N = 2 (in most cases and in some cases N = 4), individual data points are given. Source data are provided as a Source Data file.The dataset from Fig. 3b clearly demonstrates large effects of guide RNA and/or target mRNA structure on editing. Repeatedly, we observed largely varying editing yields when targeting different tyrosine phosphorylation sites (5′-UAY codons, Y = U, C) or different lysine modification sites (5′-AAR codons, R = A, G) on the same transcript side-by-side. On the SRC transcript, we found editing yields of >70% (Y419C) versus 23% (Y530C), on JAK2, we observed 50% (Y1007C, Y1008C) versus 20% (Y570C), and on CDK9 we detected 40% (K48R) versus 13% (K44R). The most dramatic effect we observed was with the E3 ligase MDM2. With a prior guide RNA design (22 nt, no LNA, but with BisBG), we found no detectable editing for the S166G site, whereas the S188G target site – only 66 nt downstream – gave editing with up to 60% yield. Both sites are highly editable 5′-UAG codons. By applying the mfold tool46, we identified a strong local secondary structure at S166, but not at S188 (Fig. 4b). In contrast, both guide RNAs did not contain noteworthy secondary structures. Thus, the different degrees of secondary structure of the two target sites seemed to cause the different editing yields. Importantly, both sites were well edited (to ca. 80%) when we switched to the improved design with 25 nt and four LNA bases, indicating that the increased length and the additional binding power provided by the LNA nucleotides help to make the guide RNA invade into this structured target site.When we analyzed the editing of the STAT3 transcript, we were wondering why we achieved excellent editing on two lysine PTM sites (ca. 75% for K140R and K685R), but why the more editable 5′-UAC codon of the important phospho-tyrosine Y705 showed almost no editing (ca. 5%). When we analyzed the secondary structure of the guide RNA with the NUPACK tool45, we found that this guide RNA folds into a very stable tetra loop hairpin with a stem of eight base pairs, which engaged three of the four LNA nucleotides (Fig. 4c). To break the structure, we adapted the guide RNA sequence. Specifically, we shortened the antisense part of the guide RNA by one nucleotide from the 5′ end and included a different 3 nt 5′-overhang that did not engage into secondary structure formation. This 24 nt guide RNA (5′−3nt + 8-C-12 nt, with four LNAs) achieved editing levels up to 70%, similar to the phosphor-tyrosine sites of other STAT members.Bystander editing ─ the editing within the guide RNA / mRNA duplex ─ is a severe engineering challenge in competing approaches that apply genetically encoded guide RNAs8,47,48,49. In contrast, the SNAP-ADAR tool blocks bystander editing outside the central base triplet by global 2′-O-methylation of the guide RNA11. However, in highly adenosine-rich codons, bystander editing of the nearest neighbor can occur12. In the PTMi screen bystander editing occasionally happened during K > R editing at the 3′ nearest neighbor in 5′-AAA(G) codons, in particular when a guanosine followed that 3′ adenosine. A typical example from the screen is the removal of the regulatory mono-ubiquitination site K644R in MALT1, where we initially obtained an on-target editing yield of 20% and a bystander editing of 8% (Fig. 4d). By incorporating a 2′-F modification opposite the bystander position, the off-target yield was reduced to below detection (<4%). Interestingly, we found that the K644R site did not benefit from the LNA modification. In fact, a guide RNA containing the 2′-F modification but not the four LNA nt achieved editing levels up to 58% with very little bystander editing (<5%). According to the NUPACK tool, the guide RNA did not fold into a very strong secondary structure. We tested if LNAs at one of the termini were better accepted than on the other and found that two LNAs at the 3′-terminus were very well accepted in contrast to two LNAs on the 5′-terminus. We speculate that LNA nucleotides can sometime also negatively interfere with editing, this might be more relevant for moderate-to-edit codons like 5′-AAA. In general, a single 2′-F nucleotide is generally recommended opposite the 3′-terminal adenosine in 5′-AAA codons.Finally, we tested the best strategy to concurrently edit two adjacent sites on the same transcript. For this, we co-transfected two out of three guide RNAs into Hela-PB-SA1Q cells, which target three distinct tyrosine residues (Y580, Y1007, Y1008) on the JAK2 transcript. In principle, one would expect the guide RNAs targeting Y1007 and Y1008 to mutually compete for target engagement. Nevertheless, the editing yields were similar for Y1007/Y1008 guide RNA co-transfection compared to other combinations, e.g., Y1007/Y570 or Y1008/Y570, where the guide RNAs are not expected to mutually compete (Fig. 4e). This suggests that guide RNAs act on the target RNA rather in a hit-and-run fashion than staying at the target for a long time. We tested this hypothesis by trying to inhibit an editing reaction by co-transfection of a BisBG-guideRNA with an excess of an editing-incompetent NH2-guide RNA of that same sequence. In agreement with our model, even a tenfold excess of NH2-guide RNA was hardly able to reduce the editing yield (Supplementary Fig. 3).Perturbation of the JAK/STAT signaling pathway by PTMiTo demonstrate the potential of PTMi to perturb signaling cues, we tested the manipulation of Interferon-α (IFN-α, type I IFN) or Interferon-γ (IFN-γ, type II IFN) induced response, which is mediated via the JAK/STAT pathway25,26,27. The canonical IFN-γ signaling results in the activation of the STAT1 transcription factor by phosphorylation of tyrosine 701, homodimerization, and nuclear translocation, which finally leads to the expression of ISGs carrying the GAS promoter sequence 29 (Fig. 5a). IFN-α, on the other hand, leads to formation of the ISGF3 complex containing pSTAT1, pSTAT2 and IRF9, which leads to the expression of ISGs carrying the ISRE promoter element28. Various functionally important PTM sites, in particular phosphorylation and acetylation sites, have been described for all members of the JAK/STAT pathway starting from the IFN receptors down to the transcription factors50,51,52. Figure 5a illustrates various PTM sites color-coded for the estimated effect of PTMi on ISG expression, being either activating (green) or inactivating (coral red). We looked particularly deeply into two well-known STAT1 mutations, which have been found in patients to be either a dominant LOF34(Y701 > C) or a dominant GOF35 (T288 > A). PTMi experiments were carried out in HeLa-PB-SA1Q cells. Such cells respond well to IFN-α or -γ treatment, resulting in pY701 STAT1 levels which are clearly detectable by Western blot (Fig. 5b, Supplementary Fig. 4). As expected, the response to IFN-γ gave higher pSTAT1 levels than IFN-α. We then studied the effect of introducing the respective LOF or GOF mutation by RNA base editing on the IFN response. When cells were transfected two times (T2), 72 h and 24 h prior to IFN treatment, with a guide RNA that introduces the LOF mutation Y701 > C (BB180, 80% editing yield), and lysed 24 h after IFN treatment, we found a clear reduction in pSTAT1 levels by Western blotting (0.5-fold of the unedited control for IFN-γ). In contrast, when cells were transfected with a guide RNA that introduces the GOF mutation T288 > A (BB478, 80%), we found a shallow increase in pSTAT1 levels (1.2-fold of the unedited control for IFN-γ), see Fig. 5b, Supplementary Fig. 4. Binding of the guide RNA or the guide RNA-SNAP-ADAR conjugate could potentially affect STAT1 levels negatively. However, neither an amino guide RNA control (incompetent to recruit SNAP-ADAR) nor an editing-incompetent but conjugation-competent control guide RNA negatively affected pSTAT1 levels, see Supplementary Fig. 4. We next characterized the effects of PTMi on the IFN-γ response by RT-qPCR and Immunofluorescence imaging. First, we measured the relative expression level of the well-known GAS-driven ISGs, CXCL9 and IRF1. Indeed, both CXCL9 and IRF1 were strongly activated (ca. 9.000-fold and ca. 40-fold, respectively) by IFN-γ in the control sample lacking PTMi. However, after introducing the dominant LOF mutation Y701 > C via PTMi, the expression of both CXCL9 and IRF1 were strongly damped (ca. 10-fold against the unedited IFN-γ control). On the other hand, introducing the dominant GOF mutation T288 > A resulted in a moderate increase of CXCL9 expression (1.5-fold against the unedited IFN-γ control) and a mild increase of IRF1 expression (1.2-fold against the unedited IFN-γ control), see Fig. 5c. This indicates that PTMi via RNA base editing allows to manipulate the JAK/STAT signaling pathway up or down depending on the selected PTM site. Then, we further characterized the signaling event by immunofluorescence imaging against total STAT1. Prior to IFN-γ treatment, total STAT1 was mainly residing in the cytoplasm. Quickly after adding the cytokine (30 min), STAT1 almost entirely localized to the nucleoplasm. Around 8 h after IFN-γ treatment, STAT1 repopulated the cytoplasm (Fig. 5d, Supplementary Fig. 5). We repeated the experiment but introduced the respective LOF (Y701C) or GOF (T288A) mutation by transfecting the respective guide RNA (BB180 or BB478), 72 h and 24 h prior to IFN-γ addition and monitored the localization of total STAT1. As expected, the Y701C LOF mutation created a STAT1 variant that was impaired to move into the nucleus, in accordance with the strong damping of ISG expression. In contrast, the GOF T288A mutation created a STAT1 variant that went well into the nucleus, but which stayed nuclear for prolonged time compared to the non-transfected IFN-γ-treated control, indicating a mechanism for increased IFN-γ signaling. Both results are well in accordance with the literature34,35,53.Fig. 5: Modulation of the interferon α- and γ-response by PTMi.a Scheme of the JAK/STAT pathway, in response to IFN-α/γ, initiating a phosphorylation cascade resulting in transcription factor activation, translocation and induction of an ISG response, figure adapted from59. Highlighted are specific PTMi target sites colored for the expected effect on the signaling outcome in coral red (downregulation) or green (upregulation), respectively. All experiments were performed in HeLa-PB-SA1Q cells, transfecting 2 pmol of the respective guide RNA 72 h and 24 h prior to IFN (2000 U/ml) stimulation for 24 h. b Western blot detection of STAT1 activation (via Y701 phosphorylation) in response to either IFN-α or -γ, following on the introduction of either a GOF (T288 > A) or LOF mutation (Y701 > C) via RNA base editing. For full blots in duplicate, see Supplementary Fig. 4. c Analysis of the effect of PTMi (STAT1, LOF/GOF) on cytokine expression (CXCL9 and IRF1) in response to IFN-γ stimulation by RT-qPCR, blue bars show expression levels in control with IFN-γ normalized to ctrl w/o IFN-γ; red bars show expression levels in samples with PTMi and IFN- γ normalized to control with IFN-γ. d Analysis of subcellular localization of total STAT1 by Immunofluorescence imaging in response to IFN-γ treatment in control cells (IFN-γ for 0 h, 30 min, 2 h, 4 h, and 8 h) and in cells after PTMi of STAT1 (LOF/GOF). Scale bars represent 25 µm. Images shown are zoomed-in, cropped and merged channels of STAT1 (amber) and nucleus (blue). For full and split channel images, see Supplementary Fig. 5. For statistical analysis of colocalization, see Supplementary Fig. 6. For detailed protocol, see Methods section. e Transcriptome-wide expression analysis (DESeq) on perturbing ISG expression in response to IFN-α or –γ treatment, via PTMi at five different JAK/STAT targets. In the control experiment (empty transfection), 379 ISGs were identified, which passed a significance threshold (p < 0.001) and a ≥ 1 log2-fold change in gene expression and were plotted for increasing induction in response to a 24 h IFN-γ stimulation. Data in panel c is shown as the mean ± s.d. of N = 2 independent experiments, individual data points are given. Source data are provided as a Source Data file.To get a broader picture of PTMi on the interferon-induced JAK/STAT pathway, we compared the editing of six different PTM sites at five different players of the pathway by double-transfection (T1 – 24 h after Doxycycline induction, T2 – 48 h post T1) of the respective guide RNA into HeLa-PB-SA1Q cells and studied global gene expression changes 24 h after IFN-α or -γ addition via next generation sequencing (DESeq analysis54). Specifically, we individually introduced the following PTM mutations: Y701 > C in STAT134 (BB180, ca. 60% editing yield), T288 > A in STAT135 (BB478, ca. 80%), Y690 > C in STAT228 (BB349, ca. 62%), Y1034 > C in JAK155 (BB355, ca. 70%), Y466 > C in IFNAR156 (BB305, ca. 82%), Y457 > C in IFNGR157 (BB304, ca. 82%). Empty transfected cells with or without IFN treatment served as negative controls. By Western blot, we showed for all targets that the binding of the guide RNA does not negatively affect target protein expression (Supplementary Fig. 7). Poly(A) + RNA was collected 24 h after interferon treatment and global changes in transcriptome expression were compared to an unedited control without interferon treatment. All experiments were carried out in duplicates. The DESeq2 pipeline58 identified roughly 11,000 genes with TPM ≥ 2 and assigned padj values to each transcript. From the control experiments (no PTMi) with versus without 24 h IFN-γ treatment, we selected 379 highly significantly (padj <0.001) differentially expressed ISGs ( ≥ 1 log2 unit gene expression change), see Excel Sheet for heatmap in the Source Data. These ISGs were plotted in a heatmap format sorted for increasing response of the unedited control to IFN-γ stimulation, see Fig. 5e, heatmap at the left. Typical ISGs, like MX1 or IFI27, were strongly upregulated ( > 7 log2 units) in response to IFN-γ treatment. Notably, when introducing the Y701 > C LOF mutation in STAT1 preceding IFN-γ treatment, the ISG response was broadly and strongly inhibited, almost back to levels of control cells untreated with interferon. The editing of STAT2 or JAK1, in contrast, had comparably subtle effects on ISG expression, which is in accordance with their limited extent of contribution for activation of GAF complex. However, subtle differences in ISG gene expression were found indicating that unique ISG expression patterns are accessible by PTMi in these players. Interestingly, the editing of the IFN-γ receptor (IFNGR1) had particularly little effect on ISG expression, even though the editing level was very high (ca. 80%). This indicates that the selected mutation was not dominant, and that the remaining 20% unedited receptor might be sufficient to drive the full ISG response. Importantly, the editing of the GOF mutation T288 > A in STAT1 leads to a completely different, highly complex modulation of ISG expression, including the activation of numerous genes, including important cytokines like IL6 and CCL2 by ≥1.8 log2 units. Clearly, PTMi enables the damping, the modulation, but also activation of the IFN-γ signaling pathway depending on the PTM site selected. Upon IFN-α treatment, the changes in ISG expression in the unedited control where less strong compared to IFN-γ treatment (Fig. 5e, heatmap at the right). The removal of the pY701 site in STAT1 had again a damping effect on ISG expression. Interestingly, this was also true for the analog site (Y690 > C) in STAT2, highlighting a clear difference between the IFN-α- and IFN-γ-driven ISG response. The dependency of ISG expression on STAT2 phosporylation in response to IFN-α stimulation, however, seems to mirror very well the requirement for both pSTAT1 and pSTAT2 in the ISGF3 complex. The editing of the other three targets had comparably little effects on ISG expression. Even though the editing yield of the IFN-α receptor (IFNAR1) was high (80%), the chosen LOF mutation might not have been dominant enough to result in a clear damping of the pathway. In accordance with the literature, the GOF mutation T288 > A in STAT1 had almost no activating effect on IFN-α-dependent ISG activation35.Overall, the data highlights that PTM interference – as induced by RNA editing – can manipulate signaling pathways in a unique, activating and/or inactivating manner. It also shows that differences in the pathways, like dependency of IFN-α but not IFN-γ on phosphorylation of STAT2, are reflected in the perturbation of the ISG signaling outcome, indicating that RNA editing can be used as a tool to interrogate the regulatory role of specific PTM sites in complex signaling networks. Importantly, all PTM sites were studied in the same cell line that had the SNAP-ADAR effector stably integrated simply by transfection with the different guide RNAs. The approach avoids the cumbersome creation of genetically modified cell lines for each PTM site in each signaling molecule and it also avoids artefacts from the overexpression of cDNAs of signaling molecules carrying specific PTM mutations.
