The ROQ domain of Roquin recognizes mRNA constitutive decay element and double-stranded RNA

Abstract

A conserved stem-loop motif of the constitutive decay element (CDE) in the 3′ UTR of mRNAs is recognized by the ROQ domain of Roquin, which mediates their degradation. Here we report two crystal structures of the Homo sapiens ROQ domain in complex with CDE RNA. The ROQ domain has an elongated shape, with three sub-domains. The 19-nt Hmgxb3 CDE is bound as a stem-loop to domain III. The 23-nt TNF RNA is bound as a duplex, to a separate site at the interface between domains I and II. Mutagenesis studies confirm that the ROQ domain has two separate RNA binding sites, one for stem-loop RNA (A site) and the other for dsRNA (B site). Mutation in either site perturbs the Roquin-mediated degradation of HMGXB3 and IL-6 mRNAs in human cells, demonstrating the importance of both sites for mRNA decay.

The cellular levels of mRNAs are regulated through many different mechanisms, which control their biosynthesis, stability and degradation. For example, microRNAs can recognize appropriate sequence motifs in the mRNAs and lead to their destruction and/or repression of their translation by the ribosome. Other sequence elements in the mRNAs can recruit protein factors to regulate their levels. A well-known example is the AU-rich element (ARE) found in the 3′ UTR of many cytokine, proto-oncogene and transcription factor mRNAs ^1-3. The ARE mediates the rapid decay of these mRNAs by recruiting deadenylation and decapping enzymes as well as other proteins, in response to defined cellular signals. Therefore, the ARE functions primarily as a regulated decay element for these mRNAs.

Recent studies have identified a constitutive decay element (CDE) in the 3′ UTR of tumor necrosis factor α (TNF-α), inhibitor of κB (IκB), inducible T cell costimulator (ICOS), and other mRNAs ^4,5. The CDE mediates the post-transcriptional, microRNA-independent repression of these mRNAs, irrespective of the cellular environment. A conserved stem-loop motif in the CDE is recognized by the ROQ domain of paralogous proteins Roquin and Roquin-2 (ref. 5), which recruit the deadenylation and decapping machineries to initiate mRNA degradation ^5,6. Roquin has a central role in repressing autoimmunity, and mice deficient in both Roquin and Roquin-2 in T cells or carrying a single-point mutation in the ROQ domain of Roquin (M199R, sanroque mutation) have increased ICOS expression and exhibit a lupus-like autoimmune phenotype ^6-11. On the other hand, overexpression of Roquin in T cells promotes the development of arthritis in a mouse model ¹². Roquin knockout mice show perinatal lethality but not autoimmunity, suggesting that Roquin may have important developmental functions as well ^9,13.

Human Roquin contains 1133 amino acid residues (125 kD, Fig. 1a). The N-terminal RING finger, ROQ and CCCH-type zinc finger (ZF) domains are well conserved among metazoans (Supplementary Fig. 1). The ROQ domain does not share recognizable sequence conservation with other proteins. It is required for binding the CDE, while the RING and ZF domains are dispensable for this interaction ⁵. The ROQ domain of Roquin-2 is 88% identical to that of Roquin (Supplementary Fig. 1) and Roquin-2 can complement Roquin’s function in repressing autoimmunity ^9,10. The C-terminal segment of Roquin following the ZF is poorly conserved in sequence (Supplementary Fig. 1), but it recruits the deadenylation machinery and is required for CDE-mediated decay ^5,6.

Figure 1. Structure of the ROQ domain of human Roquin in complex with the constitutive decay element (CDE) of Hmgxb3 mRNA.

(a). Domain organization of human Roquin. Domains I, II and III of the ROQ domain are colored yellow, cyan and green, respectively. ZF: zinc-finger. (b). Schematic drawing of three RNA molecules used in this study, Hmg₁₉, TNF₂₃, and TNF_ds. The numbering scheme of the nucleotides is also shown. (c). Two views of the structure of the ROQ domain of human Roquin in complex with Hmg₁₉. The three sub-domains of the ROQ domain are colored in yellow, cyan and green, respectively, and the RNA in orange. The side chain of Met199 (site of the sanroque mutation) is shown in red for carbon atoms. (d). Overlay of the structures of domain III of ROQ domain (green) in complex with Hmg₁₉ (orange) and domain II of ROQ domain (cyan) with TNF₂₃ duplex (yellow). All the structure figures were produced with PyMOL (www.pymol.org).

To understand the molecular basis for CDE recognition by the ROQ domain, we have determined the crystal structures of this domain from human Roquin in complex with two different CDEs, from Hmgxb3 and TNF mRNAs. The ROQ domain is actually composed of three sub-domains, I, II and III, and the structural analysis reveals previously unrecognized similarity of domains II and III to the helix-turn-helix (HTH) and winged-helix (WH) motifs. The Hmgxb3 CDE stem-loop interacts with the WH motif of domain III, which primarily recognizes the 5′ arm and the loop of the RNA (A site). Two bases of the 3-nt loop are flipped out to make close contacts with the protein. Unexpectedly, the TNF RNA is bound as a duplex, due to its higher AU content in the stem and hence lower stability of the stem-loop conformation. This RNA is located in a separate binding site (B site), at the interface between domains I and II. Our mutagenesis, biochemical and cellular experiments confirm the importance of both RNA binding sites for the functions of Roquin in mediating mRNA decay.

Results

Overall structure of the ROQ domain

We determined the structure of the ROQ domain of human Roquin in complex with a 23-nt CDE of human TNF-α (TNF₂₃) at 1.9 Å resolution (Table 1). TNF₂₃ contains the 17-nt conserved stem-loop as well as 3-nt flanking sequences at its 5′ and 3′ ends (Fig. 1b) ⁵. We next determined the structure of the ROQ domain in complex with a 19-nt CDE of mouse Hmgxb3 (Hmg₁₉) at 2.9 Å resolution, in a different crystal form. This CDE is more GC rich in the stem but has the same 3-nt loop as TNF₂₃ (Fig. 1b). Efforts at crystallizing the ROQ domain alone have not been successful.

Table 1. Data collection and refinement statistics.

	ROQ-Hmg19 complex	ROQ-TNF23 complex
Data collection
Space group	P21	P212121
Cell dimensions
a, b, c (Å)	49.6, 170.0, 51.1	90.8, 93.1, 100.2
α, β, γ (°)	90, 97.0, 90	90, 90, 90
Resolution (Å)	40–2.9 (3.0–2.9) 1	40.0–1.9 (2.0–1.9)
Rmerge (%)	9.4 (41.5)	6.9 (43.9)
I / σI	9.1 (1.7)	19.4 (3.4)
Completeness (%)	97 (88)	100 (100)
Redundancy	2.1 (1.9)	3.8 (3.8)
Refinement
Resolution (Å)	40–2.9	40–1.9
No. reflections	17,984	128,927 2
Rwork / Rfree	19.5 / 25.1	17.7 / 21.2
No. atoms
Protein	4701	4783
Nucleic acid	678	767
Ligand/ion	8	20
Water	2	885
B factors (Å2)
Protein	54.4	22.1
Nucleic acid	51.1	45.3
Ligand/ion	57.9	34.0
Water	41.9	34.0
r.m.s. deviations
Bond lengths (Å)	0.010	0.011
Bond angles (°)	1.4	1.2

One crystal was used for each data collection.

¹Values in parentheses are for highest-resolution shell.

²The Friedel pairs were kept as separate reflections during the refinement.

The structure of the ROQ domain is mostly helical, with 13 helices (αA-αM) and a small three-stranded β-sheet (β1-β3) (Fig. 1c). The structure has an elongated shape, with dimensions of approximately 25 Å × 40 Å × 75 Å, and can be divided into three sub-domains. Sub-domain I has six tightly packed anti-parallel helices, three from the N-terminal end of the ROQ domain (αA-αC) and three from the C-terminal end (αK-αM). The ROQ domain has generally been defined in the literature as containing residues 131-360. Our structure shows however that the domain actually covers residues 90-400 (Fig. 1a), and the extra ~40 residues at the N and C termini contribute helices αA and αM, respectively.

Sub-domain III is formed by residues in the middle of the ROQ domain (195-271) and contains three helices (αE-αG) and the three-stranded β-sheet (Fig. 1c). Unexpectedly, the backbone fold of this domain is similar to that of the winged-helix motif, especially the WH-B domain of the RING ubiquitin-protein ligase cullin ^14-16 and related proteins ¹⁷. For example, the rms distance among equivalent Cα atoms of this domain and the WH-B domain of cullin 1 is 2.3 Å (Supplementary Note 1, Supplementary Fig. 2), although the amino acid sequence identity of the aligned residues is only 11%, based on an analysis with the program DaliLite (Z score of 9.2) ¹⁸. Another structural homolog of this domain is the WH motif of elongation factor SelB, which is responsible for selenocysteine incorporation during translation ¹⁹ (Supplementary Fig. 2). The rms distance is 2.7 Å and the sequence identity is 13% (Z score of 5.6).

Sub-domain II, with four helices (αD, αH-αJ), connects domains I and III, and there are no direct contacts between them (Fig. 1c). It has tight contacts with domain III, with ~900 Ų surface area burial, whereas the interface with domain I is more limited (~200 Ų surface area burial) and mostly hydrophilic in nature. To our surprise, helices αH-αJ form another helix-turn-helix (HTH) motif in the ROQ domain, and its structural homologs include other HTH and winged-helix proteins, such as the Zα domain of human adenosine deaminase (ADAR1, Supplementary Fig. 2) ²⁰. The rms distance is 1.8 Å for 48 equivalent Cα atoms, and the sequence identity is 14% (Z score of 4.9).

In fact, the arrangement of helices αH-αJ in domain II is also similar to that of helices αE-αG in domain III, even though domain II lacks the three-stranded β-sheet (the ‘wing’) compared to domain III (Fig. 1d, Supplementary Note 1). The rms distance is 2.1 Å for 44 equivalent Cα atoms, and the sequence conservation is 16% (Z score 3.8, indicating a more remote similarity between the two domains). Helix αD of domain II is not a part of the HTH motif but has close interactions with helices αH and αI.

Binding mode of the Hmgxb3 CDE

We chose the Hmgxb3 CDE for our studies as it has four G-C base pairs in the stem and is therefore expected to be more stable as a stem-loop than the TNF CDE (Fig. 1b). The stem of Hmg₁₉ contains six Watson-Crick base pairs and assumes a slightly distorted A-form RNA structure (Fig. 2a). A U₁-G₁₇ wobble at the base of the stem was proposed in an earlier report ⁵, but this base pair was not formed and in fact G₁₇ is disordered (Fig. 2a). These two nucleotides are present in TNF and Hmgxb3 but are not conserved in the other CDEs (Supplementary Fig. 3).

Figure 2. Interactions between the Hmg₁₉ CDE and the ROQ domain.

(a). Composite omit F_o-F_c electron density map at 2.9 Å resolution for the Hmg₁₉ RNA, contoured at 3σ. (b). Interactions of the 3-nt UGU loop and the top base pair of the stem with domain III of the ROQ domain. Hydrogen bonds and ionic interactions are shown as red dashed lines. (c). Molecular surface of the ROQ domain near the binding site for the 3-nt UGU loop, showing the close contacts between domain III of the ROQ domain and the tri-loop. (d). Interactions between the 5′ arm and U₁ of Hmg₁₉ with domains III and II of the ROQ domain.

The majority of the interactions between Hmg₁₉ and the ROQ domain are mediated through the 3-nt loop and the 5′ arm of the stem, which contact primarily domain III (helices αF and αG of the HTH motif and the β2-β3 loop of the wing) of the ROQ domain (Figs. 2b-d). In contrast, the 3′ arm of the stem has few contacts with the protein and weaker electron density (Fig. 2a).

We will refer to this binding site for the Hmg₁₉ RNA as the A site (Fig. 1c). It is located in an electropositive surface patch in the ROQ domain (Supplementary Fig. 4). The protein-RNA interface involves hydrogen-bonding and ionic interactions with the backbone phosphate groups of the RNA, and π-stacking interactions for a few of the bases (Supplementary Note 2). Both pyrimidine bases of the U₈G₉U₁₀ tri-loop are flipped out and have close contacts with the protein (Figs. 2b, 2c). The base of G₉ is stacked with the last base pair of the stem as well as the guanidinium group of Arg219 (αF). The base of U₁ is flipped ~180° from the helical pattern of the stem (Fig. 2a) and is π-stacked with the side chain of Trp184, in helix αD of domain II (Fig. 2d).

Amino acids in this interface are highly conserved among the ROQ domains (Supplementary Fig. 1). However, none of the nucleotides of Hmg₁₉ appears to be recognized specifically by the A site. On the other hand, this site recognizes the shape of the 3-nt loop (Fig. 2c) and is unlikely to accommodate other types of loops. This provides a degree a selectivity in the interactions between the ROQ domain and Hmg₁₉, and explains the fact that all known CDEs have a tri-loop with pyrimidines at the first and third positions and purine at the second position ⁵.

Binding mode of the TNF₂₃ RNA

Unexpectedly, our crystallographic analysis revealed that the TNF₂₃ RNA is not in a stem-loop conformation in the crystal. Instead, we found the two TNF₂₃ molecules in an anti-parallel, double helical structure, formed by the stem regions of the two RNAs (Supplementary Fig. 5). The 3-nt loop gives rise to an unpaired region in the middle of this duplex, which has weaker electron density, suggesting that this region is somewhat flexible. In fact, one nucleotide in this region (U₁₀) is disordered in the crystal (Supplementary Fig. 5). Our further experiments showed that the formation of this duplex was due to the lower stability of the stem-loop conformation (low GC content) and the high concentration that was needed for crystallization (Supplementary Note 3). Under physiological conditions, the TNF CDE exists as a stem-loop and is expected to be recognized by Roquin through the A site in domain III.

The TNF₂₃ duplex contains six Watson-Crick base pairs from each of the stem regions of the TNF₂₃ RNA (Supplementary Fig. 5). The two ROQ domains in the asymmetric unit have contacts with the base-paired regions of the TNF₂₃ RNA but have no interactions with the nucleotides in the loop region in the middle of the duplex (Supplementary Fig. 5). The RNA is positioned in an electropositive surface depression at the interface of domains I and II (Supplementary Fig. 4), and we will refer to this binding site as the B site (Fig. 3a).

Figure 3. Structure of the ROQ domain of human Roquin in complex with the CDE of TNF-α mRNA.

(a). Two views of the structure of the ROQ domain of human Roquin in complex with the TNF₂₃ RNA. Only half of the TNF₂₃ RNA duplex is shown. (b). Overview of the interface between TNF₂₃ RNA (orange) and the ROQ domain (yellow and cyan). (c). Interactions between the 5′ arm of TNF₂₃ and domain II of the ROQ domain. Water molecules are shown as red spheres. (d). Interactions between the 3′ arm of TNF₂₃ and domain I of the ROQ domain

Both arms of the base-paired region interact with the ROQ domain (Fig. 3b, Supplementary Note 4, Supplementary Fig. 6). For the 5′ arm, nucleotides U₄, U₅ and U₆, near the middle of the base-paired region, have hydrogen-bonding interactions with residues in helix αJ of domain II (Fig. 3c). In addition, the interactions with the base of U₄ are consistent with pyrimidines, as they involve the carbonyl oxygen on the C2 atom that is shared between U and C. For the 3′ arm, nucleotides A₁₄, A₁₅ and C₁₆ near the end of the duplex contact residues in helices αB and αC of domain I (Fig. 3d). These residues that interact with the RNA are highly conserved among the ROQ domains (Supplementary Fig. 1). However, neither arm of the RNA appears to make strong contacts with the ROQ domain. Therefore, simultaneous binding to both arms may be required for association with the B site, indicating that it is likely for dsRNA. Moreover, the exact sequence of the RNA is not specifically recognized in this binding site either.

Helix αJ in domain II that mediates the interactions with TNF₂₃ is the second helix of the HTH motif. However, the binding mode of TNF₂₃ to this domain is rather different from that of nucleic acids to other HTH motifs (Supplementary Fig. 2). Moreover, the position of TNF₂₃ relative to domain II is also strikingly different from that of Hmg₁₉ relative to domain III (Fig. 1d). The helical axes of the two RNAs are at a nearly 90° angle, most likely due to the involvement of domain I in the binding of TNF₂₃.

Mutations in the interfaces abolish RNA binding

We next introduced mutations in the ROQ domain residues that are in the interfaces with RNA and assessed their effects on binding by the electrophoretic mobility shift assay (EMSA). Mutations in the interface with Hmg₁₉ (A site) included K239E T240A (Fig. 2d) and Q247A Y250A R251E (Fig. 2b), and those in the interface with TNF₂₃ (B site) included Q318A S319A of domain II (Fig. 3c), R135E K136E R164E of domain I (Fig. 3d), and R135E K136E D322A K323A of both domains. The mutants produced similar elution profiles from cation exchange chromatography as that for the wild-type ROQ domain, suggesting that the mutations did not disrupt the overall structure of the domain.

The Hmg₁₉ and TNF₂₃ RNAs were incubated with the wild-type and mutant ROQ domain proteins in the presence of 200 mM NaCl. The EMSA results showed that mutations in the A site blocked binding to both Hmg₁₉ and TNF₂₃, while mutations in the B site had only small effects on binding of the two RNAs (Fig. 4a). These results confirm the structural observations and suggest that both Hmg₁₉ and TNF₂₃ predominantly adopt the stem-loop conformation under these experimental conditions (1 μM RNA concentration).

Figure 4. Biochemical characterizations of the ROQ domain-RNA interactions.

(a). Electrophoretic mobility shift assay (EMSA) results for wild-type and mutant ROQ domains with Hmg₁₉, TNF₂₃ and TNF_ds RNAs. The NaCl concentration was 200 mM. (b). EMSA results at 50 mM NaCl concentration. Please see Supplementary Fig. 7 for the uncropped images of the middle panel.

To confirm that the B site is truly for binding dsRNA, we created another RNA molecule by replacing the 3-nt loop of TNF₂₃ with two GC base pairs (TNF_ds, Fig. 1b). The two separate chains were mixed and annealed to form the dsRNA, and mutations in the B site completely blocked binding to this RNA (Fig. 4a). Binding of wild-type ROQ domain as well as the mutants in the A site to this RNA was observed, but the complex migrated as a smear on the gel, likely due to its poor stability. We then decreased the NaCl concentration to 50 mM and observed a clear band for the complex in the EMSA experiment, while mutations in the B site still abolished the interactions with the TNF_ds RNA (Fig. 4b). The lower ionic strength also stabilized the complex with the stem-loop RNA, such that the effects of mutations in the A site were reduced. This is consistent with the structural observations where the interface is mostly ionic and hydrophilic in nature (Fig. 2b). Finally, the ssRNA species of TNF_ds did not cause a gel shift (data not shown), proving that the B site is a dsRNA binding site.

Two separate RNA binding sites in the ROQ domain

The structures of the Hmg₁₉ and TNF₂₃ complexes demonstrate that the ROQ domain has two separate RNA binding sites, with no overlaps between them (Fig. 5a). The binding mode of Hmg₁₉ RNA in the A site clearly indicates that a dsRNA cannot be accommodated there, as the loop makes intimate contacts with the ROQ domain (Fig. 2c). Our mutagenesis and EMSA results confirm that the A site in domain III is for binding stem-loop RNA, while the B site at the interface of domains I and II is for binding dsRNA.

Figure 5. Two separate RNA binding sites in the ROQ domain.

(a). Overlay of the structures of the Hmg₁₉ complex (in color) and TNF₂₃ complex (gray) of the ROQ domain, based on domains II and III. A 30° change in the orientation of domain I is indicated with the red arrow. (b). EMSA results showing the formation of the ROQ-TNF_ds-Hmg₁₉ ternary complex. Wild-type ROQ domain was pre-incubated with labeled TNF_ds, and then with increasing concentrations of unlabeled Hmg₁₉. The NaCl concentration was 50 mM. (c). EMSA results with mutations in the A site. The mutations blocked the formation of the ternary complex. (d). A model showing possible linkages between the RNA molecules in the B site and A site. A single-stranded RNA (gray, 7 nts) was modeled to connect the 3′ end of the RNA in the B site to the 5′ end of the stem-loop RNA in the A site (~30 Å distance). A linker between the 3′ end of the RNA in the A site to the 5′ end of the RNA in the B site would need to span ~60 Å (gray dots).

The position of domain I relative to domains II and III in the Hmg₁₉ complex is substantially different from that in the TNF₂₃ complex, corresponding to a rotation of ~30° (Fig. 5a). This conformational change reduces the size of the groove between domains I and II, such that the B site in the Hmg₁₉ complex cannot accommodate dsRNA due to steric clashes. However, this change is likely due to the limited interface between domains I and II, and the consequent flexibility of domain I, rather than being caused by Hmg₁₉ binding.

To obtain direct experimental evidence that the ROQ domain can bind both stem-loop RNA and dsRNA at the same time, we pre-incubated wild-type ROQ domain with fluorescently labeled TNF_ds RNA, and then introduced increasing concentrations of unlabeled Hmg₁₉ RNA. The EMSA result clearly showed a faster migrating species in the presence of Hmg₁₉, corresponding to the formation of the ROQ-TNF_ds-Hmg₁₉ ternary complex (Fig. 5b). In comparison, mutations in the A site that blocked Hmg₁₉ binding also abolished the formation of the ternary complex (Fig. 5c). We also reversed the sequence of incubation of the two RNAs and obtained the same results (data not shown). Therefore, the A and B sites of the ROQ domain can function independently of each other and bind RNA individually (as observed in the crystal structures) as well as simultaneously.

Both RNA binding sites are important for mRNA decay

To assess the functional significance of the two RNA binding sites in promoting decay of mRNAs targeted by Roquin, we simultaneously knocked down human Roquin and Roquin-2 in 293T cells and complemented with exogenous wild-type and mutant Roquin (Fig. 6a), which were confirmed to be stable and expressed at comparable protein levels (MZ and MK, data not shown). We tested the stability of HMGXB3 and PPP1R10 mRNAs, which were shown to be responsive to Roquin in mouse NIH3T3 cells ⁵, as well as the pleiotropic cytokine IL-6 mRNA which may possess a CDE (Supplementary Fig. 3). Reduction of endogenous Roquin and Roquin-2, by 70% and 90% respectively, led to the stabilization of the human HMGXB3 and IL-6 mRNAs that was reversed upon complementation with shRNA-resistant wild-type Roquin (Fig. 6b). In contrast, despite comparable levels of expression (Fig. 6a), neither the A site nor the B site mutant Roquin complementation could restore the decay of these mRNAs, demonstrating that both sites are necessary for efficient Roquin-mediated decay of target mRNAs in cells. The effect of Roquin knockdown on IL-6 stability is smaller, consistent with the presence of other mechanisms that regulate this mRNA ²¹. As expected, an mRNA lacking a predicted CDE, PAQR8, was not responsive to changes in Roquin levels. Surprisingly, the PPP1R10 mRNA was also nonresponsive to a reduction in Roquin levels or complementation with Roquin in human cells, indicating some functional differences between the human and mouse cells for this mRNA.

Figure 6. Functional study of Roquin in human 293T cells.

(a). mRNA levels as measured by quantitative real-time PCR. Roquin and Roquin-2 mRNA levels in 293T cells are shown after control knockdown (Con^KD), Roquin/Roquin-2 double knockdown (Roquin1/2^KD) as indicated above the figure, and complementation by add back with shRNA-resistant Roquin (WT), and Roquin (A site) and Roquin (B site) mutants. Complementation with the vector plasmid lacking the Roquin cDNA is denoted (Vector). Error bars, s.d. (n=3 cell cultures) (b). Plot showing the stability of the HMGXB3, IL-6 and PPP1R10 mRNAs. Degradation of the mRNAs was measured by quantitative real-time PCR in the same 293T cells as in panel a. Level of mRNA remaining following actinomycin D-directed transcriptional silencing at the indicated times is shown. Error bars, s.d. (n=3 cell cultures). Half-lives (t_1/2) are denoted in the figure and presented relative to GAPDH mRNA and derived from three independent experiments. Comparison of the various complementation data relative to the control are indicated by the bars, and p-values from comparison of the decay rates are presented with asterisks (* p<0.05, ** p<0.01, *** p<0.001, from two-tailed extra sum-of-squares F test).

Discussion

Our structural studies reveal that the ROQ domain actually consists of three sub-domains. Moreover, although the ROQ domain does not share recognizable sequence homology with other proteins, the structures show that it contains two HTH motifs in tandem, a WH motif in domain III and an HTH motif in domain II. These motifs are known to bind DNA and RNA as well as mediate protein-protein interactions ²². Our mutagenesis and biochemical studies show that the WH motif in domain III is crucial for binding stem-loop RNA (the A site), while the HTH motif in domain II contribute to binding dsRNA (the B site). The tight interface between domains II and III suggests that they may function together as a unit. Domain I of the ROQ domain, with a unique arrangement of helices, is also involved in dsRNA binding. Therefore, all three domains of the ROQ domain mediate or contribute to RNA binding.

The presence of two RNA binding sites in the ROQ domain is consistent with and explains the current data on Roquin. Immunoprecipitation experiments with Roquin have identified a large number of mRNAs, many of which do not contain the conserved CDE stem-loop ⁵. Within the ICOS 3′ UTR, Roquin also recognizes sequences in addition to the CDE ^6,23. Therefore, Roquin may bind sequences other than the CDE stem-loop, and the B site may have an important role in recognizing such RNAs.

The distance from the 3′ end of the dsRNA in the B site to the 5′ end of the stem-loop in the A site is ~30 Å, which can be bridged by a linker of at least 7 nts (Fig. 5d). On the other hand, the distance from the 3′ end of the stem-loop in the A site to the 5′ end of the dsRNA in the B site is ~60 Å, and a longer linker will be needed to connect these two ends (Supplementary Fig. 3). Alternatively, the two binding sites in the ROQ domain may function independently and bind separate RNA molecules.

Our studies show that domain III has a crucial role in recognizing the CDE stem-loop. The functional importance of this domain is further supported by the fact that it also contains the M199R mutation that produces the sanroque phenotype in mice ⁷. The Met199 residue is located at the N-terminal end of helix αE, but it is not in the interface with RNA (Fig. 1c). The side chain is partially exposed to the surface (Supplementary Fig. 6), and the mutation to Arg can be accommodated without significantly perturbing the structure. The structural observations are consistent with the data that the M199R mutant has similar binding affinity for the CDE as the wild-type protein ⁵, and that the mutation did not affect the stability of Roquin ⁷ but partially reduced its ability to repress ICOS expression ⁸. The role of the M199R mutation in the sanroque phenotype is likely due to a different mechanism, for example disrupting interactions with a protein partner as the Arg side chain introduced by the mutation is expected to change the surface shape and electrostatic properties of that region of domain III.

Overall, our structural and biochemical studies have defined the molecular basis for the recognition of the CDE stem-loop by the ROQ domain, and unexpectedly revealed a separate site for dsRNA recognition by this domain. Our functional studies showed that both binding sites are important for Roquin-mediated decay of target mRNAs. The structures also provide a foundation for understanding the functions of Roquin and the CDE in mediating mRNA decay as well as other cellular processes.

Online Methods

Protein expression and purification

Residues 89-407 of human Roquin (containing the ROQ domain) were sub-cloned into the pET26b vector (Novagen), and the resulting recombinant protein carried a C-terminal hexa-His tag. E. coli BL21 Star (DE3) cells transformed with this plasmid were induced with 0.5 mM IPTG and grown at 37 °C for 18 h. The soluble protein was purified through nickel-agarose affinity (Qiagen), cation-exchange (SP Sepharose Fast Flow, GE Healthcare) and gel-filtration (Sephacryl S-300, GE Healthcare) chromatography. The purified protein was concentrated to 20 mg/ml and stored at −80 °C in a solution containing 20 mM Tris (pH 8.5), 250 mM NaCl, 5 mM DTT and 5% (v/v) glycerol. The His tag was not cleaved for crystallization.

The selenomethionyl protein sample was produced in E. coli B834 (DE3) cells, and the bacteria were grown in defined LeMaster media supplemented with selenomethionine ²⁴. The purification procedure was the same as that for the native protein.

The RNA samples were purchased from Integrated DNA Technologies (IDT). The TNF₂₃ and Hmg₁₉ RNAs were diluted to 300 and 50 μM concentration, respectively, in a buffer containing 20 mM Tris (pH 8.5), 250 mM NaCl and 4 mM MgCl₂ and the solution was heated at 98 °C for 5 min and then slowly cooled to room temperature. To form the ROQ-TNF₂₃ complex, 60 nmol of the protein and 72 nmol of the RNA were mixed in a solution (total volume 3 ml) containing 20 mM Tris (pH 7.9), 150 mM NaCl and 5 mM MgCl₂. The Hmg₁₉ complex was prepared with 13 nmol of ROQ domain and 15 nmol of RNA in the same buffer (total volume 1.5 ml). The mixtures were incubated on ice for 30 min prior to gel filtration chromatography. Fractions corresponding to the complex were collected and concentrated to 7.5 mg/ml (TNF₂₃ complex) or 6.4 mg/ml (Hmg₁₉ complex) in a buffer containing 20 mM Tris (pH 8.5), 250 mM NaCl, 5 mM DTT and 5% (v/v) glycerol.

Protein crystallization

Crystals of the ROQ-TNF₂₃ complex were obtained with the sitting-drop vapor-diffusion method at 20 °C. The reservoir solution contained 100 mM sodium citrate (pH 5.7), 1.5 M LiCl, and 18-21% (w/v) PEG 6000. 1 μl of the protein-RNA complex solution at 1.5 mg/ml concentration (diluted with the storage buffer without glycerol) was mixed with 1 μl of the reservoir solution. Crystals grew to full size in one week. They belong to space group P2₁2₁2₁, with unit cell parameters of a=90.8 Å, b=93.1 Å, c=100.2 Å. There are two molecules of the ROQ domain and a TNF₂₃ duplex in the asymmetric unit.

Crystals of the ROQ-Hmg₁₉ complex were obtained with the hanging-drop vapor-diffusion method at 20 °C. The reservoir solution contained 100 mM Tris (pH 8.2), 0.3 M calcium acetate and 12-14% (w/v) PEG 8000. 1 μl of the protein-RNA complex solution at 3 mg/ml concentration was mixed with 1 μl of the reservoir solution. Crystals grew to full size in two weeks. They belong to space group P2₁, with unit cell parameters of a=49.6 Å, b=170.0 Å, c=51.1 Å, and β=97.0°. There are two copies of the ROQ-Hmg₁₉ complex in the asymmetric unit.

The crystals were cryo-protected by the reservoir solution supplemented to 35% (w/v) PEG6000 (TNF₂₃ complex) or PEG8000 (Hmg₁₉ complex) and flash-frozen in liquid nitrogen for data collection at 100K.

Data collection and structure determination

The structure of the ROQ-TNF₂₃ complex was determined by the selenomethionyl anomalous diffraction method ²⁵. A single-wavelength anomalous diffraction (SAD) data set to 1.9 Å resolution was collected on a selenomethione-substituted crystal using an ADSC Quantum-315 CCD at beamline X29A of the National Synchrotron Light Source (NSLS). The diffraction data were processed and scaled with the HKL package ²⁶. The data processing statistics are summarized in Table 1.

The Se atoms were located with the program SHELXD ²⁷. The reflections were then phased with the program SOLVE and the phases were improved with the program RESOLVE ²⁸. Most of the protein residues were automatically built by RESOLVE, and the model was completed by manual building with the program Coot ²⁹. The structure refinement was performed using the programs CNS ³⁰ and Phenix ³¹.

A native data set to 2.9 Å resolution on the ROQ-Hmg₁₉ complex was collected at the X25 beamline of the NSLS, on a Pilatus 6M detector. The structure was determined by the molecular replacement method, with the program COMO ³². The three regions of the ROQ domain were searched independently in the calculation.

ROQ domain mutants

ROQ domain mutants were made with the QuikChange kit (Stratagene) and verified through sequencing. The mutant proteins were purified by nickel affinity and cation exchange chromatography.

Electrophoretic mobility shift assay (EMSA)

The Hmg₁₉, TNF₂₃ and TNF_ds 5′ arm RNAs with 6-FAM fluorescent label at the 5′ end and the unlabeled TNF_ds 3′ arm were purchased from IDT. The RNAs were diluted to 5 μM and heated at 98 °C for 5 min and then slow-cooled to room temperature. The protein-RNA mixtures were prepared in a buffer containing 20 mM Tris (pH 8.2), 5 mM MgCl₂, and 50 or 200 mM NaCl. The samples were loaded to 0.8% native agarose gel. The electrophoresis was performed in 1× TAE buffer (pH 8.0) at 4 °C, and then the RNA was visualized on a UV illuminator.

Plasmid construction

The FLAG-tagged pTK-IREShyg-FLAG plasmid was constructed in three steps. First, the polylinker of pIRESpuro3 (Clontech) was replaced with the FLAG-tag containing polylinker in pcDNA3 (ref. 33) using Sac I and Not I restriction sites to generate pIRES-FLAG-puro3 vector. Second, the puromycin resistant gene in pIRES-FLAG-puro3 was removed by Sma I and Xba I digestion and replaced with the hygromycin B resistant gene derived from pTRE2hyg (Addgene) using Stu I and Xba I to generate pIREShyg-FLAG vector. In the last step, IREShyg-FLAG cassette from the pIREShyg-FLAG vector was PCR amplified with the addition of adaptors containing NheI and NotI restriction sites at 5′ and 3′ end, respectively. Then, the Renilla luciferase cDNA within pRL-TK (Promega) was replaced by the NheI-NotI flanked IREShyg-FLAG cassette to generate the pTK-IREShyg-FLAG plasmid. The full-length human Roquin cDNA was PCR amplified using primers 5′ GAATTCGATATCCATGCCTGTACAAGCTCC 3′ and 5′ GAATTCTGCGGCCGCTTAAGGAGCAGAACTGG 3′ containing the EcoRV and NotI restriction endonuclease sites respectively and inserted into the same restriction sites of the pTK-IREShyg-FLAG plasmid in frame and downstream of the Flag-tag to generate pTK-IREShyg-FLAG-Roquin. shRNA-resistant Roquin was generated by PCR mutagenesis using primers 5′ AACGAGTAGTTAATTCACAATACGGCACACAGCCACA 3′ and 5′ TGTGCCGTATTGTGAATTAACTACTCGTTCTTGGAGGTAGGG 3′ to alter the shRNA target sequence to generate pRoquin (WT). shRNA-resistant mutants, pRoquin (A site) and pRoquin (B site), were constructed by replacing the wild-type ROQ domain region within pRoquin (WT) with the corresponding A site mutant (Q247A Y250A R251E) and B site mutant (R135E K136E D322A K323A) using BlpI/BstEII and BlpI/MfeI restriction sites, respectively. All constructs were confirmed by sequencing.

Generation of lentiviral particles

All shRNA plasmids and packaging plasmids were purchased from Sigma. 293T cells were transfected by pLKO.1-shRoquin (Sigma #TRCN0000144045) / pLKO1-shRoquin-2 (Sigma # TRCN0000294353), and pCMV-VSVG and psPAX2 with Lipofectamin2000 (Life technology) to generate viral particles. Culture medium was harvested two days post transfection and frozen at −80°C for storage.

Determination of mRNA half-lives

293T cells infected by control shRNA expressing lentivirus (Con^KD) or both Roquin and Roquin-2 specific shRNA lentivirus (Roquin1/2^KD) were transiently transfected with pTK-IREShyg-FLAG vector, pRoquin (WT), pRoquin (A site) or pRoquin (B site). Cells were treated with actinomycin D (5 mg/L, Sigma) 48 hours post transfection/infection, cells harvested and RNA isolated (TRIzol Reagent; Life Technology) at the indicated time intervals. RNAs were then reverse transcribed with Moloney murine leukemia virus reverse transcriptase (Promega) and oligo(dT) primer according to the manufacturer’s instructions. Values were quantified by real-time PCR using SYBR green PCR core reagent (Applied Biosystems) and the abundance of specific mRNAs were quantified using the standard-curve method according to the recommendations of the manufacturer. mRNA levels were normalized to the GAPDH mRNA and plotted against time. The primers used for real-time PCR are listed in Supplementary Table 1. Real-time PCR was carried out with an ABI Prism 7900HT sequence detection system. Assuming equation of first order decay rate was fitted to the data points by linear regression: y=100 × e^{(b × t)}, where y stands for the percentage of mRNA remaining and the time. mRNA half-lives were determine by t_1/2=(−ln2)/b. p-values for the decay rate b in Figure 6b were derived using extra sum-of-squares F test calculated by PRISM® software.