Characterization of cooperative PS-oligo activation of human TLR9

Abstract

Single-stranded phosphorothioate oligonucleotides (PS-oligos) can activate TLR9, leading to an innate immune response. This can occur with PS-oligos containing unmethylated CpG sites, the canonical motif, or PS-oligos that do not contain those motifs (non-CpG). Structural evidence shows that TLR9 contains two PS-oligo binding sites, and recent data suggest that synergistic cooperative activation of TLR9 can be achieved by adding two separate PS-oligos to cells, each engaging with a separate site on TLR9 to enhance TLR9 activation as a pair. Here, we demonstrate and characterize this cooperativity phenomenon using PS-oligos in human cell lines, and we introduce several novel PS-oligo pairs (CpG and non-CpG pairs) that show cooperative activation. Indeed, we find that cooperative PS-oligos likely bind at different sites on TLR9. Interestingly, we find that PS-oligos that generate little TLR9 activation on their own can prime TLR9 to be activated by other PS-oligos. Finally, we determine that previous models of TLR9 activation cannot be used to fully explain data from systems using human TLR9 and PS-oligos. Overall, we reveal new details of TLR9 activation, but we also find that more work needs to be done to determine where certain PS-oligos are binding to TLR9.

Graphical abstract

Pollak and colleagues characterize novel mechanisms of PS-ASO-mediated TLR9 activation. They show that two separate PS-ASOs can cooperatively activate TLR9 by each oligo binding to a separate site on TLR9. They also find that current models of TLR9 activity do not apply to PS-ASO activation of human TLR9 in particular.

Introduction

Cellular defense against invading pathogens relies on innate immune responses that are initiated by activation of pattern recognition receptors, which include Toll-like receptors (TLRs).^1,2 These receptors can also be activated inadvertently by nucleic acid-based therapeutics, which can significantly hamper drug development.^3,4,5,6 TLR7/8/9 are a subfamily of structurally related endosomal receptors that respond to specific single-stranded nucleic acid sequences.⁷ TLR7 and TLR8 respond to RNA, and TLR9 responds to DNA.^8,9 While phosphorothioate oligonucleotide (PS-oligo) sequences containing unmethylated CpG (CpG) motifs are considered TLR9 agonists, various CpG-containing PS-oligos can produce a wide range of TLR9 activation levels depending on the amount of the CpG site(s) on the PS-oligo.^10,11 How these PS-oligos interact with and can differentially activate human TLR9 remains incompletely understood.

Structural analysis of TLR9 showed base-specific contacts between TLR9 and the CpG sequence in a 12-mer oligo (hereafter this TLR9 binding site is called the “CpG site”) (Figure 1).¹² Confoundingly, this PS-oligo shows extremely weak activation of TLR9 in comparison with other, longer PS-oligos containing multiple CpG sites. Further structural and functional assays determined that an additional PS-oligo binding site (hereafter called the “second site”) exists on TLR9, distinct from the CpG binding region.¹³ This led to the hypothesis that PS-oligos must bind to both binding sites on TLR9 to achieve full activation of TLR9, which has been deemed cooperative activation. This hypothesis was also inspired by data showing that TLR7 and TLR8 also require oligos binding at two distinct regions for their activation. Furthermore, these two oligo binding regions are in similar locations in TLR7/8/9.⁷

Figure 1.

Previously published models describing TLR9 activation

One TLR9 protomer is shown in blue and another in orange. Each protomer has identical interactions with oligos. (A) PS-oligos with a single CpG motif can only interact with the “CpG site,” causing weak TLR9 activation. (B) PS-oligos with a 5′-xC motif and a CpG motif can cause augmented TLR9 activation by engaging with the “second site” on TLR9, referred to here as “model 1” of cooperative activation. This may occur with either four (left) or two oligos (right). (C) Two separate PS-oligos, each binding to a different TLR9 site, can also achieve augmented TLR9 activation, referred to here as “model 2” of cooperative activation.

One model proposed to explain this TLR9 two-site cooperative activation hypothesis posits that a single PS-oligo can contain properties that allow it to bind to both sites on TLR9 to induce high levels of activation (model 1; Figure 1).^10,13 These properties include a CpG site and a 5′-xC motif. Accordingly, weakly activating CpG PS-oligos (without the 5′-xC motif) only engage with the CpG site. However, functional studies have yet to demonstrate this mechanism, making it uncertain whether model 1 is correct.

Another possible model posits that two separate PS-oligos can cooperatively activate TLR9 by each PS-oligo binding to one of the two different sites (model 2). One study of cooperativity supporting model 2 was done using PS-oligos and mouse cell lines,¹³ but the difference between mouse and human TLR9 activation has been well documented,¹⁰ making it uncertain whether this mechanism can occur with human TLR9. Another study¹⁴ used human cells but showed that only phosphodiester (PO)-linked oligos (which undergo degradation) can generate cooperative TLR9 activity after 18-h incubation in cells. PS-linked versions of the oligos generated cooperative activation (more than 2-fold over baseline) but only using greater than 80 μM concentrations for 18 h,¹⁴ conditions that likely cause a variety of cellular events not involving TLR9. Furthermore, there was an 800:1 ratio between the PS-oligos used to generate cooperative activation, unlike the ∼1:1 ratio seen with the mouse system¹³ and shown here, making it unclear how this would work mechanistically. A final model of TLR9 activation (however, not⁴ specifically defined as cooperative) posits that a single oligo can bind simultaneously to both sites (at a 1:1 ratio of TLR9 to oligo), where the spacing between the motifs on the oligo is critical.¹⁰ Overall, we find that there is no clear evidence of cooperative activation of human TLR9 by CpG PS-oligos (for either model 1 or 2) or, therefore, a comprehensive mechanistic understanding of human TLR9 activation by PS-oligos.

We have demonstrated previously that gapmer PS-ASOs (antisense oligonucleotides) without CpG sites (non-CpG) can activate TLR9,^4,15,16,17 which occurs rarely and in a sequence-specific manner. While some non-CpG PS-ASOs that activate TLR9 contain T-rich sequences,^18,19 others lack such motifs.⁴ Furthermore, CpG motifs at the 5′ end of PS-oligos appear to contribute most to their TLR9 signaling.¹¹ Overall, however, a comprehensive model to predict TLR9 signaling capacity from PS-ASO sequence or backbone chemistry currently does not exist. Indeed, our recent study systematically exploring the roles of sequence, 2′ modifications, and backbone chemistries for ASO-dependent TLR9 signaling, including the role of gap size, did not reveal any novel patterns.¹⁷ Gapmer PS-ASOs are chemically modified oligos that can effectively reduce target RNA levels via an RNaseH1-dependent mechanism and are used to treat several diseases.^20,21 However, preclinical studies have revealed that some PS-ASOs can cause undesired effects,²² such as TLR9 activation.⁵ Where and how certain non-CpG PS-ASO interact with and activate TLR9 remains poorly understood. A further confounding factor is that TLR9 binding affinity does not correlate with TLR9 activation for CpG and non-CpG sequences.^4,23 A better understanding of these processes can help streamline PS-ASO drug development so that TLR9-activating PS-ASOs can be discarded from preclinical development.

Interestingly, we recently discovered that adding two particular non CpG PS-ASOs together to cells causes significant cooperative TLR9 activation, reminiscent of model 2.⁴ Here, we sought to better understand this phenomenon and to develop a better overall understanding of human TLR9 activation. First, we further characterize the non-CpG-based PS-ASO cooperative activation shown previously. Next, we show that these cooperative PS-ASOs likely bind to different regions of TLR9, as anticipated. We then explore the extent to which other PS-oligos can either act cooperatively or competitively with one another, finding a variety of examples of both cases. To contextualize our results, we tested current models of CpG-based TLR9 activation (Figure 1) and found that these models are not sufficient to explain data using PS-oligos and human TLR9. Finally, we determined a novel set of CpG PS-oligos that cooperatively activate human TLR9. Overall, we unravel new insights into CpG- and non-CpG-dependent activation of human TLR9, but we find that more work needs to be done to fully understand human TLR9 activation by PS-oligos.

Results

Cooperative activity of non-CpG PS-ASOs

We have shown previously that combining equimolar amounts of either PS-ASO 05 (05) or PS-ASO 18¹⁸ with PS-ASO 95 (95) (all three are 3-10-3 cEt gapmers) generated significant enhancement of TLR9 signaling⁴ in comparison with each of the PS-ASOs individually, suggesting cooperative activation for these two pairs of PS-ASOs. This work was done in Bjab cells, which have been established as an accurate predictor of patient PS-ASO innate immune responses¹⁶ (via readout of CCL22 mRNA expression) and are therefore used for preclinical mechanistic studies. These three oligos have been determined previously to require TLR9 for their innate immune signaling,^4,16 and all have been shown to produce the same pattern of inflammatory responses,⁴ including CCL22 mRNA induction, suggesting that it is unlikely that other innate immune receptors are playing a role in our studies. Furthermore, we note that neither 18 nor 95 contain a 5′ xCx motif, in contrast to the cooperativity pairs reported previously.¹³

Here, we systematically characterize this cooperativity phenomenon by varying the dose of each PS-ASO within the pairs (Figure 2A) to determine the extent of cooperativity in Bjab cells. For example, combining 0.8 μM 18 with 1.6 μM 95 causes TLR9 activation roughly 13-fold over the activation seen with 18 or 95 by themselves, more than shown previously⁴ (Figure 2A). Overall, the optimal concentration for maximal TLR9 activation (under these conditions) appears to be ∼0.8 μM for 05 and 18 and ∼2.0 μM for 95 (Figure 2).

Figure 2.

Cooperative TLR9 activation via non-CpG PS-ASOs

(A–C) Sequences of 3-10-3 cEt PS-ASOs used, with blue indicating 2′ cEt modifications and black indicating DNA. Shown are relative qRT-PCR levels of CCL22 mRNA from Bjab cells following 2.5-h incubation with the indicated PS-ASOs. Data below the horizontal lines indicate co-incubated PS-oligos. The relative EC₅₀ values and maximum values (Max) were determined via GraphPad Prism and are present under the curves. Averages of three or more experimental replicates are shown.

Next, we investigated the maximum activity and the EC₅₀ (half maximal effective concentration) values for each PS-ASO on its own or with its cooperative pair. The EC₅₀ value for 95 was significantly lower when combined with either 05 or 18 than when administered as a single PS-ASO (by roughly 9-fold and 4-fold, respectively), signifying cooperative activation (Figure 2B). Interestingly, however, 95 showed roughly similar maximal activity with or without addition of 05 or 18. One explanation for this is that 95 can engage with the two sites, but it may have a lower affinity for one of the sites. On the other hand, the maximal activity of 05 or 18 on its own, even at high concentrations, is much lower than when 95 is added (Figure 2C). This suggests that 05 and 18 cannot engage with both sites to achieve maximal TLR9 activation, unlike 95. This is consistent with prior observations showing 95 to be a stronger TLR9 agonist than 05 or 18,⁴ which also need a longer time to produce TLR9 activation. Overall, we find that adding two PS-ASOs enhances the EC₅₀ of TLR9 activation over adding a single PS-ASO, and previous models suggest that this is due to the ability of both PS-ASOs to collectively interact with both TLR9 binding sites.¹³ These data corroborate model 2 (Figure 1) for non-CpG PS-ASOs.

Cooperative PS-ASOs occupy different TLR9 binding sites

Cooperative TLR9 activation is thought to require PS-oligos to bind to different sites on TLR9. Therefore, we sought to determine whether 05 and 95 bind to different regions of TLR9 using the NanoBRET binding assay.²⁴ First, we incubated Alexa fluorophore-labeled 05 (Alexa-05) with luciferase-tagged TLR9 and titrated in unlabeled 05. This resulted in a decrease of the BRET ratio, showing that the two (labeled and unlabeled) PS-ASOs compete to bind to TLR9, as obviously expected (Figure 3A). On the other hand, 95 showed limited ability to compete with Alexa-05 for TLR9 occupancy in this assay. This result suggests that 05 and 95 primarily bind to different regions of TLR9. In the reciprocal experiment, Alexa-95 was much more effectively displaced by 95 in comparison to 05, reinforcing the notion that these PS-ASOs primarily bind to different regions of TLR9 (Figure 3B). We note that 05 shows some ability to compete with Alexa-95 for TLR9 binding, but this occurs only at high concentrations. This binding may not be involved in 05’s ability to activate TLR9 and instead may be a non-specific, non-productive binding event.

Figure 3.

Differential TLR9 binding sites of cooperative PS-ASOs

(A) NanoBRET assay for binding, measuring the BRET ratio of TLR9 and Alexa-05 with addition of the indicated unlabeled PS-ASOs at the indicated concentrations. (B) Same as (A) but with Alexa-95 incubated with TLR9. (C) NanoBRET assay for binding, measuring the BRET ratio of TLR9 with addition of the indicated Alexa-PS-ASOs at the indicated concentrations. (D) Relative qRT-PCR levels of CCL22 mRNA from Bjab cells following a total of 2.5 h incubation with 1.5 μM of the indicated PS-ASOs added at the indicted times. Averages of three or more experimental replicates are shown. (E) NanoBRET assay for binding, measuring the BRET ratio of PC4 and Alexa-95 with addition of the indicated unlabeled PS-ASOs at the indicated concentrations. (F) NanoBRET assay for binding, measuring the BRET ratio of TLR9 and Alexa-789 with addition of the indicated unlabeled PS-ASOs at the indicated concentrations. The relative binding kds (M) were determined via GraphPad Prism and are present under the curves.

Next, a direct binding analysis was performed, comparing Alexa-05 and Alexa-95 binding with TLR9. We found that the amplitude was larger for 05 than 95 (Figure 3C). Because the amplitude of the BRET signal varies according to an inverse sixth power of the distance between the PS-ASO fluorophore to the protein luciferase, one interpretation of this result is that Alexa-05 binds to a site closer to the luciferase (the amino terminus of TLR9) than is the case for Alexa-95, providing additional evidence that the two PS-ASOs primarily bind to different sites on TLR9, further corroborating model 2. We note that the NanoBRET binding assay suggests that there are multiple PS-ASO binding sites on TLR9, and other sites may exist beyond the two previously determined,¹³ which may represent a limitation of this method. Further studies will be required to specifically define and characterize each of the binding sites.

Next, we tested whether the order of addition of the PS-ASOs was consequential for cooperative activation to see whether the binding of one PS-ASO was required for the other PS-ASO to bind, as reported previously.¹³ To do this, we added one PS-ASO, waited for 0.5 h, then added the other PS-ASO of the cooperative pair (and vice versa) (Figure 3D). The data show that the order of PS-ASO addition resulted in essentially no difference in TLR9 activation. This suggests that the PS-ASO binding events may not have significant effects on one another, consistent with the binding data (Figures 3A and 3B). Interestingly, this is in line with one study¹⁴ of TLR9 cooperative activation but not consistent with another study.¹³ The latter study explicitly showed that one PS-oligo needed to be bound to TLR9 for another PS-oligo to bind at the second site; if this were the case, then we would anticipate that pre-incubating one PS-ASO would result in more activation than pre-incubating the other PS-ASO, which did not occur.

As a control, we repeated the binding assay shown in Figure 3B but using a different protein, PC4, which has a single, well-defined PS-ASO binding pocket²⁵ (Figure 3E). Here, 05 successfully competed for binding with Alexa-95, in contrast to the result shown in Figure 3B. This shows the TLR9-specific nature of 05’s inability to bind to the location where Alexa-95 is bound. In another control, we incubated a larger, 20-mer Alexa-labeled PS-ASO (Alexa-789) that does not generate TLR9 signaling and may bind non-specifically to several regions of TLR9 (Figure 3F). Here, 05 and 95 can compete for binding of Alexa-789. This shows that Alexa-05 and Alexa-95 likely engage primarily in site-specific binding, unlike Alexa-789.

PS-ASO pairs can lead to cooperative activation or competitive inhibition of TLR9

We next sought to determine whether other non CpG 3-10-3 cEt PS-ASO pairs can induce cooperative TLR9 activation. We first tested whether any of 14 TLR9-activating PS-ASOs studied previously¹⁷ generated cooperative activation in combination with 05, 18, or 95 (Figure S1). No significant cooperative activation was determined, particularly in comparison with the examples shown in Figure 2, which are also plotted for reference.

We next chose to investigate how the TLR9 response from 05, 18, or 95 can be altered with addition of PS-ASOs that show minimal TLR9 activation on their own. We found a variety of outcomes, including various levels of cooperativity and competition (Figures 4A and 4B). Most interestingly, we found that 148 (the PS-ASO numbers here were shortened to the last three digits) was cooperative (green circles) with 95 but competitive (red circles) with 05 or 18. In turn, PS-ASOs that were cooperative with either 05 or 18 (131 and 031, respectively) were competitive with 95. This further suggests that 95 has different TLR9 binding properties than 05 and 18, which explains these different functional consequences. In addition, the data show that 05 and 18 likely have slightly different TLR9 binding properties, which can be explained by their slightly different activity profiles⁴ (see above). Finally, these data suggest that 148 and 131 bind to different TLR9 locations even though neither of them generates meaningful TLR9 activation. This suggests that even non-activating PS-oligos can adopt a variety of TLR9 binding modes and that their interactions with TLR9 can prime TLR9 for cooperative activation to occur when another PS-oligo binds to TLR9. This is reminiscent of the activities of 05 and 18 shown in Figure 2.

Figure 4.

PS-ASOs can display cooperative activation or competitive inhibition of TLR9

(A) Green and red circles indicate cooperative activation and competitive inhibition, respectively. Shown are relative qRT-PCR levels of CCL22 mRNA from Bjab cells following 2.5-h incubation for 95 and 4-h incubation for 05 and 18 with 1.5 μM of the indicated PS-ASOs. Data below the horizontal lines indicate co-incubated PS-ASOs. (B) Same as (A), but triangles represent ascending doses of 0.5 and 1 μM. (C) NanoBRET assay for binding, measuring the BRET ratio of TLR9 and Alexa-95 (left) and Alexa-05 (right) with addition of the indicated unlabeled PS-ASOs at the indicated concentrations. (D) NanoBRET assay for binding, measuring the BRET ratio of TLR9 with addition of the indicated Alexa-PS-ASOs at the indicated concentrations. Unlabeled 148 and 131 were incubated at set concentrations. (E) Relative qRT-PCR levels of CCL22 mRNA from Bjab cells following 2.5 h incubation with 1.5 μM of the indicated PS-ASOs (CpG 11 and CpG 5). Data below the horizontal lines indicate co-incubated PS-ASOs. Triangles represent ascending doses of 0.5 and 1 μM. Averages of three or more experimental replicates are shown.

To further confirm that different PS-ASOs can bind to different sites on TLR9 for these cooperative pairs, we utilized a similar NanoBRET-based binding strategy as above. Here, 131 competed with the Alexa-95 PS-ASO for TLR9 binding much better than the cooperative 148, as expected given that 131 blocks 95’s TLR9 activation (Figures 4C and 4D). The reciprocal experiment was performed with Alexa-05. In this case, 148 competed with Alexa-05 better than 131, which is again consistent with the functional analysis. The same binding patterns were observed using a different experimental setup, one where the Alexa PS-ASO was titrated in while 148 or 131 concentrations was kept constant.

Comparison of single- and double-CpG PS-oligos

We next sought to contextualize our observations of non-CpG PS-ASO TLR9 cooperativity by comparing the effects of non-CpG oligos with the effects of previously studied CpG-based oligos. To determine whether CpG PS-oligos and non-CpG PS-oligos share the same binding sites in TLR9, we asked whether 148, 031 and 131 could reduce TLR9 activation by CpG PS-oligos (here, we refer to PS-oligos containing unmethylated CpG sites as “CpG” with various arbitrary numbers corresponding to different sequences, while non-CpG oligos only have numbers associated with them; Table S1). We first tested a CpG PS-oligo (CpG 11) that contains a CpG site and 5′-xC motif, which should bind to both sites (based on model 1).¹³ Indeed, all three PS-ASOs competed with the CpG PS-oligo (Figure 4E). This indicates that our non-CpG PS-ASOs are likely binding to the same two sites as the CpG PS-oligo.

We next tested a CpG PS-oligo (CpG 5) that should only interact with the CpG site (Figure 4E). Unexpectedly, all three PS-ASOs were competitive with this CpG PS-oligo as well, suggesting that CpG 5 engages both TLR9 sites, in contrast to model 1. Given that no functional studies were done to confirm the model predicting that PS-oligos with a single CpG site do not engage with the second site,¹³ we assessed the activation levels of PS-oligos for wild-type TLR9 and TLR9 with mutations that were shown to affect binding.

Previous structural and binding studies showed that W96A disrupts binding at the CpG site while F375A and Y535A are involved in binding at the second site. Based on model 1, PS-oligos with a single CpG site should only be affected by the W96A mutation, while PS-oligos with a CpG site and a 5′-xCx motif (hereafter called a “double-CpG” motif) should be affected by all three mutants (Figure S2A). We expressed these mutations and wild-type (WT) TLR9 in HEK293 cells engineered with a nuclear factor κB (NF-κB) reporter system for these studies. We chose this system because it was referenced as evidence of confirmation of model 1.^11,13 Single-CpG PS-oligos generated less TLR9 signaling than double-CpG ones, as expected (Figure S2B).¹⁰ Interestingly, we found that single- and double-CpG PS-oligos were affected by all three mutants in essentially the same pattern. These data suggest that single-CpG PS-oligos require the second site for activity, in contrast to model 1.

We next compared the EC₅₀ and maximum activation of single- and double-CpG PS-oligos to further characterize their differences (Figure S2C). For two sets of size-matched PS-oligos, we found that the second CpG site caused a roughly 2-fold difference in maximum activation but produced a similar EC₅₀ value as PS-oligos with a single site. Next, we found that increasing the PS-oligo size caused a decrease in EC₅₀ but resulted in no change in maximum activation, which was found for series of single- and double-CpG PS-oligos (Figure S2C). These data build on studies showing that PS-oligos with single CpG sites show less activity than those with double CpG sites,¹⁰ and here we show that one important difference between these PS-oligos (single vs. double CpG) is their extent of maximum activation of TLR9. While it is unclear what accounts for this difference, mutations to TLR9 that show different binding properties for these oligos¹³ do not appear to reflect function effects.

Finally, we tested our non-CpG cooperative pairs (05 or 18 with 95) using these TLR9 mutants (Figure S3). We found that these mutations also decreased TLR9 activation for non-CpG PS-ASO pairs, suggesting that the CpG and non-CpG PS-oligos utilize similar binding sites for cooperative TLR9 activation, corroborating data shown in Figure 4E.

Overall, it appears that model 1 does not explain activation of human TLR9, and further studies are necessary to understand why, for example, double-CpG-containing PS-oligos have higher maximum levels of TLR9 activation than single-CpG-containing PS-oligos.

Cooperative TLR9 activation via CpG PS-oligos

Next, we sought to recapitulate previously published observations of CpG-based cooperativity to better understand human TLR9 activation overall and to contextualize our non-CpG-based cooperative activation results. Using sequence information from published studies^13,14 but changing the systems so that PS-oligos and human TLR9 were used, we were unable to observe any indication of cooperative TLR9 activation (Figure S4), even using longer time points (data not shown). This was shown in Bjab cells, our standard TLR9 activation assay, as well as in HEK293-TLR9 cells, which are commonly used by other groups studying TLR9 activation. This shows that mouse and human cells and TLR9 proteins themselves differ (despite high sequence conservation) and urges caution when extrapolating conclusions from one species to another. In addition, the functional differences between PS and PO oligos are highlighted.

In the course of our studies, we serendipitously discovered examples of cooperative activation using CpG PS-oligos in human cells. We found that PS-oligos CpG 8 and 57 added together generated significant cooperative activation (Figure 5A). To study this further, first we altered both PS-oligos to define the characteristics of these two PS-oligos that promote cooperative activation. Shortening 57 by removing nucleotides from either the 5′ or 3′ end still showed cooperative activity, but it appears that the sequence at the 3′ end of 57 contributes more to cooperativity than the 5′ end (Figure S5A). Next, we tested the importance of the sequence of 57. To do this, we generated a series of PS-oligos in which we mutated guanosines (Gs) to adenosines (As) in 57. Changing all Gs to As resulted in a drop in TLR9 activity, but cooperative activation was still achieved. Mutating select Gs to As resulted in minor changes to cooperativity in comparison with the parent PS-oligo, 57. Similar results were seen when changing the sequence in 57-8, which is a shorter version of 57 (Figure S5B). Overall, it appears that many different sequences may be able to generate cooperative activation of TLR9.

Figure 5.

CpG oligo cooperativity

(A–C) Relative qRT-PCR levels of CCL22 mRNA from Bjab cells following 2.5-h incubation with the indicated PS-ASOs. Data below the horizontal lines indicate co-incubated PS-oligos. The relative EC₅₀ values and maximum values (Max) were determined via GraphPad Prism and are present under the curves. (D) Right: NanoBRET assay for binding, measuring the BRET ratio of TLR9 with addition of the indicated Alexa-PS-ASOs at the indicated concentrations. Unlabeled CpG 8 and CpG 13 were incubated at set concentrations of 200 nM. Averages of three or more experimental replicates are shown.

We next altered the size and sequence of the CpG 8 PS-oligo to determine how its cooperativity changed (Figure S6). Interestingly, we found that, for CpG PS-oligos larger than 16 nt, 57 was competitive. On the other hand, for PS-oligos less than 16 nt, with either one or two CpG sites, 57 was cooperative. Next, we tested 6-57 (which is shorter than 57) with the same panel of CpG PS-oligos and determined the same results as with 57. Therefore, the size of the CpG PS-oligo is a critical determinant of cooperativity, unlike for 57 (or its derivatives). Overall, it appears that the sizes of the PS-oligos are important factors in cooperative activation of TLR9, and further studies will be necessary to understand the rules for the sequence dependence of this phenomenon.

Next, we determined how the EC₅₀ values and maximum activity changed comparing each PS-oligo (CpG 8, 57, and 6-57) with or without its cooperative pair. We found that CpG 8 by itself or with its cooperative pairs reached a similar maximum plateau; however, the EC₅₀ value of CpG 8 was lower with the cooperative pairs (Figure 5B). 57 and 6-57 reached a much lower maximum as single agents in comparison with their cooperative pairs, but the EC₅₀ values showed little change (Figure 5C). These data appear to be similar to the results shown in Figure 2, where one PS-oligo (95 or CpG 8) likely can interact with both TLR9 sites, but the other PS-oligo partner (05 or 57) shows limited activation capacity by itself and acts to prime the other PS-oligo.

Finally, we sought to determine whether 6–57 and CpG 8 bind to different regions of TLR9, using a similar NanoBRET approach as above (Figure 5D). Indeed, we found that 6-57 binding was unaffected by CpG 8, but CpG 13 (whose activity is blocked by 6-57; Figure S6) indeed displaced 6-57, again consistent with the functional results. Overall, we discovered several different examples of CpG PS-oligo-based cooperative activation of human TLR9 (using reasonable PS-oligo concentrations), which, to our knowledge, is the first example of this phenomenon and further supports model 2.

Comparison between cooperative oligo pairs and further characterization of cooperativity

Different classes of canonical CpG oligos can produce different types of downstream TLR9 responses and can lead to differing TLR9 activation levels.^26,27 We sought to compare the TLR9 signaling from the cooperative oligos described here with oligos from two classes of well-studied canonical CpG oligos, types C²⁸ and B.²⁹ Type C CpG oligos contain palindromes that confer secondary structural features that are important for their ability to activate TLR9(11), unlike CpG-B oligos. First, we found that the CpG and the non-CpG cooperative oligos (concentrations were used that reflect saturating levels shown in previous figures) generate a similar extent of activation, suggesting that each pair interacts with and activates TLR9 in a similar manner, likely by binding to the same set of sites (Figure S7A). This level of activation is similar to that achieved by the CpG-C oligo (2,395²⁸) although the cooperative oligos were less potent. The CpG-B oligo (PS 2,006²⁹) was more potent (as shown previously¹¹) and achieved a higher level of activation than the other oligo types. This suggests that this oligo is able to engage with different (additional) sites on TLR9 than the other oligos or that it can cause additional TLR9 conformational changes that lead to higher activation (Figure S7A).

Next, the role of ASO backbone chemistry in cooperative activation was assessed. PS backbones afford improvements in ASO therapeutic index in comparison with naturally occurring PO backbones;³⁰ however, PS backbones can cause higher TLR9 signaling than PO ones (although exceptions exist^17,31). Synthetic and naturally occurring PO-oligos can be digested by endosomal DNase II, which is required for TLR9 activation in some cases.³² Here, we found that replacement of the full PS backbone with POs resulted in significant decreases in TLR9 activation for CpG8, 57, or the pair added together (Figure S7B). Replacement of two PS bonds in the gap-wing junction with PO linkages for 05 and 95 has been shown previously to reduce TLR9 signaling.⁴ Here, we find that the cooperative activation is also diminished in comparison with the full PS version of the ASOs (Figure S7C). While it is unclear exactly why, PS backbones are important for activation of individual and cooperative pairs of the oligos studied here. This suggests that DNase II is not playing a role for these oligos.

Finally, we tested cooperative TLR9 activation in an alternative cell type to test whether these cooperativity observations extend beyond Bjab cells and TLR9-expressing HEK293 cells. Here, we used TLR9-overexpressing THP-1 cells. Oligo pairs 05 and 95 or 18 and 95 produced more of this TLR9 signaling in comparison with the oligos added individually (Figure S8A). Also, combining CpG 8 and 57 produced greater activation in comparison with those oligos added individually (Figure S8B). These results show that cooperative TLR9 activation can occur in multiple cell lines.

Discussion

Using human cells and PS-containing oligos, here we discovered and characterized several examples of PS-oligo pairs capable of cooperative TLR9 activation. In addition to PS-ASO pairs with non-CpG sequences, we identified several cooperative CpG PS-oligos. We corroborated the functional results with binding studies showing that cooperative PS-oligos indeed bind to primarily different sites on TLR9. Our work also shows that extrapolation of results from mouse cells to human cells may not be appropriate, and we provide alternative PS sequences that do show TLR9 cooperative activation in human cells.

We investigated the activities of the individual PS-oligos that were involved in TLR9 cooperative activation. For multiple examples, we found that one PS-oligo (i.e., 95 or CpG 8) was capable of generating the same maximal activation as found with the cooperative pair but only when administered at high concentrations (Figures 2 and 5). Interestingly, the second PS-oligo (i.e., 05 or 57) showed limited TLR9 activation on its own (Figures 2 and 5). This suggests a mechanism where a PS-oligo (i.e., 05 or 57) can interact with TLR9 and act to prime TLR9 to respond to a second PS-oligo (i.e., 95 or CpG 8). We speculate that both oligos binding together somehow cause a confirmational change to TLR9 that causes activation of downstream signaling. Given the extent of naturally occurring nucleic acids in the endosomes of cells,³² this mechanism may play a role in all types of TLR9 activation. Furthermore, this priming mechanism also likely occurs for TLR7 and TLR8, where oligoribonucleotides need to be bound at two sites for activation to occur.⁷

To show that PS-oligos engage with different binding sites in TLR9 to produce cooperative activation, we determined that PS-oligos showed limited or no ability to compete with their respective cooperative PS-oligo pair for TLR9 binding (Figures 3, 4, and 5). In one example, interestingly, we note that certain concentrations of 131 added to Alexa-05 caused an increase in BRET ratio relative to no addition of 131 (baseline) (Figure 4C). One explanation for this is that the binding of 131 improves the binding affinity of Alexa-05 for TLR9 or changes the conformation of Alexa-05 on TLR9. This phenomenon was not observed for other PS-oligo-TLR9 interactions seen in this study, suggesting that, in these other examples, the PS-oligos do not affect the binding properties of one another. These observations may explain why previous studies have shown that PS-oligos either can¹³ or cannot¹⁴ affect one another’s binding, and this will need to be investigated on a case-by-case basis in future cooperativity studies.

Taken together, our cooperativity data show strong evidence confirming model 2 to describe augmented activation of TLR9 via PS-oligo pairs, including non-CpG PS-ASOs. However, it is still unknown where each of the PS-oligos are binding to TLR9 for any of the cooperative pairs. Previous binding and structural studies formed the basis for model 1 to explain the difference in activity between single- and double-CpG PS-oligos.¹³ Here we show that PS-oligos with a single CpG site cannot generate the TLR9 activation levels of a double-CpG-site containing oligo, even at high concentrations (Figure 5). This expands previous observations that double CpG PS-oligos show greater activity than single CpG PS-oligos.¹⁰ The previous study¹³ relied on mutations to specific amino acids of TLR9 to propose model 1, which we used in our study here. However, when we performed the complementary functional studies to test the model, the model appeared to be inaccurate, at least for human TLR9 (Figure 5). Instead, we found that single- and double-CpG PS-oligos are impacted by the same TLR9 mutations, suggesting that both PS-oligo types (single- and double-CpG PS-oligos) engage with both TLR9 sites and that the PS-oligo sequence requirements for model 1 are therefore inaccurate.

Overall, why and how CpG and non-CpG PS-oligos can display a variety of maximal TLR9 activation levels remains an unanswered question. It is possible that generating alanine mutants of what are considered key residues for TLR9 binding (determined via structural studies) may not be the correct approach to solve this long-standing question. Answering this question may lead to determining to which TLR9 sites individual PS-oligos are binding during TLR9 activation.

In conclusion, we determined and characterized several novel examples of cooperative TLR9 activation in human cells using PS-oligos. We also challenge recently published models of TLR9 activation in the context of PS-oligos and human cells.

Materials and methods

Materials

Antibodies used were as follows: TLR9 (5845, Cell Signaling Technology) and GAPDH (32233, Santa Cruz Biotechnology). The sequences and chemistry of oligos used are listed in Table S1. Sequence Information about the oligos in Figure S1 could not be included but can be made available upon request. Oligos from Figure S1 called, for example, “S1 10” are referred to as “10” in Pollak et al.¹⁷

RNA preparation and qRT-PCR

Total RNA was prepared using an RNeasy Mini Kit (QIAGEN) from cells grown in 96-well plates using the manufacturer’s protocol. qRT-PCR was performed in triplicate using TaqMan primer probe sets as described previously.³³ Briefly, ∼50 ng total RNA in 5 μL water was mixed with 0.3 μL primer probe sets containing forward and reverse primers (10 μM of each) and fluorescently labeled probe (3 μM), 0.5 μL RT enzyme mix (QIAGEN), 4.2 μL RNase-free water, and 10 μL of 2× PCR buffer in a 20-μL reaction. Reverse transcription was performed at 48°C for 10 min, followed by 94°C for 10 min, and then 40 cycles of PCR were conducted at 94°C for 30 s and 60°C for 30 s within each cycle using the StepOne Plus RT-PCR system (Applied Biosystems). The results were analyzed by the relative quantity ddCt (dela-delta Ct) method. The mRNA levels were normalized to the amount of total RNA present in each reaction as determined for duplicate RNA samples using the Ribogreen assay (Life Technologies).

Western blot analyses

Cell pellets were lysed by incubation at 4°C for 30 min in RIPA buffer (50 mM Tris-HCl [pH 7.4], 1% Triton X-100, 150 mM NaCl, 0.5% sodium deoxycholate, and 0.5 mM ethylenediaminetetraacetate [EDTA]). Proteins were collected by centrifugation. Approximately 20–40 μg of protein was separated on 6%–12% NuPAGE BisTris gradient sodium dodecyl sulfate-polyacrylamide-gel electrophoresis gels (Life Technologies) and transferred onto polyvinylidene fluoride (PVDF) membranes using the iBLOT transfer system (Life Technologies). The membranes were blocked with 5% non-fat dry milk in 1× PBS at room temperature for 30 min. Membranes were then incubated with primary antibodies at room temperature for 2 h or at 4°C overnight. After three washes with 1× PBS, the membranes were incubated with species-appropriate horseradish peroxidase (HRP)-conjugated secondary antibodies (1:2,000) at room temperature for 1 h to develop the image using Immobilon Forte Western HRP Substrate (Millipore). Uncropped gels can be seen in Figure S9.

Cell culture

Bjab cells were maintained in Roswell Park Memorial Institute (RPMI) 1640 medium supplemented with 20% fetal bovine serum (FBS), 0.1 μg/mL streptomycin, and 100 U/mL penicillin. Prior to treatment, Bjab cells were washed in RPMI 1640 medium and placed in a V-bottom 96-well dish at 50,000 cells per well prior to the indicated PS-ASO addition by free uptake in serum-free RPMI 1640 medium. Following the indicated treatment time, cells were pelleted and lysed, total RNA was prepared, and levels of RNAs were quantified using qRT-PCR. RNA levels were normalized to 100 for UTC (untreated control).

HEK-Blue Null1 cells and THP1-Dual hTLR9 cells (InvivoGen) (THP1-TLR9) (InvivoGen) are stable commercial cell lines that expresses an optimized secreted embryonic alkaline phosphatase (SEAP) reporter gene in the HEK293 cells and a Lucia luciferase in the THP-1 cells. The SEAP reporter gene is controlled by an interferon β (IFN-β) promotor fused to five NF-κB and AP-1 binding sites. WT and mutant TLR9s (mutagenesis was performed using the Agilent QuickChange II XL Kit) were generated using the pLVX-Puro lentivirus vector (Addgene) and Lenti-XTM Packaging Single Shots (Takara) according to the manufacturer’s instructions.

HEK293-TLR9 cells and THP-1 TLR9 cells were grown according to the manufacturers’ instructions under standard conditions of 37°C, 5% CO₂.

NanoBRET binding assay

N-terminal NLuc fusions were created using the vector pFN31K NLuc CMV-neo (Promega). Briefly, coding sequences of proteins were amplified from plasmids from Origene (TLR9, catalog number RC207510; PC4, catalog number RC204999) using PCR primers complementary to the full-length cDNAs. The PCR products were ligated to XhoI and EcoRI sites of the pFN31K NLuc CMV-neo vector as described previously.²⁴ Fusion proteins were expressed by transfecting the plasmid into 6 × 10⁵ HEK293 cells using Effectene transfection reagent (QIAGEN) according to the manufacturer’s protocol. Following 24-h incubation, cells were removed from the plate by trypsinization, washed with 1× PBS, and resuspended in 250 μL Pierce immunoprecipitation (IP) lysis buffer (Thermo Scientific). Lysates were incubated 30 min at 4°C while rotating, and then debris was pelleted by centrifugation at 15, 000 rpm for 5 min. The fusion protein was purified by adding 20 μL HisPur Ni-NTA magnetic beads (Thermo Scientific) and 10 mM imidazole and then incubating at 4°C for 2 h. Beads were washed five times with 1× PBS, 10 mM imidazole, and 0.01% Tween 20. Fusion protein was eluted from the beads in 100 μL 1× PBS and 200 mM imidazole, followed by dilution with 200 μL IP buffer.

BRET assays were performed in white 96-well plates as described previously.⁴ 200 nM Alexa 594-linked PS-ASO 1024789(24) as well as the indicated PS-ASOs were incubated at room temperature for 15 min in 1× binding buffer (100 mM NaCl, 20 mM Tris-HCl [pH 7.5], 1 mM EDTA, and 0.1% NP-40) with 10⁶ RLU/well of Ni-NTA-purified NLuc fusion protein. Following incubation, NanoGlo substrate (Promega) was added at 0.1 μL/well. Readings were performed for 0.3 s using a Glomax Discover system using a 450/8-nm band pass for the donor filter and 600-nm long pass for the acceptor filter. BRET was calculated as the ratio of the emission at 600/450 nm (fluorescent excitation emission/RLU).

Data and code availability

Data are available upon request from the authors.