INTRODUCTION
Splicing of precursor messenger RNAs (pre-mRNAs) stands at the core of eukaryotic gene expression, responsible for the removal of introns to generate mature mRNAs
1-4. This process is catalyzed by the spliceosome, a dynamic ribonucleoprotein complex that must cycle through precise assembly, activation, catalysis, and disassembly
5. Most multicellular eukaryotes employ two parallel splicing systems
6: the U2-type major spliceosome, which processes the vast majority of introns, and the U12-type minor spliceosome, which excises a rare but evolutionarily conserved class of introns with the characteristic 5′ splice site (5′SS) and branch point sequence (BPS), but without a defined polypyrimidine tract upstream of the 3′ splice site (3′SS)
7,8. The evolutionary maintenance of two distinct splicing machineries must serve a non-redundant biological purpose. This rationale is evident in the specialized regulatory role of minor splicing: U12-type introns are highly enriched in essential genes and function as solitary, slow-processing units within transcripts
6,9. Their delayed excision establishes a conserved rate-limiting checkpoint
10, potentially adding a critical layer of regulation to the expression of vital genes.
The minor spliceosome comprises five small nuclear ribonucleoprotein particles (snRNPs), four of which (U11, U12, U4atac, and U6atac) are unique to this pathway
11-13. Its assembly initiates with recognition of 5′SS and BPS by the U11/U12 di-snRNP, followed by recruitment of the pre-assembled U4atac/U6atac•U5 tri-snRNP
14. Despite this established framework and its physiological importance, the molecular mechanism of the minor spliceosome assembly and catalysis has remained largely unknown. Studies have long depended on genetic and bioinformatic analyses of their substrate introns, which have successfully mapped related genes and disease links
15-17. However, these indirect approaches provide limited information about the inner workings of the spliceosomal machine itself. A major obstacle to direct biochemical dissection is its extreme scarcity
8,14: minor snRNPs are present at merely 1% of the level of their major counterparts, with U6atac being even rarer
18. This severe low abundance has historically precluded biochemical isolation and functional reconstitution, leaving fundamental questions about its assembly, catalytic mechanism, and regulation unresolved. Therefore, direct insight into its operational dynamics is essential not only to elucidate its mechanism but also to define the functional niche that ensures its evolutionary persistence alongside the major spliceosome.
Early efforts resulted in the isolation of the U11/U12 di-snRNP
19, which likely represents the most stable and abundant subcomplex. More recently, the optimization of U12-type
in vitro splicing systems enabled the visualization of the assembly intermediates, including the minor pre-B and B
act complexes
20,21, thereby clarifying early steps in the minor splicing pathway coupled with other genetic and structural advances
22-26. However, these advances stop at the threshold of catalysis. The complexes that execute the two-step transesterification reaction, the chemical core of splicing, have remained entirely uncharacterized. This fundamental gap has precluded any mechanistic insight into critical features, including catalytic metal coordination, active-site configuration, substrate delivery, and step-specific control by protein factors, representing the central unresolved problem to completely understand minor splicing.
In this study, we employed a markedly improved U12-type in vitro splicing system to capture the human minor spliceosome in both catalytic states, the branching-completed C complex and the exon-ligation-ready C* complex, and determine their cryo-electron microscopy (cryo-EM) structures at 2.9 Å and 3.0 Å resolution, respectively. These high-resolution views enable us to directly visualize the catalytic core of the minor spliceosome, offering unprecedented biochemical insight into its unique principles for catalyzing two transesterification reactions. Our findings elucidate the molecular logic underlying U12-type splicing and reveal how conserved splicing principles are uniquely adapted in this parallel eukaryotic splicing machinery, thereby resolving a central and long-standing question in RNA splicing.
RESULTS
Capturing the minor spliceosome during catalysis
We previously assembled the human minor pre-B and B
act complexes using the MINX-U12 pre-mRNA
20,21. However, all attempts to assemble sufficient catalytic-state minor spliceosomes using this pre-mRNA failed, presumably due to its low splicing efficiency
in vitro. To overcome this obstacle, we optimized the substrate through parallel screening of pre-mRNA scaffolds and splicing regulatory elements (Fig. 1a; Supplementary Fig. S1a, b). We first refined the core intronic sequences (mainly BPS and 3′SS). Subsequently, replacement of the 3′ exon with that of
SCN4A and incorporation of the ENH1 enhancer were found to boost the U12-type splicing efficiency (Supplementary Fig. S1c, d). These elements thus appear to function as
cis-regulatory elements that stimulate U12-type splicing. Combining these modifications yielded the high-performance substrate MSE-U12 (Fig. 1a, b; Supplementary Fig. S1d).
Using the helicase-deficient PRP16 mutant (K561A), which is known to arrest the major spliceosome after the first step of catalysis
27, we sought to capture the human minor C complex using MSE-U12 pre-mRNA (Supplementary Fig. S2a). The purified sample contained U12, U6atac, U5 snRNAs, and the lariat-3′ exon intermediate; the particles appeared intact on the cryo-EM micrograph (Supplementary Fig. S2b, c). In parallel, we attempted to trap the human minor C* complex by lowering the pH of the splicing reaction
28 (Supplementary Fig. S2d–f). Cryo-EM analysis of both samples yielded high-resolution reconstructions (2.9 Å for C complex; 3.0 Å for C* complex), affording an atomic-level view of most regions especially for the catalytic core (Supplementary Figs. S3–S8). Furthermore, the application of local refinement strategies improved maps in the most distal areas with resolvable secondary structure features (Fig. 1c, d; Supplementary Figs. S3 and S4). Aided by mass spectrometry (MS) analysis, these data enabled the construction of high-fidelity atomic models, thereby providing a comprehensive structural framework for elucidating the catalysis of the minor spliceosome (Supplementary Tables S1–S5).
Structures of the human minor C and C* complexes
The structure of the human minor C complex contains 49 proteins, three snRNAs (U12, U6atac, and U5), and two MSE-U12 fragments (5′ exon and lariat-3′ exon intermediate) (Fig. 1c). The protein components include those from U5 and U12 snRNPs, NineTeen Complex (NTC), NTC related (NTR), the intron binding complex (IBC), the exon junction complex (EJC), and 13 splicing factors. Notably, MMTAG2, a protein linked to multiple myeloma
29 but with an unknown role in splicing, is identified here as a step-I factor in minor splicing. The DEAH-box ATPase/helicase PRP16 is unambiguously resolved and is located close to the intron sequences downstream of the BPS. Unexpectedly, the step-II factor SLU7 is already recruited in this state. In contrast to the major C complex
27, CIR1, which tethers U2 snRNP to PRP8 RNaseH, is absent here.
The structure of the minor C* complex comprises 59 proteins, three snRNAs, a free 5′ exon, and a lariat-3′ exon intermediate (Fig. 1d). Compared to the minor C complex, six step-I factors dissociate, and nine proteins are recruited, including the ATPase/helicase PRP22. Among the newly incorporated components, we identify three previously mechanistically undefined proteins, WDR25, FAM204A, and RBM41, as U12-specific step-II splicing factors. Their recruitment coincides with large-scale, PRP16-driven rearrangements of peripheral modules that collectively prime the complex for exon ligation. Surprisingly, we observe the stable anchoring of the heptameric LSm2-8 complex between the PRP19 core and Aquarius, where it protects the U6atac 3′ end. This specific interaction is absent during catalysis in the major spliceosome
30,31. Despite these changes, core modules including U5 snRNP, NTC, NTR, EJC, and four splicing factors remain structurally invariant. In contrast to the major C* complex, the minor C* complex lacks seven step-II factors essential for the major spliceosome catalysis: PRP17, SDE2, PRKRIP1, FAM50A, CXORF56, NOSIP, and TLS1
32.
Together, the distinct composition and architecture of the minor spliceosome reflect unique strategies for catalytic center organization and substrate handling, which establish a specialized mechanistic framework for its splicing pathway.
Catalytic mechanism of U12-type splicing
The atomic models of the minor C and C* complexes provide a near-complete visualization of the RNA conformational dynamics and reveal the unique features that govern both steps of U12-type splicing. The catalytic step-I conformation is assembled by U12, U6atac, and U5 snRNAs, together with the 5′ exon and the intron of MSE-U12 (Fig. 2a; Supplementary Fig. S5a and Table S4). The 5′SS is recognized via duplex formation between the intron sequences downstream of the AU dinucleotide and a specific region of U6atac snRNA, the A12AGGA16 box and flanking nucleotides, termed 5′SS-pairing site (5′SS-PS) (Fig. 2a, b). Concurrently, the BPS forms an extended duplex with U12 snRNA, with the invariant branch point adenosine (BP-A) flipped out. The 5′ exon is anchored by loop I of U5 snRNA. The covalent linkage between the 2′-oxygen of BP-A and the phosphorus of nucleotide A1 at the 5′SS confirms the completion of the branching reaction to form the lariat (Fig. 2b).
Within this overall scaffold, the active site harbors a fully assembled heteronuclear metal-ion core organized by the U6atac intramolecular stem loop (ISL) by helix I of the U12/U6atac duplex (Fig. 2b; Supplementary Fig. S5b, c). A defining feature is the stable presence of the catalytic metal M2, which activates the 2′-OH nucleophile of BP-A, in contrast to its absence in the major C complex
27,34 (Supplementary Fig. S5d). M2 is coordinated by four phosphate groups from U6atac snRNA and A1 of 5′SS. Notably, following the completed branching reaction, the nucleophile no longer coordinates M2, as indicated by their separation of ~2.5 Å (Fig. 2c). The 3′-terminal guanine (G-1) of 5′ exon undergoes a configurational flip, displacing its newly liberated 3′-OH group (
i.e. the leaving group) ~10 Å from the A1 phosphate and away from the metal M1, which stabilizes the leaving group during branching (Fig. 2c). A putative potassium ion, coordinated by U6atac snRNA in a position analogous to the K1 site in group II introns
35, likely stabilizes the two catalytic metals (Supplementary Fig. S5c).
Transition to the minor C* complex involves significant structural resolving, including the previously unassigned regions downstream of the U6atac central stem loop (nucleotides 66–86, 91–101, and 104–125) (Fig. 2d; Supplementary Fig. S5e and Table S5). This is accompanied by an ~80° rotation of the entire BPS/U12 duplex, reconfiguring the active site for the second-step catalysis (Fig. 2e). The 2′-5′ linkage is displaced outward, creating space for the 3′SS and the 3′ exon, the first two nucleotides of which become visible (Fig. 2f; Supplementary Fig. S5f).
Within the remodeled active site, metal M2, which stabilizes the leaving group during exon ligation, adopts a distinctive coordination involving the phosphate of the first 3′ exon nucleotide and three phosphates from U6atac, contrasting with the simpler coordination seen in the major C* complex
32,36 (Fig. 2g; Supplementary Fig. S5g). Simultaneously, the nucleophile of exon ligation (
i.e. 3′-OH of the 5′ exon) is repositioned to coordinate M1 and resides ~3.3 Å from the scissile phosphate, priming the complex for the second transesterification.
MMTAG2 stabilizes the branching conformation
During branching, we identify the conserved nuclear protein MMTAG2
29, previously unrecognized in splicing, as an essential structural component that stabilizes the active site conformation (Fig. 3a; Supplementary Fig. S6a). Notably,
MMTAG2 was originally identified as a candidate oncogene in multiple myeloma
29,37,38, suggesting a potential molecular link between the regulation of minor splicing and tumorigenesis. In the structure of the minor C complex, residues 6–83 of MMTAG2 are unambiguously resolved in an extended conformation that integrates deeply into the step-I assembly, where it directly binds to PRP8 RNaseH, CWC25, CWC22, and the intron sequences immediately downstream of the BPS (Fig. 3a, b; Supplementary Fig. S6b, c).
Structurally, the N-terminal segment (residues 9–13) of MMTAG2 forms a β-strand that pairs up with two β-strands of PRP8 RNaseH to create a hybrid β-sheet, a mechanism that contributes to locking the BPS/U12 duplex into its branching conformation by stabilizing PRP8 RNaseH together with FRG1 (Fig. 3b, c; Supplementary Fig. S6d). Downstream of this β-strand, MMTAG2 and PRP8 RNaseH jointly form a positively charged surface that lines the intron exit channel (Fig. 3d). MMTAG2 specifically recognizes the intron through hydrogen bonds (H-bonds) from Asn33 and Arg27 to the nucleobase of U210, while the guanidinium group of Arg27 forms a cation–π interaction with U211, collectively maintaining the local intron conformation (Fig. 3e). Further stability is provided by a C-terminal α-helix (residues 63–82) of MMTAG2, which packs against the central helix of CWC25, with this helical unit anchored by the MA3 domain of CWC22 (Supplementary Fig. S6c).
Through these multifaceted interactions, MMTAG2 is positioned to stabilize both the BPS/U12 duplex and the surrounding protein network of the active site during branching. While conserved step-I factors such as YJU2, CWC25, and ISY1 stabilize core RNA elements in a manner analogous to the major spliceosome
27,39(Fig. 3f; Supplementary Fig. S6e–g), MMTAG2 appears to provide a distinct, pathway-specific stabilization critical for U12-type branching. Intriguingly, the step-II factor SLU7 is already recruited in this complex
via its zinc-finger domain bound to the PRP8 N-domain (Fig. 3a; Supplementary Fig. S6h). Thus, MMTAG2 emerges as a dedicated architectural factor for the minor spliceosome catalysis and a compelling molecular candidate linking U12-type splicing to oncogenic processes.
WDR25 and FAM204A stabilize the exon-ligation conformation
For exon ligation, the minor spliceosome is stabilized by a distinct set of proteins, which includes three newly structurally characterized U12-specific factors, WDR25, FAM204A, and RBM41, in the structure of the minor C* complex (Fig. 4a). Among them, WDR25 and RBM41 had been previously implicated in minor splicing
40, whereas FAM204A was uncharacterized. Notably, database analyses indicate both WDR25 and FAM204A as prognostic markers in multiple carcinomas
41, underscoring their potential clinical relevance and implicating a broader involvement of minor splicing in oncogenic pathways. WDR25 and FAM204A spatially replace and functionally mimic the major step-II factors PRP17/CDC40 and SDE2, respectively
32. Occupying a substantial region within the minor C* complex, they contact PRP8, Cactin, CDC5L, SYF1, and SYF3, cooperatively locking the BPS/U12 duplex in the exon ligation conformation (Fig. 4b).
FAM204A adopts an extended conformation in its C-terminal half (residues 120–233), which is clearly resolved in our structure (Supplementary Fig. S7a, b). A coiled-coil segment and flanking helices interact with SYF1 C-terminus (Supplementary Fig. S7c), while the C-terminal tail projects deep into the splicing active site (Fig. 4b; Supplementary Fig. S7d). WDR25 contains a C-terminal WD40 domain and an N-terminal fragment (residues 172–182, referred to as N-loop), which is anchored to ARMC7 through conserved, specific interactions (Fig. 4a; Supplementary Fig. S8a–d). Despite limited sequence identity, the WD40 domains of WDR25 exhibit high structural similarity to those of PRP17 and occupy equivalent positions in the C* complexes of both spliceosomes
32,36 (Fig. 4b; Supplementary Fig. S8e, f).
WDR25 contacts the tip of the BPS/U12 duplex and, together with the C-terminal region of FAM204A, forms a positively charged surface that stabilizes the duplex (Fig. 4c). Four conserved arginine residues in WDR25 (Arg460/Arg463/Arg464/Arg465) mediate H-bonds to the backbones of both U12 snRNA and the intron, while Arg503 interacts directly with the intronic nucleobases of G188 and A189 (Fig. 4d; Supplementary Fig. S9). A lysine-rich stretch in FAM204A further donates H-bonds to backbone phosphates of the intron and U12 snRNA (Fig. 4d). Thus, the WDR25–FAM204A pair collectively stabilizes the BPS/U12 duplex, facilitating the 3′SS docking during U12-type exon ligation.
Beyond the duplex stabilization, WDR25 and FAM204A also directly maintain the splicing active site during exon ligation (Fig. 4e). The two proteins sandwich and stabilize the β-finger of PRP8, which together with the 1585-loop (also known as α-finger
42) clamps the lariat junction (Fig. 4e, f). In addition, the C-terminus of FAM204A interacts with the 1585-loop and intercalates between helix Ia of the U12/U6atac duplex and the BPS/U12 duplex (Fig. 4e, g).
In summary, WDR25 and FAM204A represent key structural determinants that ensure the fidelity of U12-specific exon ligation. Their established role as cancer prognostic markers, coupled with their essential function in stabilizing the catalytic step-II conformation of the minor spliceosome, positions them as promising molecular links between splicing regulation and disease mechanisms, including tumorigenesis.
Conserved mechanism of association and activation for PRP16 and PRP22
Our structures of the minor C and C* complexes enable the unambiguous assignment of the DEAH-box helicases PRP16 and PRP22 to their respective catalytic states (Fig. 5; Supplementary Fig. S10a–e). Both helicases are peripherally positioned along the pre-mRNA 3′ exit channel and resemble the association mode of PRP2 in the minor B
act complex
21 (Fig. 5a, b; Supplementary Fig. S10f).
In analogy to PRP2 that engages the spliceosomal RNP core via its C-terminal domain (CTD, binding SF3B1/SF3B3) and RecA2 (contacting BUD13), PRP16 employs its CTD to interact with the PRP8 core domain and YJU2, while its RecA2 binds CWC25 in the minor C complex (Fig. 5a, c; Supplementary Fig. S10c). In the minor C* complex, PRP22 similarly uses its CTD to bind the PRP8 core, but its RecA2 contacts the N-terminal domain of RBM41, which is clearly resolved in our structure and, in turn, bridges PRP8 RNaseH and the step-II factor SLU7 (Fig. 5b, d). Moreover, the C-terminus of PRP22 extends along the lateral surface of PRP8 core, placing it adjacent to the 1585-loop that directly anchors the 3′SS in the active site (Fig. 5b, d; Supplementary Fig. S10e).
A key mechanistic divergence lies in helicase activation. Unlike PRP2, which strictly requires the G-patch coactivator GPKOW (Spp2 in yeast) that simultaneously tethers the RNP core and PRP2
43 (Supplementary Fig. S10f, g), PRP16 and PRP22 function without one. The structural basis for this coactivator independence has remained unclear, largely due to the poorly resolved N-terminal extensions in previous studies
27,32,34,44-46. Our structures now reveal that PRP16 and PRP22 operate
via a built-in, coactivator-independent mechanism.
This intrinsic mechanism is mediated by an N-terminal extension harboring two conserved motifs: N-fragment I and N-fragment II. N-fragment I, comprising an α-helix followed by a flexible loop, anchors to the RNP core; N-fragment II, containing two α-helices and an extended intervening loop, simultaneously engages the helicase’s own CTD and RecA2 domain (Fig. 5a–d; Supplementary Fig. S10b, e, g). This binding mode mirrors the tethering function of GPKOW/Spp2 but is encoded entirely within the helicase itself. The sequences of N-fragment II are conserved between PRP16 and PRP22, with conserved residues contacting both the CTD and RecA2 domains (Fig. 5e). Most strikingly, the binding site for N-fragment II on PRP16 or PRP22 precisely overlaps with the site occupied by GPKOW/Spp2 on PRP2, explaining why no external G-patch protein is required.
Taken together, our structural and mechanistic analysis reveals that PRP16 and PRP22 have evolved an integrated regulatory architecture that bypasses the need for a separate G-patch coactivator. Concurrently, the recruitment of PRP22 is specifically assisted by RBM41, which physically links the helicase to the SLU7-dependent step-II machinery, highlighting a unique mechanism for coordinating helicase action with catalytic progression in the minor spliceosome.
DISCUSSION
Specified proteins orchestrate a conserved catalytic core
Our structures resolve the molecular basis of both catalytic steps in U12-type splicing, revealing that the reactions are driven by a conserved catalytic core: a two-metal-ion center organized by the U6atac ISL and the U12/U6atac helix I (Fig. 2), which together adopt a near-identical conformation to the corresponding U6 ISL and U2/U6 helix I in the major spliceosome. Meanwhile, despite the RNA-based catalytic core, the newly identified pathway-specific proteins add distinct layers of stabilization: MMTAG2 adopts an extended conformation to support the active site during branching (Fig. 3), whereas WDR25 and FAM204A directly position and stabilize the BPS/U12 duplex during exon ligation (Fig. 4).
Extending this principle of specific regulation, our structures uncover how the previously functionally uncharacterized protein RBM41 mediates the recruitment of the helicase PRP22 to the catalytic minor spliceosome. RBM41 utilizes its N-terminal domain, which we show is composed of seven α-helices, to directly interacts with PRP22, consistent with a previous study
47, while simultaneously contacting BRR2, Cactin, SLU7, and PRP8 RNaseH (Supplementary Fig. S11). Through an extended structural module formed by helices α1/α2/α6/α7 and associated β-strands, RBM41 effectively bridges PRP22 with the step-II machinery (Supplementary Fig. S11c–e), thereby integrating helicase activity into the exon-ligation network. The C-terminal RRM of RBM41, reported to bind U12 snRNA 3′ stem loop
47, is not visible in our structure, nor is the U12 snRNA 3′ region.
Together with the minor B
act structure
21, our current study also outlines a coherent remodeling pathway of protein displacement and recruitment that drives the U12-type catalytic progression (Supplementary Fig. S12). During the B
act-to-B* transition, PRP2 and its cofactor GPKOW mediate the dissociation of the SF3b complex, SCNM1, and a set of splicing factors, thereby freeing the 5'SS and the BPS/U12 duplex. Subsequent rearrangements, along with the recruitment of six shared splicing factors and the unique component MMTAG2, translocate the BPS/U12 duplex into the vicinity of the splicing active site, enabling branching. Following PRP16 action, the step-I factors are dissociated, making way for nine step-II factors. Among these, RBM41, WDR25, and FAM204A are specific to the minor spliceosome, whereas several step-II factors essential in the major spliceosome (including PRP17, SDE2, PRKRIP1, FAM50A, CXORF56, NOSIP, and TLS1) are absent, highlighting the compositional specialization of the minor spliceosome.
U6atac snRNA as a central organizer for U12-type catalysis
Structural analysis reveals that the unique protein composition of the minor spliceosome is likely governed by the catalytic component U6atac snRNA. Unlike U6 snRNA, U6atac lacks a 5′ stem loop and does not form helix II, two hallmarks of the major spliceosome catalytic center
32,48 (Fig. 6a). Instead, the sequences downstream of ISL fold into a central stem loop, a single-stranded segment that binds along the positively charged surface of SYF3, and a 3′ stem loop anchored by SYF1 (Supplementary Fig. S13a, b). Notably, the RNA segment emerging from the 3′ stem loop enters the heptameric LSm2-8 complex (Fig. 6b, c; Supplementary Fig. S5e). In contrast to the major spliceosome, where the LSm2-8 complex dissociates during activation
30, it remains stably bound in the minor spliceosome through interactions with PRP19 core, SYF1, and Aquarius (Supplementary Fig. S13c).
These distinct structural features create a binding landscape that selectively recruits minor-specific factors. For example, in the major C* complex, PRP17 anchors its N-terminus to the 5′ stem loop of U6 snRNA and the NTR components RBM22 and BUD31, whereas its functional partner SDE2 is partly stabilized by SYF2, which binds to U2/U6 helix II (Fig. 6d, e). In the minor C* complex, the 5′ end sequences of U6atac snRNA are encapsulated by the RBM48–ARMC7 complex (Fig. 6b, c), which occludes the binding site for the N-domain of PRP17 but provides a binding interface for the N-loop of WDR25. Overall, of the nine unique U12-type splicing factors (excluding previously characterized U11/U12 di-snRNP proteins), five are directly linked to the specific conformation of U6atac snRNA, accounting for over half of these specialized components. Therefore, U6atac snRNA serves as a central organizer for the minor spliceosome.
Architectural insights define a regulatory hub in development
Moving beyond catalysis, our structural elucidation reveals that the minor spliceosome emerges not only as a parallel catalytic machine but also as a critical regulatory node in gene expression in multicellular eukaryotes. The identification of multiple disease-associated proteins within the machine further underscores its physiological importance. Under normal conditions, minor spliceosome-specific components are expressed at low levels
18,49, consistent with the rate-limiting observation of U12-type splicing
10. Mutations or defects in these components frequently lead to developmental disorders
50, indicating that precise regulation of minor spliceosome activity is essential for normal development.
In this study, we identify MMTAG2, a protein highly expressed in multiple myeloma and known to contribute to tumorigenesis
29,37,38, as a U12-type splicing factor. This finding suggests that elevated expression of minor spliceosome-specific components (
e.g., MMTAG2) may relieve the natural bottleneck of U12-type splicing, thereby promoting checkpoint bypass under malignant conditions. In this light, the minor spliceosome acts as a tunable gateway, susceptible to both loss-of-function and gain-of-function perturbations: its proper regulation ensures normal development, whereas its dysregulation drives tumorigenesis. Moreover, the sequence-dependent kinetic bottleneck that we observe across different substrates also reflects a built-in feature that may tune the expression of U12-type intron-containing genes. Our work thus provides new molecular handles for understanding how splicing fidelity is maintained in health and disrupted in disease.
MATERIALS AND METHODS
Preparation of the pre-mRNA
The U12-type pre-mRNAs for the
in vitro splicing assay were modified from the
MINX and
SCN4A genes. MINX-U12 was the substrate we used to purify the human minor B
act complex
21. MINX-U12E was generated by adding an exonic splicing enhancer (ESE) sequence to the 3′ end of the 3′ exon of MINX-U12 (Supplementary Fig. S1a, upper). The ESE sequence is 5′-AGGAUCCGGAAGAAUU-3′, which is reported as ENH1
51. MS-U12 preserves the majority of MINX-U12 sequences but harbors modifications in the BPS and 3′SS regions, and contains a 43-nucleotide (nt) fragment of the SCN4A exon in place of the original 3′ exon. MSE-U12 was generated by inserting the ENH1 enhancer element at the 3′ end of MS-U12 (Supplementary Fig. S1a, bottom). Three tandem MS2-binding RNA aptamers were inserted between the 5′SS and the BPS. The DNA templates for
in vitro transcription were produced by PCR. The RNA substrates were synthesized by T7 runoff transcription and analyzed by urea-PAGE (Supplementary Fig. S1b).
In vitro splicing assay and RT-PCR
Nuclear extract from HeLa S3 cells was prepared for
in vitro splicing as described
52. The
in vitro splicing reaction was assembled in a 20-μL volume, containing 1 nM pre-mRNA substrate, 50% nuclear extract, 20 mM HEPES-KOH (pH 7.9), 65 mM KCl, 2 mM ATP, 20 mM creatine phosphate, and 3 mM MgCl
2. To examine the specificity of the U12-type pre-mRNA substrates, we depleted U1, U2 and U6 snRNAs (to examine the impact on U2-type splicing), or U11, U12 and U6atac snRNAs (to examine the impact on U12-type splicing) in the nuclear extract using endogenous RNaseH at 30 °C for 30 min prior to the splicing reaction (Supplementary Fig. S1c, d). This was achieved by incubating the reaction with 1.5 μM antisense DNA oligonucleotides targeting each of six snRNAs. The sequences of the antisense DNA oligonucleotides (Sangon Biotech) were reported previously
21. The splicing reaction was incubated at 30 °C for gradient durations of 0 min, 30 min, 60 min and 120 min, followed by proteinase K digestion. RNA from the
in vitro splicing assay was extracted using phenol:chloroform:isopentanol at a volume ratio of 25:24:1 (Coolaber Science & Technology). Reverse transcription was performed using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems™) and random hexamers. The RT-PCR products were resolved on 3% (w/v) agarose gel and stained by GoldView (Beijing SBS Genetech Co., Ltd.) (Supplementary Fig. S1c, d).
Expression and purification of human PPR16
Purification of the spliceosomal ATPase/helicase was performed as described
20, 43. The optimized coding sequence of full-length (FL) human PRP16 (DHX38) was synthesized by GenScript
®. The DNA fragment of the ATPase-deficient, dominant-negative mutant PRP16-K561A, which blocks splicing after catalytic step I and traps the major spliceosome at the C complex stage
27, was generated by site-directed mutagenesis using the FL PRP16 sequence described above. The PRP16-K561A fragment was cloned into the pESC-TRP vector with a C-terminal Flag tag. The construct was transformed into the
Saccharomyces cerevisiae (
S. cerevisiae) strain JDY52 (
trp-)
53 by the lithium acetate method. Correct transformants were selected on minimal medium lacking tryptophan (Coolaber Science & Technology). Yeast cells were grown to an OD
600 of 1.8–2.0 at 30 °C in the medium supplemented with 2% (w/v) raffinose, and then pelleted and resuspended in fresh medium supplemented with 2% (w/v) galactose to induce overexpression of PRP16 mutant. After 14–16 h in the galactose medium, the yeast cells were collected and resuspended in lysis buffer that contains 25 mM HEPES-KOH (pH 7.4), 500 mM NaCl, 1.5 mM MgCl
2 and 20% glycerol, and then disrupted by the SPEX 6870 Freezer Mill. The recombinant proteins were purified using the anti-Flag M2 affinity gel (Millipore) and then fractionated through a heparin column (GE Healthcare) to remove the non-specifically bound nucleic acids. Subsequently, the eluted proteins were applied to gel filtration (Superdex-200 10/300 GL, GE Healthcare) in the buffer containing 20 mM HEPES-KOH (pH 7.9), 500 mM NaCl and 5% (v/v) glycerol. The peak fractions were analyzed on an SDS-PAGE gel (GenScript
®) (Supplementary Fig. S2a) and stored at –80 °C.
Assembly and purification of the human minor C complex
The protocol for the minor C complex assembly was modified from those established for the major C complex
27 and the minor pre-B complex
20 (Supplementary Fig. S2a). Briefly, the splicing reaction was performed in a volume of 40 mL or multiples thereof, containing 20 mM HEPES-KOH (pH 7.9), 65 mM KCl, 2 mM ATP, 20 mM creatine phosphate, 3 mM MgCl
2, 0.5 μM PRP16-K561A, 10 nM pre-mRNA, 450 nM MS2-MBP and 50% splicing extract. The pre-mRNA was pre-bound to MS2-MBP for 30 min on ice. To reduce contamination by the major spliceosome, we depleted U1, U2 and U6 snRNAs in the nuclear extract as above. The reaction mixture was incubated for 2 h at 30 °C, and then centrifuged at 3,000×
g for 15 min to remove aggregates. The supernatant was loaded onto amylose resin (NEB), and washed using the G100K buffer (10 mM HEPES-KOH, pH 7.9, 100 mM KCl, 1.5 mM MgCl
2, 0.01% NP40 and 5% (v/v) glycerol). The spliceosomal complexes were eluted using 30 mM maltose. The eluate was loaded onto a 10–30% glycerol gradient with 0–0.05% glutaraldehyde (Sigma), and centrifuged at 23,000 rpm for 10 h at 4
oC in a SW32Ti rotor. The fractions containing the minor C complex were quenched by 25 mM Tris (pH 7.6), pooled and dialyzed against Buffer D (20 mM HEPES-KOH, pH 7.9, 50 mM KCl, 1.5 mM MgCl
2, 0.01% NP40) to remove glycerol prior to EM analysis. The dialyzed sample was analyzed on a 10% urea PAGE gel for detecting RNA elements and concentrated for cryo-EM studies (Supplementary Fig. S2b, c).
Assembly and purification of the human minor C* complex
The protocol for assembly and purification of the minor C* complex was modified from that for the major C* complex
28 and the minor B
act complex
21 (Supplementary Fig. S2d). Briefly, the splicing reaction was performed in a volume of 40 mL or multiples thereof, containing 4 mM HEPES-KOH (pH 7.9), 16 mM MES-NaOH (pH 6.4), 65 mM KCl, 2 mM ATP, 20 mM creatine phosphate, 3 mM MgCl
2, 10 nM pre-mRNA, 450 nM MS2-MBP and 50% splicing extract. Purification procedure is similar with that of the minor C complex. The final sample was analyzed on a urea PAGE gel and concentrated for cryo-EM studies (Supplementary Fig. S2e, f).
MS analysis
About 40 µL of the minor spliceosome samples were mixed with 10 µl of 5× SDS sample loading buffer (GenScript Biotech, China) supplemented with 150 mM dithiothreitol. The sample was incubated at 95 °C for 5 min and resolved using a 4–12% gradient SDS-PAGE gel. The proteins were subjected to in-gel proteolytic digestion as described
54. Peptides were purified using Pierce C18 Spin Tips (ThermoFisher Scientific) prior to LC-MS/MS analysis using Ultimate 3000 nanoLC system coupled with Q Exactive HF-X Hybrid Quadrupole-Orbitrap (ThermoFisher Scientific). About 500 ng of peptides were separated over 90 min using a linear LC gradient of 3–28% (buffer A: 2% acetonitrile, 0.1% formic acid; buffer B: 98% acetonitrile, 0.1% formic acid) at a flow rate of 300 nL/min. The top 20 peptides were subjected to MS2 analysis. MS2 spectra were acquired at a resolution of 30,000 (at m/z 200) in the Orbitrap using an AGC target of 1e5, and max IT of 80 ms. Dynamic exclusion was applied with a repeat count of 1 and an exclusion time of 25 s. The resultant mass spectrometric data were analyzed using pFind
55(Version 3.1.5) against the
Homo sapiens FASTA database downloaded from UniProtKB (version on 27-Apr-2020), which contains 20,365 reviewed protein sequences. Cysteine carbamidomethyl was set as fixed modification and methionine oxidation was set as variable modification. A summary of MS analysis for the human minor C and C* complexes are listed in Supplementary Table S1.
EM data acquisition and pre-processing
Cryo-EM grids for data collection were prepared using Vitrobot Mark IV (FEI Company) at 8
oC and 100% humidity, largely as described
56. Briefly, 4-μL aliquots of the sample at a concentration of ~0.2 mg/mL were applied to Quantifoil R2/1 grids coated with a homemade continuous carbon film of ~2 nm thickness, which were glow-discharged for 30 s using the “Low” setting of the Plasma Cleaner (Harrick, Plasma Cleaner PDC-32G). After blotted for 1.5 s using the standard Vitrobot filter paper (Ø55/20mm, Ted Pella), the grids were plunged into liquid ethane cooled by liquid nitrogen.
The grids were loaded onto a FEI Titan Krios electron microscope equipped with a GIF Quantum energy filter (slit width 20 eV) and operating at 300 kV with a nominal magnification of 81,000×. Images were recorded using a Gatan K3 detector (Gatan Company) in the super-resolution mode, with a pixel size of 0.53865 Å (Supplementary Fig. S2b, e). Each image was dose-fractionated to 32 frames with a dose rate of ~22.49 counts/s per physical pixel (~19.38 e–/s per Å2) and a total exposure time of 2.58 s. Total electron dose for each image is about 50 e–/Å2. All the data were collected using EPU (ThermoFisher Scientific) with a preset defocus range from –1.3 to –1.9 μm. Using the sample of human minor C complexes, we collected a total of 34,090 micrographs, comprising an earlier dataset (dataset 1) of 15,788 micrographs and a later dataset (dataset 2) of 18,302 micrographs. For human minor C* complexes, a total of 58,431 micrographs were collected, containing an earlier dataset (dataset 1) of 22,181 micrographs and a later dataset (dataset 2) of 36,250 micrographs.
The image stacks were motion-corrected using MotionCor2
57 and binned to a pixel size of 1.0773 Å. The dose-weighted micrographs were processed for Patch CTF estimation in CryoSPARC (v4.3.0). Micrographs were discarded if showing obvious ice contamination, excessive drift, or damage.
Cryo-EM data processing
The data processing pipeline is summarized in Supplementary Figs. S3 and S4. For the human minor C complex, we used dataset 1 for initial data analysis (Supplementary Fig. S3a). To avoid missing the spliceosomal particles, we used a reduced threshold for Topaz
58 auto-picking, yielding 2,569,998 particle coordinates. Particles were initially extracted using a pixel size of 8.6184 Å and a box size of 80 pixels. To identify spliceosomal particles, we performed a heterogeneous refinement using reference volumes representing the human minor B
act complex (EMDB code: EMD-30875
21), the major C complex (EMDB code: EMD-6864
39), the major C* complex (EMDB code: EMD-6721
36), the major P complex (EMDB code: EMD-9645
59), the major ILS complex (EMDB code: EMD-9647
59), the major tri-snRNP (EMDB code: EMD-6581
60), the ribosome and three bad classes from previous results, which were low-pass filtered to 20 Å. We further performed 2D classifications on each class from the heterogeneous refinement to enrich particles with a reasonable size. 330,292 particles were selected and subjected to an additional round of heterogeneous refinement with the initial references. Only class derived from C reference shows fine features. The resulting 49,347 particles were re-extracted with a 1× binning factor and processed for non-uniform (NU) refinement, yielding a reconstruction at an average resolution of 3.54 Å, representing the human minor C complex, featured by RBM48–ARMC7 bound at 5′ end of U6atac snRNA.
To enhance the accuracy of particle picking, we performed template picking on datasets 1 and 2 using templates created from the 3.54 Å reconstruction, resulting in 2,007,824 and 2,408,114 particles, respectively. Subsequently, multiple rounds of seed-facilitated 2D classification were performed to maximize the utilization of the collected datasets and achieve rapid preliminary particle alignment. We initially generated 500,000 simulated high-quality particles using the 3.54 Å reconstruction as our seed dataset. The template-picked particles from datasets 1 and 2 were divided into four and five subgroups respectively, with each subgroup containing a similar number of particles as the seed dataset. Each subgroup was then mixed with the simulated seed dataset and subjected to 2D classification to select particles with fine features. After removing simulated particles, 402,437 particles from dataset 1 and 987,302 particles from dataset 2 were re-extracted with a 4× binning factor and processed through similar seed-facilitated 2D classifications, yielding a total of 1,014,725 particles (263,354 from dataset 1 and 751,371 from dataset 2).
These particles were re-extracted with 2× binning factor and subjected to three parallel runs of heterogeneous refinement to obtain more particles. The 3.54 Å map was rescaled to a pixel size of 2.1546 Å and used as a “good” reference, while three other maps generated during processing of dataset 1 served as bad references. Particles from good classes were merged and duplicated particles were removed. The resulting 357,030 particles were re-extracted at a pixel size of 1.0773 Å and processed for NU refinement, yielding a reconstruction at an average resolution of 3.43 Å. After one round of global 3D classification without image alignment, 210,708 particles were selected, processed for NU refinement and CTF refinement, improving the resolution to 3.07 Å for the entire human minor C complex. We further selected 86,306 particles through an iterative particle sorting algorithm call CryoSieve
61, resulting in a reconstruction at an overall resolution of 2.92 Å (Supplementary Fig. S3a, b). The local resolution reaches 2.5 Å in the core region.
In contrast to the core region, the peripheral regions of the human minor C complex exhibit flexibility. To improve map quality, we performed localized 3D classification and refinement on eight regions using region-specific soft masks. This approach yielded high-resolution densities for seven regions: the RBM48–ARMC7 region (3.07 Å; 58,941 particles), the PRP8 RNaseH-like domain (3.48 Å; 37,176 particles), the EJC (3.59 Å; 51,437 particles), PRP16 (3.29 Å; 50,530 particles), and the region encompassing BRR2 and PRP16 (3.97 Å; 72,597 particles), the IBC region (4.41 Å; 85,373 particles), the SYF1−SYF3 region (3.21 Å; 85,373 particles). However, flexibility limited the resolution of the PRP19-associated region to a level that revealed only its global architecture. Overall, these targeted refinements significantly enhanced the local resolution of peripheral features compared to the global 2.92-Å map.
For the human minor C* complex, a similar data processing procedure was carried out with slight modification (Supplementary Fig. S4a). After multiple rounds of heterogeneous refinement and 2D classification of 3,162,134 Topaz-picked particles, 39,581 particles, yielding a reconstruction at 4.38-Å resolution. The same procedure of template picking, seed-facilitated 2D classification, hetero-refinement, and 3D classification was performed as described above. A total of 145,908 particles were selected, re-extracted with a pixel size of 1.0773 Å, and subjected to NU and CTF refinement, yielding a 3.04-Å reconstruction. Local resolutions were improved via focused classification and refinement for several regions: RBM48–ARMC7 (3.19 Å; 88,127 particles), SYF1/3 (3.48 Å; 107,934 particles), EJC (4.01 Å; 61,936 particles), PRP22 (4.23 Å; 95,438 particles). For BRR2, the BPS-proximal WD40 protein, and IBC, we performed NU refinements using the region-specific particle subsets to improve local map quality. Focused classification of the PRP19-associated region (22,629 particles) revealed its overall fold and the LSm heptameric complex, with a local resolution of 7.71 Å after refinement. (Supplementary Fig. S4a,b).
The above reported resolutions were calculated according to the FSC 0.143 criterion with a high-resolution noise substitution method
62. Prior to visualization, all density maps were sharpened by applying a negative B-factor that was estimated using automated procedures
63 in cryoSPARC. Local resolution variations were estimated using cryoSPARC (Supplementary Tables S2 and S3).
Model building and refinement
Model building of minor C and C* complex was carried out using COOT
64 and UCSF Chimera
65. We combined appropriate modeling methods including docking and manual adjustment,
de novo modeling, homology modeling, AI-facilitated modeling, and rigid-body docking according to resolution levels of the various local maps to generate the atomic models (Supplementary Tables S4 and S5).
For the minor C complex, components were identified using the atomic coordinates of the human minor B
act complex (PDB code: 7DVQ
21) and the human major C complex (PDB code: 8I0W
34). For initial model building, we first aligned the locally refined maps with the consensus map to integrate high-resolution information. This composite map was then used to guide the docking of relevant structures from the human minor B
act complex (PDB code: 7DVQ
21), the major C complex (PDB code: 8I0W
34), and the AlphaFold-predicted models
66 of PRP16 and FRG1 into the maps using UCSF Chimera
65. The docked models were subsequently truncated as needed, manually adjusted and extended (if necessary) using COOT
64. Structural models for the BRR2 and IBC regions were generated by rigid-body docking of coordinates from the major C complex, while the U12 Sm ring and associated RNA were docked from the minor B
act complex. A model of the complex comprising the PRP19-tetramer, SPF27, the C-terminal fragment of CDC5L, and the N-terminus of PRP46 was generated by docking a refined Alphafold3-predicted complex model
67, which was manually truncated and adjusted against the local map for this region in the minor C* complex.
After these steps, several unassigned EM density patches remained, including regions adjacent to the PRP8 RNaseH-like domain and the C-half of CWC22. The A
2-Net method
68 was applied to its density to recognize amino acid side chains, which, combined with MS data, identified the protein as MMTAG2. The AlphaFold model of the N-terminal fragment of MMTAG2 fitted well into the density and was manually adjusted, revealing that its downstream sequence extends into the CWC22-proximal density. An AlphaFold3-based interaction screen of the CWC22–MMTAG2 interface identified a fragment of CWC25 as the interacting partner occupying this density. Such a method was used to model other missing fragments, including CDC5L (residues 461–487), SRm300 (residues 122–161), and CCDC12 (residues 107–150), into the remaining densities.
Model building of the human minor C* complex followed a procedure analogous to that described above. Based on the EM maps of the human minor C* complex, we selected appropriate segments of the atomic coordinates from the human minor B
act complex (PDB code: 7DVQ
21) and the major C* complex (PDB code: 8C6J
32), and the AlphaFold2 models of SYF1, SYF3, PRP22, Aquarius, and PPIE, docked them into the EM densities and performed manual adjustments, extensions, or rigid-body docking as required. To determine the identity of the unassigned density in the vicinity of the BPS/U12 duplex, A
2-Net-facilitated modeling
68 was employed. This approach generated the models for FAM204A and WDR25. Additionally, the A
2-Net method was used to model interacting fragments of SKIP and PLRG1 with the RBM48–ARMC7 complex.
To identify the binding partner of PRP22 RecA2 domain in the minor C* complex, we employed AlphaFold3
67 to screen against PRP22. This analysis predicted a complex between PRP22 and the N-terminal region of RBM41, which was validated by its exceptional fit to the adjacent density adjacent to PRP22 and consistent with proximity-labeling data
47. Additional fragments, including CDC5L (residues 289–327, 359–387, 461–487), SRm300 (residues 122–161), and CCDC12 (residues 107–150), were incorporated based on AlphaFold3 predictions. For the PRP19-associated region, a complex model was predicted using AlphaFold3, docked as a rigid body, and manually refined according to the local map. The model for the 3′ region of U6atac snRNA and associated LSm2-8 heptameric complex was generated by homology modeling based on the
S. cerevisiae Lsm2-8−U6 snRNA complex structure (PDB code: 4M7A
69) and prediction of the structure of the 3′ stem loop of U6atac snRNA by trRosettaRNA
70. After docking the two models, we manually built extensions for U6atac snRNA (residues 66–79, 117–120).
The entire models of the human minor C and C* complexes were respectively refined against the 2.92-Å map and 3.04-Å map using PHENIX
71 in real space with secondary structure and geometry restraints. Model overfitting was monitored by model-to-map FSC curves (Supplementary Figs. S3c, S4c). Model quality was assessed using the Molprobity scores and the Ramachandran plots (Supplementary Tables S2 and S3). Molprobity scores were calculated as described
72. Components included in the final coordinates are summarized in Supplementary Tables S4 and S5.
DATA AVAILABILITY
The atomic coordinates have been deposited in the Protein Data Bank (PDB) with the following accession codes: 9XTT for the minor C complex and 9XU3 for the minor C* complex. The EM maps have been deposited in the EMDB with the following accession codes: for the minor C complex, EMD-67250 (overall map), EMD-81104 (with stably bound RBM48 region), EMD-81123 (SYF region), EMD-81125 (PRP16 region), EMD-81129 (with stably bound PRP8-RH), EMD-81130 (EJC region), EMD-81127 (helicase region), EMD-81204 (IBC region); for the minor C* complex, EMD-67263 (overall map), EMD-81139 (with stably bound WDR25 region), EMD-81137 (RBM48 region), EMD-81138 (SYF region), EMD-81140 (with stably bound BRR2 region), EMD-81143 (with stably bound IBC region), EMD-81144 (PRP22 region), EMD-81145 (EJC region), EMD-81142 (PRP19-LSm region). All data and materials are available from the corresponding author upon reasonable request.
The Author(s) 2026. Published by Higher Education Press. This is an Open Access article distributed under the terms of the CC BY license (https://creativecommons.org/licenses/by/4.0/).
