The pre-mRNA splicing and transcription factor Tat-SF1 is a functional partner of the spliceosome SF3b1 subunit via a U2AF homology motif interface

Abstract

The transcription elongation and pre-mRNA splicing factor Tat-SF1 associates with the U2 small nuclear ribonucleoprotein (snRNP) of the spliceosome. However, the direct binding partner and underlying interactions mediating the Tat-SF1–U2 snRNP association remain unknown. Here, we identified SF3b1 as a Tat-SF1–interacting subunit of the U2 snRNP. Our 1.1 Å resolution crystal structure revealed that Tat-SF1 contains a U2AF homology motif (UHM) protein–protein interaction module. We demonstrated that Tat-SF1 preferentially and directly binds the SF3b1 subunit compared with other U2AF ligand motif (ULM)-containing splicing factors, and further established that SF3b1 association depends on the integrity of the Tat-SF1 UHM. We next compared the Tat-SF1–binding affinities for each of the five known SF3b1 ULMs and then determined the structures of representative high- and low-affinity SF3b1 ULM complexes with the Tat-SF1 UHM at 1.9 Å and 2.1 Å resolutions, respectively. These structures revealed a canonical UHM–ULM interface, comprising a Tat-SF1 binding pocket for a ULM tryptophan (SF3b1 Trp³³⁸) and electrostatic interactions with a basic ULM tail. Importantly, we found that SF3b1 regulates Tat-SF1 levels and that these two factors influence expression of overlapping representative transcripts, consistent with a functional partnership of Tat-SF1 and SF3b1. Altogether, these results define a new molecular interface of the Tat-SF1–U2 snRNP complex for gene regulation.

Introduction

Tat stimulatory factor 1 (Tat-SF1)⁴ was originally identified as a host cofactor that activates Tat-directed HIV-1 transcription (1, 2). Subsequent studies revealed that Tat-SF1 normally stimulates human transcription elongation (3, 4) as a complex with P-TEFb, SPT5, and RAP30 (2, 5–7). The Saccharomyces cerevisiae Tat-SF1 homologue, CUS2, promotes transcription elongation in combination with the RNA unwindases PRP5 and PRP11 (8). Beyond transcription, CUS2 is well-established as an assembly factor for the U2 small nuclear (sn)RNA in the early stages of pre-mRNA splicing (9–13). Like CUS2, human Tat-SF1 influences pre-mRNA splicing (14–18). Reduced Tat-SF1 levels typically promote intron retention rather than exon-skipping or alternative splice sites. Most recently, critical Tat-SF1 functions in pre-mRNA splicing have emerged for embryonic stem cell differentiation (17, 18).

Tat-SF1 co-immunoprecipitates with the U2 small nuclear ribonucleoprotein particle (17–20), which anneals with the pre-mRNA branch point site in the early steps of spliceosome assembly. Yet, the direct interaction partner(s) of Tat-SF1 in the human spliceosome remains unknown. Clues arise from experimental observations coupled with the primary sequences of Tat-SF1 and the U2 snRNP components. Primary sequence analysis suggests that Tat-SF1 contains two modular domains, an RNA recognition motif (RRM) and a U2AF homology motif (UHM) (Fig. 1A). The UHM is a protein-interaction module with specialized features for recognizing short “U2AF ligand motifs” (ULM) of target proteins (21). Among U2 snRNP subunits, SF3b1 singularly comprises ULMs that could mediate Tat-SF1 UHM association; namely, five motifs near the N-terminal region of the SF3b1 protein (Fig. 1, B and C).

Figure 1.

SF3b1 and Tat-SF1 splicing factors contain ULM and UHM motifs. Human Tat-SF1 (A, blue) and SF3b1 (B, green) domain organization are shown. Residue ranges of constructs used for crystallization (structure) or isothermal titration calorimetry (ITC) are given in parentheses. p14, p14a binding site. C, the aligned sequences of five ULM repeats from human SF3b1 are shown in the inset below. The boundaries of synthetic peptides (in parentheses and underlined) terminate at the second TP motif, which for ULM1/ULM2 and ULM2/ULM3 also corresponds to the start of the next ULM. The central tryptophan (yellow) is preceded by characteristic basic (R/K) amino acids (green).

Here, we explore the hypothesis that a UHM–ULM interface couples Tat-SF1 to the SF3b1 subunit of the U2 snRNP. We determined a 1.1 Å resolution structure that details the Tat-SF1 UHM-fold. We show that Tat-SF1 directly and specifically recognizes the ULM region of SF3b1, and determine high resolution structures of Tat-SF1 UHM complexes comparing complexes with high- and low-affinity SF3b1 ULMs. The Tat-SF1–SF3b1 ULM interactions are important for association of the proteins in cell extracts. We demonstrate that SF3b1 regulates Tat-SF1 expression levels, and that these splicing factors both regulate representative transcripts. Altogether these results elucidate a functional UHM–ULM-mediated partnership of Tat-SF1 with the SF3b1 subunit of the U2 snRNP spliceosome particle.

Results

Tat-SF1 contains a U2AF homology motif

To confirm that Tat-SF1 contains a UHM and detail its fold, we determined the crystal structure of this region at 1.1 Å resolution (Fig. 2, Table 1). Analysis using the DALI 3D comparison server (22) shows that the structure most closely matches established members of the UHM family, including CAPERα (PDB ID 4OZ1, Z-score 13.3) and U2AF⁶⁵ (PDB ID 4Z2X, Z-score 13.2). The core βαββαβ-fold also resembles an RRM (e.g. the U1A prototype, PDB ID 3K0J, Z-score 12.8), which shares a common topology and appears to be an ancestor of the UHM. A 3₁₀ helix at the N terminus of the Tat-SF1 UHM is stabilized by crystal packing, yet may have relevance in the context of the full-length protein-fold.

Figure 2.

Structure of the Tat-SF1 UHM at 1.13 Å resolution. A, view into the β-sheet. Arrows indicate the β-strands corresponding to RNP1 and RNP2 of the RRM-like fold. B, view rotated 90° about the y axis. Interactions between the C-terminal Tat-SF1 UHM extension and degenerate RNP residues are shown in surface and stick representation. The UHM is colored in a rainbow gradient from blue to red at the N and C termini.

Table 1.

X-ray data collection and refinement statistics

Values in parentheses refer to the highest resolution shell: apo 1.17–1.13 Å; ULM4 1.96–1.89 Å; ULM5 2.17–2.10 Å. For R_free, 7% of the reflections were randomly-selected then excluded from all stages of refinement.

Tat-SF1 UHM ligand	Apo	ULM4	ULM5
Data collection
Wavelength (Å)	0.83	1.13	0.98
Resolution range (Å)	30.20–1.13	35.05–1.89	37.31–2.10
Space group	P 43	P 32 21	P 21 21 21
Unit cell lengths (Å)	30.20, 30.20, 98.74	58.37, 58.37, 97.31	46.32, 58.71, 125.96
Unit cell angles (°)	90 90 90	90 90 120	90 90 90
Total reflections	178,395	90,661	101,323
Unique reflections	32,586	15,748	20,337
Multiplicity	8.7 (1.9)	5.7 (1.8)	5.0 (5.0)
Completeness (%)	98.8 (99.6)	99.3 (94.5)	98.4 (97.0)
Mean I/(I)	13.6 (2.9)	24.3 (2.7)	10.1 (1.8)
Rmerge (%)	5.4 (25.2)	4.4 (45.2)	8.7 (65.0)
Rp.i.m. (%)	3.3 (25.2)	2.0 (24.3)	4.3 (36.1)
CC1/2	0.99 (0.84)	0.99 (0.81)	0.98 (0.99)
Refinement
Rwork/Rfree (%)	12.7/16.0	17.1/19.8	17.5/21.9
Tat-SF1 UHM in ASU	1	1	2
SF3b1 ULM in ASU	–	1	1
No. non-hydrogen atoms
Tat-SF1 UHM	899	799	1573
SF3b1 ULM	–	72	39
PEG/Glycerol	13	9	13
Ions	5	–	–
Waters	114	105	151
R.m.s.d. bonds (Å)	0.007	0.010	0.013
R.m.s.d. angles (°)	0.83	0.97	1.16
Ramachandran (%)	98.9/1.1/0	100/0/0	97.3/2.7/0
favored/allowed/outlier
Clashscore (percentile)	2.76 (93)	2.93 (99)	0.94 (100)
〈B-factor〉 (Å2)	19.3	44.8	43.8
Tat-SF1 UHM	16.8	41.6	42.1
SF3b1 ULM	–	69.5	89.9
PEG/glycerol	45.2	61.0	80.1
Ions	20.7	–	–
Water	36.4	51.3	46.5
PDB code	6N3D	6N3E	6N3F

The canonical RNA interface of the RRM family involves nucleobase stacking with aromatic residues in two ribonucleoprotein consensus motifs (RNP1 and RNP2) of the β-sheet surface. In all prior UHM structures (23–28), the RNP motifs are degenerate and masked by C-terminal extensions of the core-fold. Accordingly, a C-terminal tail of the Tat-SF1 UHM (namely Trp³⁴⁷ and Tyr³⁵³) packs against the putative RNP surface (Val³¹⁴ and Ile²⁶⁷). These key RNP residues have diverged from the typical aromatic residues of RRMs, consistent with specialized roles for the Tat-SF1 UHM in protein rather than RNA interactions.

Tat-SF1 preferentially and directly binds SF3b1 compared with other ULM-containing proteins

We next tested whether the Tat-SF1 UHM binds a ULM-containing partner (Fig. 3). Most ULM-containing proteins are pre-mRNA splicing factors, including U2AF⁶⁵ (29), SF1 (30), and SF3b1 (31–33). The inner nuclear membrane protein MAN1 (34) and the poly(Q)-expanded protein ATX1 (35) also contain ULMs. Considering the influence of Tat-SF1 on pre-mRNA splicing (14–18) and its co-immunoprecipitation with the SF3b1-containing U2 snRNP (17–20), we focused on testing Tat-SF1 association with SF3b1 compared with other ULM-containing splicing factors.

Figure 3.

Tat-SF1 directly and preferentially associates with the ULM region of SF3b1. GST-pulldowns of purified recombinant proteins were resolved by SDS-PAGE and stained with Coomassie® Brilliant Blue. Three lanes were loaded for each experiment: I, input (1 μg); W, final wash of bound resin; E, elution. The sizes of molecular weight standards (STD) and subunits are indicated. GST is a control for nonspecific binding. A, GST-fused Tat-SF1 (residues 1–360) retains the SF3b1 ULM region (“SF3b1^ULM”, residues 190–344) and B, shows little retention of UHMK1-phosphorylated (P) SF1. Likewise, C, GST-U2AF⁶⁵ΔRS (residues 85–475) shows little Tat-SF1 retention, whereas GST-SF3b1^ULM significantly retains Tat-SF1.

To ensure direct interactions, we used GST pulldown assays of purified recombinant proteins (Fig. 3). Our Tat-SF1 construct comprises the RRM and UHM domains that are sufficient for snRNP co-immunoprecipitation (19) (residues 1–360, Fig. 1A). We compared Tat-SF1 retention by the ULM-containing region of SF3b1 (SF3b1^ULM, residues 190–344, Fig. 1B) with nearly full-length U2AF⁶⁵ that lacks an aggregation prone arginine-serine repeat region (ΔRS), each N-terminally tagged with GST. To avoid overlapping molecular weights, we conversely compared GST–Tat-SF1 retention of SF3b1^ULM with SF1. The SF1 protein was phosphorylated (P) in vitro with UHMK1 as described (25, 36), which subtly enhances SF1-U2AF⁶⁵ association (36, 37) and is the major form of SF1 in proliferating cells (36, 38).

GST–Tat-SF1 retains a substantial fraction of the input SF3b1^ULM and has little detectable association with (P)SF1 (Fig. 3, A and B). Although GST–U2AF⁶⁵ΔRS retains a slight fraction of untagged Tat-SF1, this interaction is likely to be nonspecific considering the nanomolar affinity of the constitutive U2AF⁶⁵-2AF³⁵ heterodimer (29). Instead, the amount of Tat-SF1 retained by GST–SF3b1^ULM is significant (Fig. 3C). This result demonstrates that Tat-SF1 specifically and directly associates with the ULM region of SF3b1.

Tat-SF1 preferentially binds the fifth SF3b1 ULM

To quantitatively characterize Tat-SF1–SF3b1 interactions, we used isothermal titration calorimetry (ITC) to measure the binding affinities of Tat-SF1 for SF3b1^ULM and each of the individual SF3b1 ULMs (Fig. 4, Table 2). Our Tat-SF1 construct for ITC experiments included residues 1–360, which supports snRNP co-immunoprecipitation (19) and avoids precipitation of the UHM fragment during titration. Unlike CAPERα (23) but similar to U2AF⁶⁵ (31), the Tat-SF1–SF3b1^ULM isotherms show no evidence of cooperative association. We chose boundaries of peptides corresponding to the SF3b1 ULMs to minimize overlap among consecutive motifs (Fig. 1B). Tat-SF1 detectably binds three of the five documented SF3b1 ULMs (ULM1, ULM4, and ULM5). The binding affinity of Tat-SF1 for ULM5 is >20-fold greater than for ULM4 and slightly greater than ULM1 (4-fold). No heats were detected during titrations of Tat-SF1 with ULM2 or ULM3. The apparent binding affinity of Tat-SF1 for the entire SF3b1^ULM region agrees with the average values of the highest affinity ULM5 and ULM1 sites (Table 2). We conclude that Tat-SF1 is capable of recognizing three local SF3b1 ULM sites, yet has lower affinity for ULM4 compared with either ULM5 or ULM1.

Figure 4.

Tat-SF1 prefers SF3b1 ULM5 by isothermal titration calorimetry analysis of binding. A, apparent equilibrium affinities (K_A) of the indicated SF3b1 ULMs binding Tat-SF1 (residues 1–360). SF3b1^ULM, SF3b1 residues 190–344. Unpaired, two tailed t-tests with Welch's correction: **, p < 0.005; ****, p < 0.0001. B–G, representative isotherms of the titrations.

Table 2.

Thermodynamics of Tat-SF1 binding SF3b1 ULMs

Average values and standard deviations of three independent titrations.

Interaction	KD	ΔH	−TΔSa
	μm	kcal mol−1
Wildtype Tat-SF1titrated with ULM regions
SF3b1ULM	1.8 ± 0.9	−66.1 ± 3.0	58.0 ± 3.0
ULM1	3.0 ± 0.5	−11.8 ± 0.3	4.1 ± 0.3
ULM2b	No detectable binding
ULM3b	No detectable binding
ULM4c	18.9 ± 4.7	−34.1 ± 25.3	27.5 ± 25.5
ULM5	0.8 ± 0.1	−18.7 ± 1.0	10.2 ± 1.2
E286K/D290K mutant Tat-SF1 titrated with ULM5
ULM5b	No detectable binding
E286K/F337A mutant Tat-SF1 titrated with ULM5
ULM5b	No detectable binding

^a Calculated using −TΔS = ΔG-ΔH, with ΔG = −RT ln (K_D⁻¹) and T = 303 K.

^b No heats detected during titration of 200 μm ULM into 20 μm Tat-SF1 protein.

^c Values are estimates due to low binding affinity.

Structures of Tat-SF1 bound to high and low affinity SF3b1 ULMs

To identify and compare the interactions of the Tat-SF1 UHM with high versus low affinity SF3b1 ULMs, we determined the crystal structures of its complexes with ULM4 and ULM5 at 1.9 Å and 2.1 Å resolutions (Fig. 5, Table 1). Seven residues (residues 291–297) of a 17-residue ULM4 peptide were observed bound to a single Tat-SF1 UHM in the crystallographic asymmetric unit (Fig. 5A). Fewer ULM5 residues are evident (four ordered residues) and one of the two Tat-SF1 UHM copies in the asymmetric unit lacks bound ligand (Fig. 5B). This difference is most likely due to the distinct packing environments of the crystal structures.

Figure 5.

Structures of Tat-SF1 UHM bound to SF3b1 ULM. A, feature-enhanced electron density maps (65) (green, 1.5 σ contour level) for SF3b1 ULM4 or B, ULM5 bound to Tat-SF1 UHM (blue). A second Tat-SF1 UHM copy that lacks bound ULM5 (Tat-SF1 UHM) is colored gray. View is rotated 180° relative to the left view of Fig. 2. C, superposed backbone traces of unliganded (magenta) Tat-SF1 UHM with ULM4-bound (blue) or ULM5-bound (yellow) UHM structures. The SF3b1 ULM4 (green) and ULM5 (yellow) are overlaid for reference. Stick diagrams show SF3b1 Trp²⁹³/Trp³³⁸ and compare Tat-SF1 Phe³³⁷ positions.

The core-fold of the apo-Tat-SF1 UHM is similar to its SF3b1 ULM-bound complexes (r.m.s.d. n 0.35–0.45 Å between 86 matching Cα atoms) (Fig. 5C). However, the empty ULM-binding site of the apo-Tat-SF1 structure is filled by a phenylalanine (Phe³³⁷) donated by the so-called “RXF” loop of the UHM (21). In the Tat-SF1–ULM complexes, this Phe³³⁷ side chain rearranges to enclose the consensus ULM tryptophan.

Tat-SF1 displays characteristic features of UHM-type interactions with the SF3b1 ULMs (Fig. 6, A and B). The ULM tryptophan (ULM4 Trp²⁹³ or ULM5 Trp³³⁸) inserts between the α-helices of the Tat-SF1 UHM. There, the ULM tryptophan is sandwiched between a salt bridge (Tat-SF1 Glu²⁹⁴–Arg³³⁵) and a T-shaped aromatic interaction (39) with the Phe³³⁷ side chain. The Tat-SF1-bound ULM4 and ULM5 peptides both adopt a U-shape, which is shared by the two other known structures of SF3b1 ULMs bound to UHMs (ULM5 bound to CAPERα (23) or SPF45 (26)). In contrast, the SF3b1 ULM trajectories differ from that of the SF1 ULM, which binds to the U2AF⁶⁵ UHM in a distinct, linear conformation that is facilitated by a C-terminal coiled-coil extension of the SF1 ULM (24, 25).

Figure 6.

Tat-SF1 UHM-SF3b1 ULM interactions. A, ULM4 and B, ULM5. Hydrogen bonds are indicated by dashed lines. Key interacting residues are labeled; SF3b1 residues are italicized. Packing interactions are indicated by semitransparent surfaces. Isothermal titration calorimetry confirms structure-guided mutations of the Tat-SF1 interface abolish detectable binding to SF3b1 ULMs. Representative isotherms for: C, E286K/D290K or D, E286K/F337A mutant Tat-SF1 (residues 1–360) titrated with SF3b1 ULM5 are shown.

At the C terminus of SF3b1 ULM4, a threonine-proline (TP) motif packs against an exposed Tat-SF1 tryptophan (Trp³³⁶) at the X position of the UHM RXF loop (Fig. 6A). Although the surrounding crystal contacts prevent an analogous interaction of Tat-SF1 Trp³³⁶ with the ULM5 TP motif, the SF3b1 ULM5 and ULM4 share identical C-terminal sequences (Fig. 1C). As such, ULM5 is likely to contact Tat-SF1 Trp³³⁶ in a solution environment. The SF3b1^ULM region contains >20 TP motifs near the ULMs. These SF3b1 sites are highly phosphorylated (phospho.ELM (40)), which could provide means to regulate SF3b1 association with UHM-containing proteins.

Positively-charged residues at the N terminus of each ULM orient toward Glu²⁹⁴, Glu²⁸⁹, Asp²⁹⁰, and Glu²⁸⁶ in an acidic α-helix of the Tat-SF1 UHM (Fig. 6, A and B). Although several N-terminal ULM residues are disordered in the electron density, an ion pair between a conserved ULM arginine (ULM4 Arg²⁹² or ULM5 Arg³³⁷) and Tat-SF1 Glu²⁸⁶ is well-defined. Considering that the Tat-SF1–interacting residues of the SF3b1 ULM3, ULM4, and ULM5 C termini are identical, the different number of basic residues in the N-terminal tails are likely to dominate the range of binding affinities. For example, the different Tat-SF1 binding energies of ULM4 versus ULM5 (ΔΔG 1.9 kcal mol⁻¹) are consistent with the loss of an ion pair between the acidic Tat-SF1 α-helix and SF3b1 ULM5 Lys³³³ (41). A prior study showed that basic residues substituted in the normally acidic U2AF⁶⁵ α-helix eliminated detectable binding to the SF1 ULM in pulldown assays (30). To investigate the importance of corresponding charge–charge interactions for the Tat-SF1–ULM complex, we introduced analogous Tat-SF1 E286K and D290K mutations, and found that these substitutions abolished detectable binding to SF3b1 ULM5 by ITC (Fig. 6C).

Integrity of the Tat-SF1 UHM is critical for co-immunoprecipitation with SF3b1

We evaluated the contribution of UHM–ULM interactions to the association of full-length Tat-SF1 and SF3b1 in human cell extracts (Fig. 7). Plasmids expressing Tat-SF1 variants or N-terminally FLAG-tagged SF3b1 were co-transfected into HEK293T cells. The associated protein complexes were immunoprecipitated using Tat-SF1–specific agarose beads. We found that wildtype (WT) Tat-SF1 retained significant levels of FLAG-SF3b1 in lysates from transfected but not mock-transfected cells. This result confirms the existence of a Tat-SF1–SF3b1 complex, in agreement with prior observations in distinct contexts (17–20). We next compared FLAG-SF3b1 co-immunoprecipitates of a Tat-SF1 variant with E286K/F337A mutations, which prevent detectable binding of Tat-SF1 to SF3b1 ULM5 by ITC (Fig. 6D). The mutant Tat-SF1 variant lacked detectable retention of FLAG-SF3b1, which supports the importance of an intact UHM for association of Tat-SF1 with SF3b1 in human cell extracts.

Figure 7.

Tat-SF1 UHM mediates association with SF3b1 in cell extracts. The amount of transiently expressed FLAG-SF3b1 (143 kDa) retained by immobilized Tat-SF1 proteins (86 kDa; migrates 135 kDa) was detected. Cell lysates (0.5%) were loaded as input controls (left lanes). The sizes of Bio-Rad Precision Plus Dual Color pre-stained markers are indicated (STD). Tat-SF1 (WT) interacts with SF3b1, whereas E286K/F337A mutations (MUT) of the UHM abolished detectable Tat-SF1 association with SF3b1.

Tat-SF1 expression is sensitive to SF3b1 levels

The association of Tat-SF1 with SF3b1 suggested that its expression level would be mutually regulated to ensure formation of a functional macromolecular complex. To test this hypothesis, we used siRNAs to reduce Tat-SF1 or SF3b1 expression levels in HEK293T cells, and evaluated changes in expression of the other factor by RT-qPCR and immunoblotting (Fig. 8, A–D). Tat-SF1 knockdown had little effect on SF3b1 levels, consistent with prior observations (18). Nevertheless, SF3b1 knockdown significantly reduced Tat-SF1 expression levels. We conclude that whereas SF3b1 serves Tat-SF1-independent functions as a major spliceosome subunit (42), a feedback mechanism modulates Tat-SF1 levels in response to SF3b1.

Figure 8.

Tat-SF1 expression levels decline following SF3b1 depletion and in turn decrease expression of similar representative transcripts as SF3b1. A and B, immunoblots demonstrating Tat-SF1 or SF3b1 knockdown following transfection of HEK293T cells with the indicated Stealth^TM siRNAs or a low G/C control siRNA (“low GC”). Whole cell lysates were analyzed by SDS-PAGE and immunoblotted for either: A, Tat-SF1 or B, SF3b1. H2B is encoded by an intronless gene and hence an appropriate loading control. C–H, relative levels of TAT-SF1, SF3B1, or the indicated transcripts were determined by RT-qPCR of total RNA isolated from HEK 293T cells following transfection with the indicated siRNAs. Amplification of the LMNB1 exon 5/6 junction is shown. The LMNB1 exon 7/8 junction gave similar results as shown in Fig. S1. Primer locations and sequences are given in Table S1. Unpaired, two-tailed t tests with Welch's correction: *, p < 0.05; **, p < 0.005; ***, p < 0.0005; ****, p < 0.0001.

Tat-SF1 and SF3b1 regulate overlapping gene transcripts

To further explore the functional outcomes of a Tat-SF1 partnership with SF3b1, we evaluated whether SF3b1 regulates expression of Tat-SF1 responsive genes. We focused on three representative transcripts (CCND1, SIN3B, and LMNB1) that have documented expression changes in response to Tat-SF1 levels (16). As expected, levels of these transcripts decreased following siRNA-mediated reduction of Tat-SF1 (Fig. 8, E–G). Notably, SF3b1 knockdown also decreased the levels of all three Tat-SF1–sensitive transcripts (by 3–5-fold). These effects are likely to be direct, because SF3b1 expression is only subtly affected by Tat-SF1 levels (Fig. 8, B and C) (18). Despite numerous studies (e.g. Refs. 14, 16, and 18), to our knowledge neither we nor others have succeeded in rescuing gene expression by reintroducing Tat-SF1, which suggests that Tat-SF1 activities are delicately balanced by localization and/or post-translational modifications. To reinforce the functional convergence of Tat-SF1 and SF3b1, we compared an additional transcript (DUSP11) that is known to be sensitive to an SF3b1-targeted small molecule, spliceostatin-A (43). Both SF3b1 and Tat-SF1 knockdown decreased levels of the DUSP11 transcript (Fig. 8H).

As a preliminary check for alternative splicing, we compared the sensitivity of a distinct splice site from LMNB1 to Tat-SF1 and SF3b1 knockdown (Fig. S1). We observed similar changes in transcript levels when probing either LMNB1 splice site. Consistently, Tat-SF1 reduction typically leads to intron retention rather than alternatively splicing (17, 18), and most intron-retained transcripts would be removed by quality control pathways (44). As such, our experiments lack discrimination between altered transcription and splicing. Regardless, the significant impact of SF3b1 knockdown for all Tat-SF1–sensitive transcripts supports altered pre-mRNA splicing, considering the well-established roles of SF3b1 in the spliceosome (42). Altogether, our results establish convergent roles for Tat-SF1 and SF3b1 in human gene expression.

Discussion

Here, we identify direct and specific coupling of the Tat-SF1 transcriptional and pre-mRNA splicing factor with the SF3b1 subunit of the U2 snRNP. Underscoring the importance of the Tat-SF1–SF3b1 complex, associated snRNPs are required for Tat-SF1 to stimulate transcription (19). Following clues from structural homology, we pinpoint a critical UHM–ULM interface for Tat-SF1–SF3b1 association in cell extracts or purified proteins. Our high resolution structures reveal that the Tat-SF1 UHM exhibits a characteristic RRM-like topology with an extended C terminus that masks degenerate ribonucleoprotein motifs. The Tat-SF1 UHM encloses a central tryptophan of the SF3b1 ULM between two α-helices, which is typically essential for UHM–ULM binding (e.g. Refs. 29–31 and 36). Tat-SF1 and SF3b1 association further depends on electrostatic interactions between an acidic UHM α-helix and a basic ULM tail.

The Tat-SF1 UHM–SF3b1 ULM structures expand panoptic views of this protein interaction motif (21). A recurrent theme of tryptophan-mediated and charge–charge interactions underscores the question of how cognate UHM–ULM pairs are determined. Although we show that the Tat-SF1 UHM is required for detectable SF3b1 association, it remains likely that auxiliary interfaces mediated by distinct protein or RNA regions could enhance the binding specificity and affinity of the Tat-SF1–SF3b1 complex in a physiological context. The binding affinities of the Tat-SF1–SF3b1 protein domains are micromolar (Table 2), consistent with other SF3b1 ULM–UHM protein complexes (21). Such relatively weak interactions could be important for dynamic associations (45), yet do not rule out enhancement by transient interfaces during spliceosome assembly. For example, the UHM-mediated association of Tat-SF1 with the SF3b1 ULMs may synergize with the U2 snRNA, which Tat-SF1 and SF3b1 bind in the context of the intact U2 snRNP. Indeed, an RRM mutation of the yeast homologue Cus2 prevents its U2 snRNA association and rescue of U2 snRNA misfolding (9, 19).

Macromolecular complexes such as spliceosome particles often employ feedback mechanisms to coordinate subunit expression (e.g. Refs. 46–48). Our finding that Tat-SF1 expression decreases following SF3b1 depletion (Fig. 8, A and D) supports functional coordination of Tat-SF1 and SF3b1. Conversely, SF3b1 levels are relatively insensitive to reduced Tat-SF1, which agrees with SF3b1 ULM associations with a variety of alternative splicing factors (23, 26, 27, 31). The SF3b1-dependent Tat-SF1 expression is analogous to a previously characterized precedent, the U2AF heterodimer (46). U2AF⁶⁵ knockdown reduces U2AF³⁵ protein levels, whereas reduced U2AF³⁵ levels are insensitive to U2AF⁶⁵. Functionally, in vitro splicing can be reconstituted by the U2AF⁶⁵ subunit alone (49–51), and moreover, Schizosaccharomyces pombe introns that are affected by U2AF⁶⁵ depletion are affected by U2AF³⁵ depletion (52). Accordingly, we here find that the levels of representative downstream transcripts respond to both Tat-SF1 and SF3b1 reductions. These effects are likely to be specific, because SF3b1 (43, 53, 54) and Tat-SF1 (16, 17, 55) knockdowns alter expression and splicing of only a subset of genes. We suggest that a feedback mechanism promotes the formation of Tat-SF1 complexes with SF3b1.

Altogether, we have identified SF3b1 as a directly interacting partner of Tat-SF1, and detailed a requisite ULM–UHM interface of this functional complex at atomic resolution. Beyond our study of Tat-SF1, the SF3b1 ULM-binding preferences have been comprehensively determined for one other UHM-containing protein, U2AF⁶⁵ (31). Both Tat-SF1 and U2AF⁶⁵ bind the SF3b1 ULM region with similar apparent affinities (approximate K_D values 2 μm/3 μm for Tat-SF1/U2AF⁶⁵ binding to SF3b1^ULM), and both show a subtle binding preference for ULM5 among the SF3b1 ULMs. Immunoaffinity purification followed by MS has shown that Tat-SF1 associates with U2 snRNP particles, but is absent from assembled, “A”-stage spliceosomes (20). In contrast, U2AF⁶⁵ is an essential splicing factor that stabilizes U2 snRNP association with pre-mRNA splice sites (56, 57), and is detected only in the A- and later “B”-stages of spliceosome assembly (20). The similar SF3b1 binding preferences of Tat-SF1 and U2AF⁶⁵, coupled with sequential Tat-SF1 and U2AF⁶⁵ functions during transcription and spliceosome assembly, call for future studies to explore whether Tat-SF1 serves as a checkpoint factor for handoff of functional U2 snRNP to U2AF⁶⁵ during spliceosome assembly.

Experimental procedures

Recombinant protein expression and purification

Human Tat-SF1 (residues 1–360 of NCBI Refseq NP_055315), Tat-SF1 UHM (residues 260–353), SF3b1^ULM (residues 190–344 of NCBI Refseq NP_036565), U2AF65 (residues 85–475 of NCBI Refseq NP_009210), or SF1 (residues 1–255 of NCBI RefSeq NP_004621) were expressed in Escherichia coli as GST fusions using either pGEX-6p or an in-house pGEX-6p variant encoding a tobacco etch virus protease site. The splicing factor proteins were purified by GST affinity followed by anion exchange or heparin affinity chromatography. For crystallization experiments, the Tat-SF1 UHM was separated from aggregates by size exclusion chromatography in 50 mm NaCl, 15 mm HEPES, pH 7.4, 0.2 mm tris(2-carboxyethyl)phosphine. Purified SF1 proteins were phosphorylated by UHMK1 as described (25) and then separated from UHMK1 by size exclusion chromatography. Synthetic peptides corresponding to SF3b1 ULMs (with residue ranges shown in Fig. 1) were purchased at 98% purity (Biomatik or Bio-Synthesis, Inc.).

GST pulldown assays

The GST fusion “bait” protein (100 μg) was incubated for 1 h on pre-packaged GSH-agarose spin columns with 0.2-ml bed volume (Thermo Scientific Pierce^TM) that were equilibrated three times with binding buffer (100 mm NaCl, 25 mm HEPES, pH 7.4, 3% glycerol, 1 mm EDTA, 10 mm DTT), and supplemented with protease inhibitor. The immobilized GST fusions were incubated with purified protein (500 μg, 1 mg/ml) in binding buffer for 45 min at 4 °C. The resin was then washed five times with binding buffer. Retained proteins were eluted by heating in Laemmli loading buffer supplemented with 10 mm GSH, and analyzed by SDS-PAGE with Coomassie® Blue staining.

Isothermal titration calorimetry

Protein samples were dialyzed into 25 mm HEPES, pH 7.4, 50 mm NaCl, 0.2 mm tris(2-carboxyethyl)phosphine, and peptides were diluted into >200-fold greater volume of dialysis buffer, prior to ITC. The thermodynamics of Tat-SF1 titrated with SF3b1^ULM or ULM peptides were measured using a VP-ITC and fit using Origin 7.0 (MicroCal, LLC). Control titrations of ULM peptides into buffer were linear and showed negligible heats of dilution, and were subtracted to correct the relatively low affinity isotherms of ULM1 and ULM4. Other isotherms were corrected for the heats of dilution using an average of three data points at the saturated plateau of the titration. The apparent stoichiometries are consistent with 1:1 binding except SF3b1^ULM, which is underdetermined and appears to bind between two and five Tat-SF1 molecules considering χ² values of different fits.

Crystallization and structure determination

For crystallization of the Tat-SF1 UHM–SF3b155 ULM4 and ULM5 complexes, 0.1 mm Tat-SF1 UHM was mixed with a 1.2-fold molar excess of SF3b155 ULM and concentrated in a 2-kDa MWCO Vivaspin® Hydrosart (Sartorius) concentrator. The Tat-SF1 UHM and its ULM complexes were crystallized by vapor diffusion using the JCSG+^TM screen (Molecular Dimensions) at 277 K. The apo-Tat-SF1 UHM crystallization conditions were optimized in hanging drop format to 0.4 m MgCl₂, 0.1 m Tris, pH 7.5, 15% PEG 8000. The respective crystallization solutions of ULM4 and ULM5 complexes contained 0.2 m magnesium formate, 20% PEG 3350, and 0.1 m Tris, pH 8.5, 15% PEG 8000. The crystals were sequentially transferred to 20% glycerol in synthetic mother liquor then flash-cooled in liquid nitrogen. Crystallographic data sets were collected remotely at Stanford Synchrotron Radiation Lightsource beamlines 12-2, 7-1, and 9-2 (58). Data were processed using the SSRL AUTOXDS script (A. Gonzalez and Y. Tsai) implementation of XDS (59) and CCP4 packages (60).

The apo-Tat-SF1 UHM structure was determined with an ensemble search model comprising the U2AF⁶⁵, SPF45, and CAPERα UHM structures generated using Ensembler (61). A molecular replacement solution using Phaser (61) via the Phenix interface (62) (TF Z-score 5.8, LLG 304) was rebuilt in Coot (63), and refined using Phenix (62). Alternative strategies to determine the apo-Tat-SF1 UHM structure were unsuccessful, including direct methods at 1.13 Å resolution and molecular replacement searches with either individual UHM crystal structures stripped of ligand coordinates (U2AF⁶⁵, PDB ID 4FXW; SPF45, PDB ID 2PEH; CAPERα, PDB ID 4OZ0) or an unpublished apo-NMR structure (PDB ID 2DIT). The ULM-bound structures were determined by molecular replacement starting from the apo-Tat-SF1 UHM structure. Crystallographic data and refinement statistics are given in Table 1.

Cell culture and transfections

HEK 293T cells (ATCC® CRL-3216^TM) were cultured in Gibco^TM minimum essential media supplemented with 10% fetal calf serum (Atlanta Biologicals) and 2 mm l-glutamine and maintained at 37 °C in a humidified atmosphere containing 5% CO₂. Cells were transfected using JetPrime® (Polyplus-transfection®) as instructed by the manufacturer.

Co-immunoprecipitation

HEK 293T cells were transfected with plasmids encoding human Tat-SF1 (modified from Origene #SC114990), N-terminally FLAG-tagged SF3b1 (modified from a gift of Benjamin Ebert (64)), or empty control vector (pCMV). After 24 h, cells were harvested and lysed in a IP buffer (50 mm Tris, pH 8.0, 75 mm NaCl, 5% glycerol, 10 mm CaCl₂, cOmplete^TM EDTA-free protease inhibitor (Sigma), 0.5 mm DTT) + 0.5% Triton X-100. Lysates were clarified by centrifugation and a fraction of each sample was set aside as input controls. The remaining lysates were diluted 1:5 with IP buffer + 0.1% Triton X-100, then incubated for 1 h on a rocking platform at 4 °C protein G-Sepharose (GE Healthcare number 17061801) pre-bound to Tat-SF1-specific antibody (Santa Cruz Biotechnology number sc-514351). Beads were collected by centrifugation and washed extensively with IP buffer. Final washes included RNase A (Sigma).

Immunoblotting

The immunoprecipitated complexes on beads were resuspended in SDS-PAGE loading buffer. For total protein analysis, cells were lysed in a buffer containing 50 mm Tris, pH 8.0, 10 mm EDTA, 1% SDS, 1 mm DTT, and protease inhibitors. Proteins were separated by SDS-PAGE, transferred to a polyvinylidene difluoride membrane, and immunoblotted with antibodies specific for Tat-SF1 (Santa Cruz Biotechnology number sc-514351), FLAG (Cell Signaling Technology number 2368), SF3b1 (MBL number D221-3), or Histone H2B (Cell Signaling Technology number 12354). Secondary antibodies included anti-mouse IgG horseradish peroxidase or anti-rabbit IgG horseradish peroxidase (GE Healthcare numbers NA931 or NA934). Chemiluminescent signal from SuperSignal WestPico chemiluminescent substrate (ThermoFisher Scientific) was detected on a Chemidoc^TM Touch Imaging System (Bio-Rad).

Quantitative-reverse transcriptase real time PCR

Cells were harvested 3 days following transfection with Stealth^TM siRNA duplexes (ThermoFisher Scientific) specific for HTATSF1 (assay IDs: HSS120650, HSS120651), SF3B1 (assay IDs: HSS146413, HSS146415), or a negative control (12935200). Total RNA was isolated and DNase I-treated using the RNeasy Kit (Qiagen). The cDNAs were synthesized by Moloney murine leukemia virus RT with random primers (Invitrogen), and analyzed by quantitative PCR with SYBR^TM Green in triplicate on a Bio-Rad CFX thermal cycler. The RT-qPCR products were quantified by the relative standard curve method and normalized to the level of 18S rRNA. Results in Figs. 7 and 8 represent at least three biological replicates. Primers are listed in Table S1.

Author contributions

S. L. and C. L. K. conceptualization; S. L., J. L. J., M. P., and C. L. K. validation; S. L., J. R. L., S. W. H., D. M., J. L. J., M. P., and C. L. K. investigation; S. L. and C. L. K. writing-original draft; S. L., D. M., J. L. J., M. P., and C. L. K. writing-review and editing; J. L. J., M. P., and C. L. K. methodology; C. L. K. supervision; C. L. K. funding acquisition.