ATP and nucleic acids competitively modulate LLPS of the SARS-CoV2 nucleocapsid protein

Dissection of N protein and their NMR characterization

Initially we attempted to study the full-length N protein. Unfortunately, even after an extensive screening of protein concentrations and buffer conditions, the NMR resonance signals of the full-length N protein were very broad as exemplified by its one-dimensional NMR proton spectra (Fig. 1c), and consequently the signals of its amide protons were too broad to be detected in HSQC spectrum even without phase separation, consistent with a just published NMR study¹⁴.

To identify the domains suitable for high-resolution NMR studies, we first dissected the full-length N protein into two large fragments consisting of differential domains, namely N (1-249) with IDR1, NTD and IDR2 as well as N (175-419) with IDR2, CTD and IDR3, together with the isolated NTD and CTD (Fig. 1b). The isolated NTD and CTD both have well-dispersed HSQC spectra even with peaks highly superimposable to those of the isolated NTD and CTD previously collected in slightly different buffers^13,38,39. In particular, in their one-dimensional proton NMR spectra (Fig. 1c), NTD has two very up-field signature signals respectively at −0.78 and −1.37 ppm while CTD has two at −0.25 and −0.58 ppm, which are all from the methyl groups with the close contact to the aromatic rings only observed in the well-folded proteins. Strikingly N (1-249) has a well-dispersed HSQC spectrum (I of Fig. 1d) and in particular, the HSQC peaks of its NTD residues are highly superimposable to those of the isolated NTD residues (II of Fig. 1d), indicating that the structures of NTD are highly similar in both isolated NTD and N (1-249).

By contrast, N (175-419) has a narrowly-dispersed HSQC spectrum (I of Fig. 1e) in which the well-dispersed peaks of the isolated CTD were completely undetected. Nevertheless, a close examination showed that the signature peaks of CTD could be still observed in the 1D spectra of the full-length (II of Fig. 1c) as well as (175-419) (Supplementary Fig. 2), implying that CTD is also similarly folded in the isolated domain, N (175-414) and the full-length N proteins. To confirm this, we further dissected N (175-419) into N (175-364) and N (247-419). As shown by their 1D spectra (Supplementary Fig. 3a), N (175-464) and N (247-419) also have the signature peaks respectively at −0.25 and −0.58 ppm characteristic of the folded CTD which are less broad than those of N (175-419). In particular, well-dispersed HSQC peaks of CTD became detectable for N (175-464) and N (247-419), which are largely superimposable to those of the isolated CTD (Supplementary Fig. 3b, c). As such, most likely due to the self-association/oligomerization or/and μs-ms conformational dynamics provoked by the presence of IDRs and CTD, their NMR signals became broadened to different degrees for N (175-419) and full-length N proteins, consistent with the recent NMR reports^14,36,43,44.

Intriguingly, even in N (1-249) while most of HSQC peaks of the non-proline NTD and IDR1 residues could be detected, HSQC peaks of several segments of IDR2 were too broad to be detected, which include Ser184-Ser197, Leu221-Leu222, and Leu224-G236, thus implying that these segments are involved in μs-ms exchanges/dynamics as recently revealed by NMR backbone dynamics^14,36. Strikingly, these segments contain several hydrophobic residues Leu and particularly are rich in Ser residues (Fig. 1a) characteristic of the prion-like domains, which, as previously found, are prone to the self-association to form amyloid fibrils⁴⁵. Similarly, a very recent study also showed that CTD with IDRs together led to the disappearance of NMR signals of the full-length N protein although NTD and CTD were shown to have no detectable interaction with each other¹⁴.

ATP specifically binds residues of both folded NTD and IDRs

Previously we have found that ATP specifically binds the nucleic-acid-binding pocket of the isolated NTD with HSQC peaks of 11 residues largely perturbed²⁵. Here we titrated N (1-249) at the same protein concentration (100 μM) and in the same buffer with ATP at 0.5, 1.0, 2.0, 4.0, 6.0, and 10.0 mM respectively. Upon further increasing ATP concentration, a small set of HSQC peaks underwent large shift as illustrated by Supplementary Fig. 4a. Noticeably, in addition to the NTD residues, several peaks of the residues within IDRs unique for N (1-249) also underwent large shift (Supplementary Fig. 4b, c). Supplementary Fig. 4d presents the expanded shift tracings of two selected residues: Ala90 within NTD and Arg32 from IDR1.

Detailed analysis led to identification of largely-perturbed residue of N (1-249) upon adding ATP. The overall pattern of shifted HSQC peaks of the NTD residues are very similar in both isolated NTD and N (1-249) (Fig. 2a). Precisely, 10 NTD residues have considerable shifts including all those previously identified 11 residues in the isolated NTD except for Leu56, namely Asn48, Ser51, Thr57, Arg89, Ala90, Arg92, Ser105, Arg107, Ala155 and Tyr172 (Fig. 2b). Comparison of the shift tracings of six representative residues showed that in the context of N (1-249) their shift tracings approached being saturated at the slightly lower ATP concentrations than for the isolated NTD (Fig. 2c). Indeed, the fitting of the shift tracings of 10 NTD residues in N (1-249) gave the average Kd of 2.0 ± 0.2 mM (Supplementary Table 1), which is only slightly smaller than that for the isolated NTD (3.3 ± 0.4 mM), which is similar to our previous observations that ATP binds the tandem-linked RRM domains (Fig. 2d) with the slightly higher affinity than the isolated ones of TDP-43 and hnRNPA1⁴⁶. Most strikingly, ATP also induced the large shifts of four Arg residues respectively within IDR1 and IDR2, namely Arg10, Arg14, Arg32, and Arg209 (Fig. 2a). The fitting of their shift tracings (Fig. 2b) gave the average Kd of 2.8 ± 0.2 mM (Supplementary Table 1), which is only slightly larger than that of the NTD residues in the same context of N (1-249).

a Residue-specific chemical shift difference (CSD) of N (1-249) (blue) and isolated NTD (red) between the free state and in the presence of ATP 10 mM. The largely perturbed residues are defined as those with the CSD values at 10 mM > 0.072 (average value + one standard deviation) (cyan line). b Shift tracings of HSQC peaks of six representative NTD residue in the context of N (1-249) and isolated NTD, as well as 4 Arg residues within IDRs in the presence of ATP at different concentrations. c The docking structure of the ATP-NTD complex we previously constructed for the isolated NTD. d Schematic representation of N (1-249) in the free state (I) and in the presence of the exceeding amounts of ATP (II).

To exclude the possibility that the observed shifts are due to the presence of Mg²⁺ in complex with ATP, we titrated MgCl₂ into N (1-249) and CTD as monitored by HSQC. No large shift was detected for both N (1-249) (Supplementary Fig. 5a) and CTD (Supplementary Fig. 5b) even with MgCl₂ concentrations up to 20 mM. The results together reveal that ATP is able to bind the nucleic-acid-binding pocket of the folded NTD in the context of N (1-249) with the complex structure very similar to what we previously constructed for the isolated NTD (Fig. 3c). Furthermore, here we decoded that ATP is also capable of binding Arg residues within IDRs with Kd of 2.8 mM (Fig. 3d), whose affinity is comparable to those for binding a list of the folded nucleic-acid-binding domains recently identified^30,33.

a Turbidity (absorption at 600 nm) curves of the average values of three repeated measurements (n = 3) with STD displayed as error bars of the full-length N and N (1-249) proteins in the presence of S2m at different ratios. b DIC images of the full-length N protein in the presence of S2m at different ratios. c DIC images of N (1-249) protein in the presence of S2m at different ratios.

ATP also binds the folded CTD in N (175-419)

Very recently, we shown that in addition to acting for dimerization/oligomerization, CTD in fact is a cryptic domain for binding ATP and S2m³¹. In particular, CTD binds ATP with Kd of 1.49 ± 0.28 mM³¹. Here we asked the question whether ATP can bind N (175-419). As evidenced by its 1D spectra in the presence of ATP at different concentrations (Supplementary Fig. 6a), ATP did specifically induce the shift of one very up-field NMR peak at -0.25 but not the one at -0.58 ppm, exactly as we previously observed on the isolated CTD³¹. This result clearly indicates that ATP is able to bind the folded CTD in the context of N (175-419). On the other hand, however, as N (175-419) have HSQC spectrum with many peaks undetectable likely due to μs-ms conformational dynamics or/and oligomerization, although ATP also triggered the broadening and disappearance of some HSQC peaks of IDRs (Supplementary Fig. 6b–e), it is impossible to assign these peaks to the corresponding residues.

S2m modulates LLPS of the full-length N and N (1-249) proteins

Previously we found that A24, a 24-mer non-specific nucleic acid, was sufficient to achieve the biphasic modulation of LLPS of N protein: induction at low concentrations but dissolution at high concentrations²⁵. Here, we selected S2m, a nucleic acid probe derived from gRNA which was previously utilized to identify nucleic-acid-binding domains of SARS-CoVs³⁷ to assess its modulation of LLPS in parallel for the isolated NTD and CTD as well as the full-length N and N (1-249) proteins as monitored by measuring the turbidity (absorption at 600 nm) and imaging with DIC microscopy, as we previously conducted on SARS-CoV-2 N protein²⁵, FUS²⁸ and TDP-43²⁹.

We first titrated S2m into the isolated NTD and CTD with protein concentrations reaching up to 200 μM but found no phase separation. By contrast, as shown in Fig. 3, S2m could biphasically modulate LLPS of the full-length N protein: induction at low ratios and dissolution at high ratios. Briefly, the N protein sample showed no LLPS in the free state. However, phase separation was induced upon addition of S2m as evident by the increase of turbidity and DIC imaging. At 1:0.75 (N:S2m), the turbidity reached the highest of 1.92 (Fig. 3a) and many liquid droplets with the diameter of ~1 μm were formed (Fig. 3b). However, further addition of S2m led to the reduction of turbidity and dissolution of the droplets. At 1:1.5 all liquid droplets were completely dissolved.

Subsequently, we titrated S2m into the N (1-249) protein under the same conditions. Again N (1-249) showed no phase separation in the free state. Nevertheless, S2m could also biphasically modulate LLPS of the N (1-249) protein. Briefly, LLPS could be induced upon addition of S2m and at 1:0.75 (1-249:S2m), the turbidity reached the highest of 1.45 (Fig. 3a) and many liquid droplets with the diameter of ~1 μm were also formed (Fig. 3c). Similar to what was observed above on the full-length N protein, further addition of S2m led to the reduction of turbidity and dissolution of the droplets. At 1:1.25 all liquid droplets were dissolved. Compared with the liquid droplets formed by the full-length N protein in the presence of S2m at 1:0.75, the droplets formed by the N (1-249) protein with S2m at 1:0.75 have similar sizes but the number were less, thus resulting in the lower turbidity (Fig. 3a).

Residue-specific NMR view of LLPS of N (1-249) modulated by S2m

So far, no high-resolution mechanism has been reported on LLPS of N protein induced by nucleic acid. Here, our successful identification of N (1-249) offered us the opportunity to gain residue-specific view of LLPS by NMR spectroscopy. To achieve this, we monitored the S2m-modulated LLPS by NMR upon stepwise addition of S2m into the N (1-249) sample at ratios 1:0.05, 1:0.1, 1:0.25, 1:0.75, 1:1, and 1:2,5, which range from the induction to complete dissolution of liquid droplets.

As shown in I of Supplementary Fig. 7, even upon addition of S2m at 1:0.05, some HSQC peaks became broadened and consequently their intensity reduced. At 1:0.1 (II of Supplementary Fig. 7) further broadening was observed for many HSQC peaks. Noticeably, at 1:0.25, HSQC peaks of all folded NTD and some IDR residues became disappeared while at 1:0.75, even the intensity of the remaining peaks of IDR residues became largely reduced (III of Supplementary Fig. 7). On the other hand, however, further addition of S2m to 1:1 led to the intensity increase for remaining IDR HSQC peaks, and at 1:2.5, the intensity of those IDR HSQC peaks further increased (V of Supplementary Fig. 7). Nevertheless, even at 1:2.5 where liquid droplets were completely dissolved, the HSQC peaks of NTD and some IDR residues still remained undetectable (VI of Supplementary Fig. 7).

Detailed analysis of the peak intensity of the spectra at different S2m ratios reveals a very interesting picture (Fig. 4): as illustrated by Fig. 4a, even with the addition of S2m at 1:0.05, the intensity of the NTD peaks largely reduced with an average of 0.51 while the intensity of IDR peaks showed the relatively small reduction with an average of 0.79. At 1:0.1, the intensity of the NTD peaks further reduced with an average of 0.31 while the intensity of IDR peaks still has an average of 0.73. Strikingly, at 1:0.25, all NTD peaks became too weak to be detected but IDR peaks still has an average of 0.54. At 1:0.75, the intensity of the remaining IDR peaks became further reduced to 0.34. Intriguingly, however, with further addition of S2m to 1:1, the intensity of IDR residues became increased with the average of 0.48 and further back to 0.76 at 1:2.5.

a Normalized intensity of N (1-249) residues in the presence of S2m at different ratios as divided by that of N (1-249) residues without S2m. b Normalized intensity of three groups of N (1-249) residues in the presence of S2m at different ratios. c Speculative model of homogeneous solution of N (1-249) (I) which is induced to phase separate upon adding S2m at low ratios (II) followed by dissolution into homogeneous solution at high ratios (III).

Furthermore, a close examination indicates that based on the patterns of the intensity changes, the N (1-249) residues can be grouped into three categories (Fig. 4b): the folded NTD residues (group 1) and 7 Arg/Lys residues within IDRs (group 2) with their HSQC peak intensity reduced rapidly upon adding S2m and becoming too weak to be detected above 1:0.25, Interestingly those two categories of residues had their HSQC peaks remaining undetectable even with further addition of S2m up to 1:2.5. By contrast, for other IDR residues (group 3), their intensity reduced with addition of S2m and reached the lowest at 1:0.75. However, with further addition of S2m, their peak intensity increased and at 1:2.5, the average intensity is back to 0.76.

In titrations with S2m, the intensity of HSQC peaks of N (1-249) appears to be mechanistically modulated by at least two processes: 1) the formation of large and dynamically crosslinked complexes between S2m and N (1-249) molecules, which is expected to slow the rotational motions and consequently lead to the broadening of HSQC peaks; and 2) the binding-induced dynamics on μs-ms time scale that also result in the broadening of HSQC peaks^{28,29,33,34,47,48,49}. As NTD is a folded domain, its residues largely behave as a coupled unit. By contrast, due to the lack of the folded structure, IDR residues behave rather independently. As such, upon binding with S2m, HSQC peaks of all NTD residues become uniformly broadened and disappeared due to the μM Kd or/and provoked μs-ms dynamics. By contrast, IDR residues have rather independent dynamic behaviors and therefore only the peaks of Arg/Lys residues directly bound with S2m become largely broadened and disappeared, while the peaks of other IDR residues only have reduced intensity due to the slowing of the rotational motions upon forming the large dynamically-crosslinked complexes^28,44. However, upon further addition of S2m, the large complexes are disrupted due to the excessive binding and consequently the intensity of most IDR residues except Arg/Lys become increased. Nevertheless, the peaks of NTD and Arg/Lys peaks remain undetectable even in the exceeding presence of S2m because these residues are still bound with S2m even after the complete dissolution of liquid droplets in the presence of the exceeding amount of S2m.

Previously RNA^50,51 and ssDNA⁵² were shown to biphasically modulate LLPS of FUS and TDP-43 respectively but the general mechanisms remain largely elusive. Here the results indicate that S2m appears to achieve both induction and dissolution of LLPS of N (1-249) mainly by specifically binding to the folded NTD and Arg/Lys residues within IDRs. In this context, a speculative model was proposed (Fig. 4c). Briefly, upon adding S2m into the homogenous solution of N (1-249) (I of Fig. 4c) at low ratios, N (1-249) and S2m are capable of dynamically and multivalently interacting with each other over both NTD and IDR Arg/Lys residues to form large and dynamically-crosslinked complexes^28,44,48 which manifest as liquid droplets (II of Fig. 4c). However, with the exceeding addition of S2m, several S2m molecules become bound with one N (1-249) molecule and consequently the large and dynamically-crosslinked complexes become disrupted, thus manifesting as the dissolution of liquid droplets into homogeneous solution (III of Fig. 4c).

ATP dissolves LLPS of the full-length N and N (1-249) proteins in the same manner

The residue-specific NMR results together imply that ATP and S2m might in fact have the highly overlapped binding sites on N (1-249), namely NTD and Arg/Lys of IDRs despite having very different affinities. If this is the case, ATP at the higher concentrations than that of S2m should be able to dissolve the S2m-induced LLPS of N (1-249) by competitively displacing S2m from being bound with the proteins. To verify this mechanism, we prepared the phase separated samples of both full-length N and N (1-249) proteins with the pre-presence of S2m at 1:0.75. Subsequently, ATP was added into the samples in a stepwise manner, as monitored by turbidity (Fig. 5a) and DIC imaging (Fig. 5b and Supplementary Fig. 8).

a Turbidity curves of the average values of three repeated measurements (n = 3) with STD displayed as error bars of the full-length N protein and N (1-249) in the presence of S2m at 1:0.75 with additional addition of ATP at different ratios. b DIC images of N (1-249) in the presence of S2m at 1:0.75 with additional addition of ATP at different ratios.

Indeed, ATP could dissolve LLPS of both full-length N and N (1-249) proteins induced by S2m, although ATP completely dissolved LLPS of N (1-249) at the slightly lower ratio (1:500) than that for dissolving LLPS of the full-length N protein (1:750). This difference is most likely due to the presence of CTD in the full-length N protein which might enhance LLPS by additional involvement in binding S2m or/and dimerization/oligomerization. Briefly, as for N (1-249), at the ratio < 1:25, ATP only has minor effect on LLPS. However, at ratio of 1:250, ATP dissolved liquid droplets as evidenced by the reduction of turbidity and disappearance of many liquid droplets. Strikingly, at 1:500, ATP could completely dissolve liquid droplets of N (1-249) protein induced by S2m. The result clearly suggests that ATP and S2m do interplay in modulating LLPS of the full-length N and N (1-249) proteins in the same manner. To exclude the possibility that the above observation is due to the alteration of the conformation of S2m by ATP-Mg complex, we titrated ATP-Mg complex into a S2m sample as monitored by 1D proton NMR spectroscopy on the protons of the base aromatic rings and NH2 which are directly involved in base pairing. The obtained results showed that addition of ATP-Mg complex even up to 10 mM triggered no shift of these NMR signals (Fig. S5C and S5D), thus unambiguously indicating that no conformational alteration occurred for S2m upon addition of ATP-Mg complex.

ATP dissolves LLPS by competing with S2m for binding the protein

To understand the high-resolution mechanism, we monitored the dissolution of the S2m-induced LLPS of N (1-249) by ATP with HSQC spectroscopy. As shown in I of Fig. 6a, the addition of ATP at 1 mM into the N (1-249) sample with the pre-existence of S2m at 1:0.75 has very minor effect on its HSQC spectrum. However, addition of ATP at 10 mM led to the restore of some HSQC peaks (II of Fig. 6a) and at 20 mM, the disappeared HSQC peaks including those of the NTD and Arg/Lys residues were re-appeared (III of Fig. 6a). Most strikingly, the HSQC spectrum in the presence of S2m at 1:0.75 with additional addition of ATP at 20 mM is highly similar to that without S2m but only in the presence of ATP at 10 mM. This observation unambiguously suggests that ATP is able to bind NTD and IDR Arg residues as well as to displace S2m from being bound with the protein. Nevertheless, even at 20 mM, many HSQC peaks, particularly from NTD residues are still weak as compared with those of N (1-249) only in the presence of ATP at 10 mM, implying that ATP even at 20 mM is still unable to completely displace the binding of S2m from N (1-249) because the binding affinity of S2m to N (1-249) is much higher (with Kd of ~μM) than that of ATP (Kd of ~mM).

a Superimposition of HSQC spectra of N (1-249) in the presence of S2m at 1:0.75 (red) with additional addition of ATP (blue) at 5 mM (I), 10 mM (II) and 20 mM (III). (IV) Superimposition of HSQC spectra of N (1-249) in the presence of ATP at 10 mM only (red) and in the presence of both S2m at 1:0.75 and ATP at 20 mM (blue). b Speculative model of N (1-249) in homogeneous solution (I) which undergoes phase separation to form dynamic liquid droplets upon induction by S2m (II), followed by dissolution of droplets into homogeneous solution upon adding the exceeding amount of ATP (III). c Illustration icons.

Here the NMR competition experiments confirm the above speculation that ATP and S2m share the highly-overlapped binding sites over both folded NTD and IDRs of N (1-249). As a consequence, S2m interacts with N (1-249) (I of Fig. 6b) to induce LLPS by forming the large and dynamically crosslinked complexes (II of Fig. 6b), which is characterized by an extensive intensity reduction and disappearance of HSQC peaks. However, the addition of ATP at the high concentrations is able to displace S2m from being bound with N (1-249), thus leading to the dissolution of the liquid droplets (III of Fig. 6b), with the reappearance of HSQC peaks including those of the NTD and Arg/Lys residues. Nevertheless, due to the fact that S2m binds N (1-249) with Kd of ~μM, which is much higher that that of ATP (Kd of ~mM), ATP at 20 mM is still unable to completely displace S2m from binding N (1-249), and consequently the intensity of HSQC peaks of N (1-249) in the presence of both S2m and ATP is still weaker than that only in the presence of ATP.

S2m modulates LLPS of N (175-419) protein

We also assess whether S2m could induce LLPS for N (175-419), N (175-365) and N (247-419). As shown in Fig. 7a, the N (175-419) sample showed no LLPS in the free state but phase separation was induced upon addition of S2m as evident by the increase of turbidity and DIC imaging. At 1:0.5, the turbidity reached the highest of 1.71 (I of Fig. 7a) and many liquid droplets with the diameter of ~1 μm were formed (II of Fig. 7a). However, as observed on N protein and N (1-249) above, further addition of S2m led to the reduction of turbidity and dissolution of the droplets. At 1:1.25 all liquid droplets were completely dissolved.

a Turbidity curves of the average values of three repeated measurements (n = 3) with STD displayed as error bars of N (175-419) and N (247-419) in the presence of S2m at different ratios (I); and DIC images of N (175-419) in the presence of S2m at different ratios. b Turbidity curves of the average values of three repeated measurements (n = 3) with STD displayed as error bars of N (175-419) in the presence of S2m at 1:0.5 with additional addition of ATP at different ratios (I); and DIC images of N (175-419) in the presence of S2m at 1:0.5 with additional addition of ATP at different ratios (II).

On the other hand, for N (175-365), addition of S2m even at 1:0.1 led to visible precipitation and consequently no DIC characterization could be performed. For N (247-419), addition of S2m even up to 1:2.5 induced no LLPS as well as precipitation as reported by turbidity (I of Fig. 7a). The results suggest that CTD needs the additional presence of IDR2 with 8 Arg and 3 Lys residues to achieve multivalent and dynamic binding with S2m to drive LLPS. By contrast, as IDR3 contains only 1 Arg and 9 Lys, the interaction of S2m with Arg/Lys is not sufficiently strong for driving LLPS as the interaction between nucleic acids and Lys is much weaker that that with Arg^23,38.

We also prepared the phase separated samples of N (175-419) with the pre-presence of S2m at 1:0.5. Subsequently, ATP was added into the samples in a stepwise manner, as monitored by turbidity (I of Fig. 7b) and DIC imaging (II of Fig. 7b). Indeed, ATP could dissolve its LLPS induced by S2m and the droplets were completely dissolved at 1:500, which is in general similar to what was observed on N (1-249) and full-length N protein.

NMR characterization of the interaction of N (247-419) and S2m

We further set to characterize the interactions of S2m with N (175-419), N (175-365) and N (247-419) by NMR. Unfortunately, as to collect good quality NMR spectra needs a protein concentration of at least 50 μM, we were unable to collect NMR spectra of (175-419) and N (175-365) as they became precipitated at 50 μM upon adding S2m even at 1:0.1. Nevertheless, we have successfully collected NMR spectra of N (247-419) at 50 μM in the presence of S2m up to 1:2.5 (Supplementary Fig. 9). Intriguingly, although the addition of S2m triggered no LLPS, S2m was shown by NMR (Supplementary Fig. 9) to interact with the folded CTD as evidenced by the broadening of several very up-field NMR signals, exactly as we previously observed on the interaction of S2m with the isolated CTD³¹. On the other hand, as judged by its HSQC spectra in the presence of S2m at different concentrations, most HSQC peaks of the IDR3 residues in N (247-419) retained even up to 1:2.5, implying that the interaction of S2m with IDR3 is indeed relatively weak as compared to that between S2m and Arg-rich IDR1 and IDR2 which resulted in dramatic disappearance of many of their HSQC peaks (Figs. 3, 4).