BORC assemblies integrate BLOC-1 subunits to diversify endosomal trafficking functions

Significance

Eukaryotic cells depend on complex molecular machines to regulate internal transport. We studied two ancient complexes, BORC and BLOC-1, which were previously considered separate entities. Our research reveals the existence and regulated assembly of specific hybrid BORC–BLOC-1 complexes, enabling the cell to fine-tune its internal logistics. Defects in these complexes have been linked to neurodevelopmental, pigmentary, and cancerous diseases, so understanding this dynamic assembly is critical. Our work establishes that thoroughly discriminating between the canonical complexes and the various hybrid assemblies is an essential prerequisite for future therapies. The efficacy and specificity of pharmacological interventions depend on our ability to target the correct complex, paving the way for precise, disease-specific strategies.

Abstract

BORC and BLOC-1 are multisubunit complexes that regulate endolysosomal trafficking. Although they are presumed to be distinct, their paralogous origins and shared subunits suggest the potential for higher-order assembly. Here, we reveal the conserved octameric architecture of BORC formed by two intertwined tetramers and present the structure of C. elegans BORC. Through cross-linking mass spectrometry of endogenous complexes, we validate this model for human BORC and demonstrate that the integrity of the complex, which is essential for lysosomal transport, relies on specific interfacial residues. We also clarify the disruptive nature of disease-causing mutations and propose that the formation and function of BORC are likely regulated by specific cues. These cues might include the phosphorylation of Snapin and a pH-sensitive histidine residue in BORCS5. Additionally, we present direct biochemical and structural evidence of BORC–BLOC-1 hybrid complexes. Finally, we link a specific hybrid complex to the regulation of transferrin receptor recycling via interaction with the EARP complex. Our work challenges the paradigm of BORC and BLOC-1 as separate entities, establishing a model of dynamic complex formation wherein modular assembly creates functional specialization to meet diverse cellular demands.

The rise in complexity of the endomembrane system stands as one of the primary determining factors of eukaryotic life, with major organelles being present at the time of the last eukaryotic common ancestor (LECA). To sustain trafficking between organelles, several molecular machines have evolved through expansion of a smaller set of primordial genes. These machineries control cargo recognition, coat formation, budding/scission, uncoating, delivery, and ultimately fusion with the targeting organelle (1–6). A specialized subset of these trafficking regulators includes the Biogenesis of Lysosome-related Organelles Complexes (BLOC-1, -2, and -3), which coordinate cargo delivery not only to conventional lysosomes but also to secretory lysosome-related organelles such as melanosomes, platelet dense granules, and lamellar bodies (7). In particular, the BLOC-1 complex has been shown to control recycling endosome (RE) biogenesis and melanosome maturation (8). Although the maintenance of the RE tubular structures requires close coordination between BLOC-1 components, AP-3, microtubule, and actin associated machineries (9), in vitro the presence of BLOC-1 is sufficient to induce tubulation of PI4P containing vesicles (10), underscoring the active involvement of the complex in membrane remodeling. Mutations in three of the BLOC-1 components (DTNBP1, BLOC1S3, and BLOC1S6) are associated with Hermansky-Pudlak syndrome types 7 to 9. Patients with this condition present with two characteristic features: a tendency to bleed (diathesis) and oculocutaneous albinism (11).

In vitro, BLOC-1 was found to form an elongated octameric complex containing BLOC1S1, BLOC1S2, BLOC1S3, BLOC1S4, BLOC1S5, BLOC1S6, Snapin, and DTNBP1 (12, 13). Three subunits (BLOC1S1, BLOC1S2, and Snapin) are shared between BLOC-1 and BLOC-1 related complex (BORC) (14). The latter primarily binds to lysosomes and plays an active role in organelle positioning (14, 15), tubulation, size regulation, and ultimately lysosome-autophagosome fusion (16). Mutations in BORC have been linked to neurodevelopmental diseases in both humans and mice (17, 18). Similar to BLOC-1, BORC forms an octameric complex composed of BORCS5, BORCS6, BORCS7, BORCS8, KxD1, and the three shared subunits (14, 19). Evolutionary reconstructions suggest that BORC and BLOC-1 evolved from a common ancestral complex through subunit paralogy prior to LECA. These ancient complexes coevolved with core eukaryotic features, with most subunits originating from the early stages of eukaryogenesis (7, 20). Given their shared subunits and similar membrane remodeling activities at different organelles, BORC and BLOC-1 may use conserved mechanistic principles despite their organelle specialization.

We employed an integrative approach combining structural biology, cross-linking mass spectrometry, and functional assays to define these principles. We used BORC complexes from C. elegans and Homo sapiens as primary models. Our findings reveal the mechanistic basis for complex assembly by identifying critical residue interactions and motifs that synchronize octamer formation with diverse cellular demands. Additionally, we present experimental evidence that supports the existence of mixed BORC–BLOC-1 assemblies.

Results

Integrative Structure-Biology Approach Reveals the Molecular Basis of BORC Complex Formation.

Given the presence of BORC and BLOC-1 subunits throughout Eukarya, we first generated AlphaFold3 models of BORC across several model species: Arabidopsis thaliana, Mixina glutinosa, Aplysia californica, Dictyostelium discoideum, Caenorhabditis elegans, and Homo sapiens. All of these species contained representatives of the 8 BORC subunits (SI Appendix, Fig. S1 A and B). These models suggest that the complex’s overall architecture is conserved across eukaryotes. Moreover, in all analyzed species, BORC seems to assemble as an elongated rod-shaped complex obtained by the juxtaposition of two tetramers formed by the intertwined alpha helices of four components each (SI Appendix, Fig. S1A). Within this assembly, all subunits orient their N termini toward the central joint region, with their C termini facing outward. One tetramer comprises BORCS8, BLOC1S1, KxD1, and BORCS6, and the other comprises BLOC1S2, BORCS5, BORCS7, and Snapin. Our models also suggest that BORC subunits have extended unstructured termini that are unlikely to contribute to complex assembly. Together, our in-silico predictions suggest that BORC has a highly conserved three-dimensional structure. These findings align with previous bioinformatic predictions (7, 20) and a recently reported structure of C. elegans BORC (21).

Based on these findings, we selected C. elegans BORC for structural studies. The choice was based on its high sequence conservation and favorable biophysical properties, notably a compact fold with fewer disordered regions than the human ortholog. Moreover, there is experimental evidence indicating that C. elegans BORC controls Arl8B function (22) and phagocytic clearance (23), just as vertebrate BORC. We designed an E. coli construct that allowed for simultaneous expression of all eight BORC subunits and included a twin STREP tag preceded by a TEV cleavage site at the C-terminus of BORCS6 and a SUMO-His₆ tag at the N terminus of BORCS5 (SI Appendix, Fig. S2A). 2D gel electrophoresis analysis of the affinity-purified BORC complex following size exclusion chromatography (SEC) confirmed the presence of all BORC components (SI Appendix, Fig. S2B). We then analyzed the sample by negative staining electron microscopy (nsEM) (SI Appendix, Fig. S2C), which revealed that the complex forms an elongated rod structure, consistent with previous data on the BLOC-1 complex (12) and the AlphaFold3 predictions of the BORC architecture (SI Appendix, Fig. S1A). The purified BORC complex was then digested to remove the SUMO tag, run on an SDS-PAGE and stained with Coomassie. Individual bands were cut and identified by mass spectrometry (SI Appendix, Fig. S2D). Rewardingly, all BORC components were present in a single SEC peak. In order to optimize the conditions for disuccinimidyl dibutyric urea (DSBU) cross-linking mass spectrometry analysis (XL-MS), we used SUMO and TEV cleaved complex and titrated the cross-linker amounts (SI Appendix, Fig. S2E). We observed the accumulation of a single band at approximately 200 kDa, accompanied by reduction of the lower molecular weight bands of the individual BORC components. XL-MS analysis of the C. elegans BORC complex led to the identification of over 1,300 cross-linked peptides. From the 344 unique peptides, 292 were interprotein and 52 intraprotein linkages (SI Appendix, Fig. S2F and Table S1). The preferential occurrence of interprotein vs. intraprotein cross-linked peptides is consistent with the elongated rod shape assumed by C. elegans BORC in nsEM (SI Appendix, Fig. S2C) and in the AlphaFold3 model (SI Appendix, Fig. S1A). We then used AlphaLink2 to generate a model of BORC based on the XL-MS analysis (Fig. 1 A and B). Feeding the algorithm with all cross-linked residues with a confidence score higher than 100, led to a model with a satisfaction value of 0.567 and a mean distance of the cross-linked residues of 40.4 Å. Since secondary structure predictions and the AlphaFold3 model indicated that the termini of the BORC subunits are unstructured, we removed the corresponding cross-linked peptides from the analysis (Fig. 1A). This improved the cross-linked satisfaction to 0.825, with a mean distance between the cross-linked residues of 31.2 Å, in line with the arm length of DSBU. We then superposed the AlphaLink2 model to the AlphaFold3 model of C. elegans BORC (Fig. 1B). The two models were nearly identical, with only a small bending of the complex in one of the tetramers (RMSD 0.944 Å).

Fig. 1.

Integrative structure-biology approach reveals the molecular basis of BORC complex formation. (A) Distances between Cα atoms observed for the detected cross-linked peptides in the Alphalink2 model. Left side, results for all cross-links with a score >100, Right hand side, same analysis excluding cross-links in unstructured termini. (B) Side view of the superposition of the AlphaFold3 and the AlphaLink2 models of C. elegans BORC. In the AlphaFold3, BORCS5 is represented in dark olive green, BORCS6 in orange, BORCS7 in yellow green, BORCS8 in saddle brown, KxD1 in dark goldenrod, Snapin in dark cyan, BLOC1S1 in medium aquamarine and finally BLOC1S2 in dark slate gray. All subunits from the Alphalink2 model are shown in dark blue. The RMSD of the analysis is indicated. (C) AlphaLink2 model fitted into the CryoEM density. BORCS5 is presented in dark olive green, BORCS6 in orange, BORCS7 in yellow green, BORCS8 in saddle brown, KxD1 in dark goldenrod, Snapin in dark cyan, BLOC1S1 in medium aquamarine and finally BLOC1S2 in dark slate gray. (D) CryoEM density fitted C. elegans BORC model with indicated cross-links with a score >100 in blue.

In parallel, we selected the SEC purified octameric BORC peak for single particle cryo-electron microscopic (cryoEM) analysis (SI Appendix, Fig. S2 G–K and Table S2). After template-based particle picking, the identified particles were cleared using iterative reference-free 2D classification. An initial map was generated using ab initio reconstruction followed by homogeneous refinement and postprocessing using DeepEMhancer, resulting in a map with a nominal resolution of 7.8 Å. In agreement with the AlphaFold3 and Alphalink2 models, the purified C. elegans BORC complex assembled as an elongated rod shape. Detailed analysis of the low-resolution structure led to the identification of three full helices present on each side of the BORC complex. We then fitted the AlphaLink2 model of C. elegans BORC to the CryoEM density (Fig. 1C). While the core of the N-terminal bitetramer assembly point fit immediately, the distal C-terminal region required further optimization. Using molecular dynamics simulations with secondary structure restraints (ISOLDE) (24), we were able to unambiguously trace the alpha-helices extending from the core toward the distal region. This analysis revealed that KxD1 and Snapin were largely unresolved at their distal C termini, indicating structural flexibility (Fig. 1C). Although the extension of the alpha-helices required refinement, crosslink-derived distance restraints determined by XL-MS for DSBU (35 Å) were satisfied (Fig. 1D). Together, our combined XL-MS and CryoEM analyses confirmed the formation of an octameric BORC complex in solution, consistent with predictions made using AlphaFold3. Importantly, our results are in agreement with the recently published BORC structure (21) and unambiguously assign the position of the eight components in the density.

Functional Validation of the BORC Structural Model.

To systematically identify critical residues in BORC assembly, we first mapped evolutionary conservation patterns across the AlphaFold3-modeled human BORC structure (SI Appendix, Fig. S3A). Combining this analysis with biophysical and functional considerations, we selected residues for mutagenesis that represent distinct structural and regulatory features (SI Appendix, Fig. S3B).

The clinically significant BORCS6 R109P variant (Fraser syndrome-associated; ClinVar) illustrates that disease relevance does not always correlate with structural impact. Located in a disordered BORCS6 segment with no predicted subunit interactions (SI Appendix, Fig. S3C), this mutant had no effect on coimmunoprecipitation of BORC components (Fig. 2A and SI Appendix, Fig. S3 J and K), suggesting that its pathogenicity operates through alternative mechanisms. In stark contrast, truncation of BORCS7 at Q88, which removes predicted interaction interfaces with BLOC1S2 and BORCS5 (SI Appendix, Fig. S3D) and causes axonal dystrophy in mice (18), completely abolished complex incorporation (Fig. 2 B and F), confirming the essential role of this C-terminal domain. Central to the BORC architecture is the interaction hub centered on BORCS5 Q111, which AlphaFold3 predicts forms hydrogen bonds with both BLOC1S2 N70 and Snapin Q51 (SI Appendix, Fig. S3E). The Q111A mutant showed broad binding defects (Fig. 2 C and F), confirming its structural importance. Nearby, Snapin S50—known to regulate synaptic vesicle trafficking when phosphorylated (25, 26)—forms a hydrogen bond with BLOC1S2 N70 (SI Appendix, Fig. S3E). Both S50A and S50E mutations impaired BORC assembly (Fig. 2 D and F), although the absence of nearby acidic residues suggests that this is due to structural disruption rather than charge repulsion. Regulatory sites showed more nuanced effects. BORCS7 S46, which hydrogen-bonds BLOC1S2 D62 (SI Appendix, Fig. S3 F and G) and is phosphorylated in vivo [PhosphoSitePlus (27)], tolerated both phosphomimetic (S45,46E) and phospho-null (S45,46A) mutations with only mild defects (Fig. 2 B and F), suggesting that phosphorylation fine-tunes rather than controls interactions. Similarly, the brain morphology-associated Snapin R55W polymorphism neither affected complex assembly (Fig. 2 D and F) nor mapped to conserved/interacting regions (SI Appendix, Fig. S3H), suggesting it might potentially disturb BORC independent functions.

Fig. 2.

In vivo functional validation of BORC’s structure. (A) Expression of SH-tagged BORCS6 wild type was induced by incubating the corresponding HEK293 Flp-In T-REx cell lines with 20 ng/mL tetracycline. Streptavidin affinity precipitates and respective inputs were analyzed by immunoblotting. SH, StrepII-HA-tag; HA, hemagglutinin. n = 2 independent biological experiments. (B–E) Experiments were performed as described in (A) for BORCS7 (n = 3), BORCS5 (n = 3), Snapin (n = 3), and BORCS8 (n = 2) respectively, (F) Quantification of the experimental data from panels (B–D). (G) Indirect immunofluorescence images of HeLa wild type, BORCS7 KO, and reconstituted cell lines kept under starved or stimulated conditions. Merged and single-channel images of endogenous lysosomal marker LAMP1 (red) and Phalloidin (green) are indicated. Representative cells are shown. n = 3 independent biological experiments. (Scale bar, 10 μm.) (H) Indirect immunofluorescence images of HeLa wild type, BORCS5 KO, and reconstituted cell lines kept under starved or stimulated conditions. Merged and single-channel images of endogenous lysosomal marker LAMP1 (red) are indicated. Representative cells are shown. n = 3 independent biological experiments. (Scale bar, 10 μm.) (I) The lysosomal distribution of HeLa wild type, BORCS7KO, and reconstituted lines was analyzed based on the experiment shown in (G), Graphic depicts the distance of each lysosome to the nucleus in µm. The vertical black line represents the median, the dotted line shows the 17 µm distance from the nucleus distinguishing peripheral from perinuclear lysosomes. A minimum of 19 cells were analyzed with more than 5,235 lysosomes per condition. (J) The lysosomal distributions of HeLa wild type, BORCS5KO, and reconstituted lines were analyzed based on the experiment shown in (H) Graphic depicts the distance of each lysosome to the nucleus in µm. The vertical black line represents the median, the dotted line shows the 17 µm distance from the nucleus distinguishing peripheral from perinuclear lysosomes. A minimum of 18 cells were analyzed with more than 4,710 lysosomes per condition.

Conservation-driven mutants revealed additional assembly principles. While BORCS8 P44A incorporated almost normally into BORC, its reduced cellular expression (Fig. 2E and SI Appendix, Fig. S3 I and K) suggested a role in preassembly stability. More strikingly, BORCS5 H99A/E mutants showed opposing effects on different binding partners (Fig. 2 C and F and SI Appendix, Fig. S3 F and G), demonstrating how single residues can bias subunit incorporation.

To validate our structural predictions in a cellular context, we performed functional rescue experiments using HA-tagged wild-type and mutant constructs in BORCS5/BORCS7 knockout HeLa cells (SI Appendix, Fig. S3L). In this manuscript, “in vivo” refers to cell culture experiments analyzing endogenous complexes, as opposed to reconstituted “in vitro” systems or “in silico” predictions. Both knockout lines exhibited impaired lysosomal redistribution upon growth factor stimulation (Fig. 2 G–J), consistent with a role for BORC in lysosomal positioning (14). While BORCS7 WT, S45,46A and S45,46E mutants all restored normal lysosomal dynamics (Fig. 2 G and I and SI Appendix, Fig. S3M), consistent with their mild biochemical defects, the C-terminally truncated variant (1-88) severely impaired outward transport (Fig. 2G and SI Appendix, Fig. S3M), confirming the structural importance of this domain. Similarly, BORCS5 WT reconstitution rescued peripheral lysosomal distribution (Fig. 2 H and J and SI Appendix, Fig. S3N), but none of the mutants tested could fully recapitulate this effect. Notably, Q111A and H99A mutants not only failed to restore normal trafficking, but actually reversed the direction of lysosomal transport under anabolic conditions (Fig. 2J and SI Appendix, Fig. S3N), suggesting that these residues mediate regulatory inputs beyond structural roles.

Taken together, these structural and functional analyses reveal three fundamental principles governing BORC organization and activity. First, we identify essential architectural nodes—in particular the BORCS7 C-terminus and the BORCS5 Q111 interaction hub—whose disruption prevents both complex assembly and physiological function. These elements represent nonredundant structural pillars that have been strictly conserved throughout eukaryotic evolution. Second, disease-associated variants segregate into distinct classes: those like BORCS7 1-88, that directly compromise complex integrity, and others (BORCS6 R109P, Snapin R55W) that likely exert pathogenic effects through alternative mechanisms. This distinction provides a framework for interpreting mutation pathogenicity in the context of multiprotein complexes. Finally, we uncover how posttranslational modifications and residue-specific interactions enable dynamic regulation—phosphorylation of BORCS7 S46 fine-tunes binding affinity without abolishing interactions, while BORCS5 H99 appears to function as a molecular switch that can redirect trafficking outcomes.

Monitoring the Existence of BORC/BLOC-1 Mixed Complexes in Cells and Addressing Their Possible Physiological Function.

The paralogous origins of BORC and BLOC-1, their shared subunits, and their conserved interaction modes suggest the potential formation of hybrid complexes. These complexes may constitute an unrecognized layer of regulatory diversity in the endolysosomal system. In this context, our mutagenesis strategy provides an ideal framework for directly testing the hypothesis of hybrid assembly formation. We performed affinity purifications of mutated BORCS7 and BORCS5 under different physiological conditions (Fig. 3A and SI Appendix, Fig. S4 A–D). We hypothesized that the mutations previously found to disrupt the BORC complex assembly should also impact the formation of mixed assemblies. Because commercially available antibodies failed to reliably detect several BORC and BLOC-1 subunits, we employed quantitative mass spectrometry as our primary detection method, validating the approach with Western blotting for those subunits where antibodies proved effective. In all interactomes analyzed, we detected the known BORC components as well as three BLOC-1 subunits—BLOC1S4, BLOC1S6, and BLOC1S3 (Fig. 3A, yellow circles). Importantly, the relative abundance of these BLOC-1 components exceeded that of BORCS8 and in some cases also KxD1 (Fig. 3A, orientation of the hits in a clockwise manner). The results therefore suggest a robust association of specific BLOC-1 components with BORC. Consistent with the steady-state interactome results, truncation of BORCS7 or mutation of Q111 of BORCS5 resulted in a marked reduction in association with all BORC subunits as well as the BLOC-1-specific components. Western blot analysis confirmed these findings (SI Appendix, Fig. S4 A–D). The BORCS5 H99A and H99E mutants showed a less homogeneous effect. Interactions with BORCS6, BLOC1S1, KxD1, and BORCS8 were significantly decreased (Fig. 3A, green circles) whereas binding to other BORC components remained unchanged. A closer look at the AlphaFold3 model of BORC revealed that BORCS5 H99 mutations could induce a separation of the two tetramers. Hence, binding of BORCS5 H99 mutants to the components of the opposite tetramer decreased while their association with subunits in the same tetramer remained intact. Both BORCS5 H99A and H99E mutations showed comparable binding patterns to BLOC-1-specific components under all experimental conditions, with no substantive changes observed.

Fig. 3.

Monitoring the existence of BORC/BLOC-1 mixed complexes. (A) HEK293 Flp-In T-REx cell lines induced to trigger the expression of SH-tagged BORCS7wt, BORCS7_1-88, SH-tagged BORCS5wt, and respective mutants or SH.GFP, were either starved for FBS or stimulated. The eluates from Streptavidin affinity precipitations were subjected to mass spectrometry analysis. Each graphic displays the comparison between the interactome of the mutant mentioned in the center of the scheme and the respective wt control. The copurified subunits are ranked clockwise by their abundance in the WT interactome. The ratio of the abundances of each interactor in the mutant vs wt interactomes is shown in a color gradient from red to dark blue. The statistical significance of the abundance ratio is indicated by the line surrounding the circle. BLOC-1 specific components are encircled in yellow. The components of BORC present in the opposite tetramer of the complex relative to the bait are encircled with a green line, n = 3 independent biological experiments. (B) HA.BORCS7 was stably expressed in the background of the BORCS7KO, BORCS6KO, or BORCS5KO. All cell lines were subjected to anti-HA immunoprecipitation under steady state conditions. Eluates and respective inputs were analyzed by immunoblotting. HA, hemagglutinin. n = 4 independent biological experiments. (C) Mass spectrometry analysis of the eluates obtained in (B). Graphic representation follows the same pictogram rules as in panel (A) with the exception that the abundance values of HA.BORCS7 are also shown. (D) In vitro affinity purification of BORC hexamers. SDS-PAGE gel was stained with coomassie. (E) HEK293 Flp-In T-REx cell lines inducibly expressing BORC components or BLOC1S6 were subjected to Streptavidin affinity precipitation. Eluates composition was analyzed by mass spectrometry. The table represents the abundance of each subunit in the different interactomes, normalized to the levels found using Snapin as a bait. n = 4 independent biological experiments. (F) Scheme representing the putative composition of mixed complexes A and B.

To further strengthen these findings, we expressed HA-tagged BORCS7 in the background of the BORCS7, BORCS6, and BORCS5 knockouts (Fig. 3 B and C). The BORCS7 knockout reconstituted with HA.BORC7 was taken as a control. We found a strong increase in the presence of BLOC1S6 when HA.BORCS7 was purified from the BORCS6 knockout background, but not from BORCS5 ablated cells (Fig. 3B). We then analyzed the obtained eluates using mass spectrometry (Fig. 3C). In the BORCS6KO cells, the interactome of BORCS7 showed a significant reduction of the BORC specific components BORCS6, BORCS8, and KxD1. In the same sample, we detected a significant increase in BLOC1S6, BLOC1S4, and BLOC1S3. Interestingly, this effect was specific to the cellular background as the BORCS7 interactome from the BORCS5 ablated cells seemed to incorporate preferentially BORC, showing a decrease of the copurified BLOC-1 subunits.

Ge et al. have demonstrated that the two BORC tetramers are stable in vitro (21). They also proposed a hierarchical assembly system using BLOC1S1, BLOC1S2, Snapin, and BORCS5 as core units to recruit the remaining BORC components. Based on these and our in vivo data, we tested whether hexameric BORC subcomplexes might be stable in vitro. To this purpose, we deleted from full BORC either KxD1 and Snapin, or KxD1 and BORCS8, or BORCS5 and BORCS7. Affinity purifications using the BORCS6 STREP-tagged subunit demonstrate that stable hexamer assembly requires an intact hemicomplex on the purification side. However, BORCS5 is not strictly necessary for this assembly.

We then systematically screened for BLOC-1 components within the BORC interactomes. We used HEK293 Flp-In T-Rex cell lines engineered for inducible expression of SH-tagged subunits: BORC-specific, BLOC-1-specific, and shared components, with SH-GFP as a negative control (Fig. 3E and SI Appendix, Fig. S4E). In order to adequately compare the integration of the noncanonical components in either BORC or BLOC-1 complexes, we decided to normalize the abundancies of each interactor to the abundance of the same protein detected in the interactome of a shared subunit (either Snapin or BLOC1S1). Using the Snapin interactome for normalization, we observed a strong enrichment of BORC specific subunits when BORC subunits were used as baits (Fig. 3E). In a similar manner, BLOC-1 components were predominantly found in the BLOC1S6 interactome. In line with mixed complexes formation, there was a considerable fraction of BORCS5 and BORCS7 on the BLOC1S6 interactome, corresponding to nearly 70 % of the amount of the same components in the Snapin interactome. The remaining subunits of BORC were nearly absent from the BLOC1S6 interactome. We detected BLOC1S4 and BLOC1S6 in the BORCS5 interactome, albeit at lower abundance. This finding reciprocally validated the BLOC1S6 interactome data and was consistent with the results of the BORCS5 mutant analyses (Fig. 3A). Notably, BLOC1S5 and DTNBP1 were also detected in BORCS6 affinity purifications. This indicates a highly selective presence of BLOC-1 subunits in the interactomes of different BORC components.

The tendencies observed for the formation of mixed complexes were independent of the interactome used for normalization (SI Appendix, Fig. S4E). However, the normalization against the BLOC1S1 interactome revealed some differences concerning the prevalence of the BORC specific subunits. It is possible that these changes might be related to tagging effects, but it is also possible that shared subunits do not integrate in the different complexes in a strictly stoichiometric manner.

Based on our results, we hypothesized the existence of two distinct mixed complexes (Fig. 3F). Mixed complex A is based on the stable hexamer observed in vitro and the interaction screen (Fig. 3 D and E). Mixed complex B is derived from the in vivo experimental data shown in Fig. 3 A and B and E. We used AlphaFold3 to model both putative assemblies (SI Appendix, Fig. S5 A–C). The resulting predictions incorporated BORC and BLOC-1 components into octameric complexes resembling the canonical BORC and BLOC-1.

To test the existence of these assemblies in vivo, we combined our STREP-tag affinity purifications with XL-MS, employing enrichable PhoX cross-linker (SI Appendix, Tables S3–S5). The analysis was performed using SH.BORCS6 and BORCS7.HS as baits. Following the same strategy as for the C. elegans BORC complex, we improved XL satisfaction by removing cross-links in unstructured termini (SI Appendix, Fig. S5D). We then mapped the remaining cross-links to the AlphaFold3 model of human BORC (SI Appendix, Fig. S5E). The vast majority of the cross-links satisfied the distance constraints, thereby validating the AlphaFold3 prediction using in vivo human BORC data.

Next, we interrogated the XL-MS results for specific interactions between BORC and BLOC-1 components. Notably, the structural model of mixed complex A was supported by a total of 11 cross-links (<5 % FDR) detected in SH.BORCS6 purifications (Fig. 4 A–C). The limited number of cross-links found is in line with the lower abundancy of these assemblies in comparison to canonical BORC. Cross-links of DTNBP1 with BLOC1S1 and BLOC1S5 exceed the 25 Å distance threshold most likely due to the flexibility of the DTNBP1 termini (Fig. 4C).

Fig. 4.

Existence of mixed complex A and possible physiological function of a mixed complex assembly. (A) AlphaFold3 prediction of mixed complex A (iPTM 0.46). BORCS6 is represented in orange, BLOC1S1 in medium aquamarine, BORCS8 in saddle brown, KxD1 in dark goldenrod, BLOC1S2 in dark slate gray, Snapin in dark cyan, BLOC1S5 in tomato and DTNBP1 in maroon. The structure is supported by a total of 11 cross-links (<5 % FDR) detected in SH-BORCS6 AP-XL-MS experiments. Cross-links between BORC exclusive subunits are depicted in dashed lines. Solid lines represent cross-links between BORC and BLOC-1 subunits with those under 25 Å Cα–Cα distance in red and those above 25 Å in blue. (B) Zoom view of the central region of mixed complex A depicted in the previous panel. (C) Cross-links between BORC and BLOC-1 components exceeding Cα–Cα distances of 25 Å are represented in black and connect flexible terminal residues of DTNBP1 and BLOC1S5 to the core of the helix bundle. Structured rigid regions (pLDDT > 59) are colored in blue and flexible or disordered regions (pLDDT < 59) in red. (D) Schematic representation of the cross-links between EARP subunits and BORC/BLOC-1 components found in the SH-BORCS6 and BORCS7SH AP-XL-MS experiments. Line thickness correlates with the number of cross-links. (E) Schematic representation of the transferrin recycling experiment, (F) Indirect immunofluorescence images of HeLa wild type, BORCS5KO, BORCS6KO, and BORCS7KO cells at different time points post internalization of labeled transferrin. Merged and single-channel images of transferrin-Alexa488 (green) and EEA1 are indicated. Representative cells are shown. n = 3. (Scale bar, 10 μm.) (G) Graphic depicts the colocalization rate between labeled transferrin and endogenous EEA1 in the different cell lines, 10 cells per genotype were analyzed, (H) Graphical representation of the levels of transferrin-Alexa488 found in the cells normalized to the levels of labeled transferrin at time point 0 of Wild type HeLa, 10 cells per genotype were analyzed. (I) Graphical representation of the putative role of mixed BORC/BLOC-1 complexes in transferrin recycling.

To investigate potential physiological roles of hybrid BLOC-1/BORC complexes, we systematically analyzed differentially enriched interactors identified in our proteomic profiling of individual subunits (Fig. 3E and SI Appendix, Fig. S4E). The proteome of BLOC1S6 contained 102 high-confidence interactors, with expected overlap with BLOC-1 components (55 shared with Snapin, 40 with BLOC1S1; SI Appendix, Fig. S5F). Remarkably, BLOC1S6 shared 30-54 interactors with BORCS5 or BORCS8, revealing a highly interconnected network among these evolutionarily conserved complexes. This robust overlap provides strong biochemical evidence for endogenous BLOC-1/BORC hybrid assemblies, suggesting functional integration beyond their canonical roles. Gene ontology (GO) term enrichment analysis of the interactomes revealed the presence of bona fide partners of either complex (e.g., Ragulator and Folliculin/FNIP in the case of BORC, AP3 complex for BLOC-1) (SI Appendix, Fig. S5G). Interestingly, we also detected AP3 components as enriched GO term in the BORCS6 interactome. The most striking finding of the analysis was the presence of EARP/GARP subunits in the interactomes of both BLOC1S6 and the BORC specific subunits BORCS6 and BORCS7. EARP/GARP are two related complexes associated with tethering containing helical rods (CATCHR) that control endosome to Golgi transport (GARP) or cargo recycling from endosomes (EARP) (28, 29). In our interactome analysis we detected Vps51, Vps52, and Vps53 but neither Vps50 nor Vps54. The latter two are the subunits that discriminate EARP from GARP. We then revisited our XL-MS experiments to determine whether cross-links could be detected between EARP/GARP and BORC or BLOC-1 subunits. Consistent with our hypothesis, we found 6 cross-linked peptides reporting interactions between BORCS5, BLOC1S6, DTNBP1, and the EARP subunits Vps50 and Vps53 (Fig. 4D). These results, obtained using tagged BORCS6 and BORCS7 as baits, are in line with the existence of another mixed complex, likely containing BORCS5, BORCS6, and BORCS7 as well as BLOC1S6 and DTNBP1.

The original characterization of EARP by Schindler et al. showed that depletion of syndetin/Vps50 disrupts transferrin receptor recycling, establishing its critical role in endosomal trafficking (28). Therefore, we tested whether depletion of the BORC subunits BORCS5, BORCS6, or BORCS7 would have a similar effect (Fig. 4 E–H). Pulse–chase experiments with labeled transferrin (Fig. 4 E–H) revealed indistinguishable uptake in all cell lines at t = 0 (Fig. 4H), indicating unperturbed endocytosis. After 30 min of chase, control cells showed reduced intracellular transferrin with decreased EEA1 colocalization (Fig. 4 G and H). In contrast, BORCS6 and BORCS7 KO cells retained elevated transferrin levels with increased EEA1 colocalization, while BORCS5 KO cells showed an intermediate phenotype. After 60 min, all lines reached comparable low levels of transferrin (Fig. 4H).

Upon arrival at early endosomes, transferrin can be recycled to the plasma membrane via two independent routes defined by the associated Rab proteins (30, 31). Rab4 mediates a fast recycling route (32, 33), whereas Rab11 regulates slow recycling to the plasma membrane that occurs via the endosomal recycling compartment (34, 35). Given the established role of EARP in Rab4-dependent rapid recycling (28), we asked whether BORC deficiency impairs transferrin delivery to Vps53-positive compartments (SI Appendix, Fig. S5 H and I). Wild-type cells exhibited two Vps53 pools: 1) perinuclear GARP complexes at Golgi membranes and 2) scattered vesicles showing partial transferrin colocalization at t = 0 (SI Appendix, Fig. S5 H and I)—likely Rab4-positive intermediates as reported (28). Selective disruption of Vps53/Rab4 compartment access, coupled with preserved Rab11-mediated recycling, underlies the kinetic dichotomy in BORC knockout cells: delayed early clearance (30 min) but uncompromised late transferrin retention (60 min). These results position BORC/BLOC-1 mixed assemblies as a critical arbiter of cargo sorting between fast (Rab4/EARP) and slow (recycling endosome) recycling pathways.

Discussion

Ge et al. recently reported the first cryoEM structure of BORC (21). Our study corroborates the overall architecture of the complex and provides a definitive subunit assignment within the C. elegans BORC structure. While Ge et al. used in vitro assays to determine the stability of BORC subcomplexes and propose a hierarchical assembly model, we focused on validating the structural framework in vivo. We demonstrated that octameric BORC stability depends on α-helical contacts between subunits. Systematic interface disruption dissociated the complex, and pathogenic mutations at specific sites impaired function, underscoring their essential role. Notably, we were also able to validate the structure of human BORC by XL-MS analysis.

While the flexible N- and C-terminal regions of BORC subunits are dispensable for core assembly, they harbor conserved motifs that are predicted to undergo binding-induced folding. These regions are critical for engaging key interactors, such as the small GTPase Arl8 and the Ragulator/LAMTOR complex (14, 15, 19, 36–38). Ge et al. demonstrated that a conserved sequence within the disordered region of BORCS5 adopts a β-strand conformation to interact with Arl8 (21). Similarly, the unstructured terminus of BORCS6 has been demonstrated to bind LAMTOR2 (15, 38). Together, results from us and others establish a bipartite architecture: rigid helical interfaces maintain BORC integrity, while disordered segments likely recruit effectors, explaining how the complex integrates stability with regulatory capacity.

As hypothesized by De Pace et al. (20), the presence of paralogs and multiple isoforms derived by alternative splicing is consistent with the formation of mixed BORC/BLOC-1 complexes. Our study now provides direct experimental evidence of such hybrid assemblies in two distinct cell lines. Specifically, we demonstrate the following: i) mutations that disrupt canonical BORC assembly also prevent interactions with specific BLOC-1 subunits, ii) the assembly of hybrid complexes depend on the availability of subunits and are affected by the depletion of BORC-specific subunits, iii) cross-complex incorporation is versatile and selective, and distinct BLOC-1 subunits preferentially associate with specific BORC components, and iv) affinity purification coupled with cross-linking mass spectrometry confirms the presence of a defined mixed BORC–BLOC-1 complex in vivo.

A review of the literature reveals several reports of interactions between BORC and BLOC-1 subunits, albeit fragmented. For example, during the initial characterization of BORC, it was found that DTNBP1 coimmunoprecipitated with BORCS6 (14). KxD1 was previously found to interact in vitro with the BLOC-1 specific components BLOC1S4 and DTNBP1 (39, 40). Moreover, BLOC1S5 coimmunoprecipitated BORCS6 while BORCS6 and BORCS8 associated with DTNBP1 (41). Finally, in Drosophila melanogaster DTNBP1 coimmunoprecipitated BORCS6 and KxD1 (42).

Our data consolidate and strengthen the support for the existence of multiple, distinct hybrid complexes that was established by earlier observations. To define the physiological role of these assemblies, we examined their interactome and found a connection to components of the Endosomal Assembly Recycling Protein (EARP) complex. We confirmed this interaction through cross-linking mass spectrometry (XL-MS) and functional assays. Based on these findings, we propose that specific BORC–BLOC-1 hybrid assemblies regulate the rapid recycling of transferrin from early endosomes via their interaction with EARP.

Functional modularity in BORC and BLOC-1 assemblies is further illustrated by subunit-specific requirements across species and cellular processes. In human cells, lysosomal size regulation depends on BORCS5 and BORCS7, but not BORCS6 (43). Notably, this functional profile aligns with the composition of the mixed complex B identified here. This suggests that this specific assembly may mediate lysosomal size control. Following the same lines, in C. elegans all BORC subunits contribute to lysosomal trafficking, but only a subset (excluding KxD1/BORCS7) is required for synaptic vesicle trafficking (22). Notably, the same dispensable subunits also only partially contribute to phagolysosomal vesiculation (23).

We propose that the formation of hybrid complexes is governed by structural plasticity and regulated by cellular levels of canonical BORC and BLOC-1, as well as the stoichiometry of individual subunits. Our in vivo data suggest that these mixed assemblies form by incorporating two subunits from one complex into a stable hexameric core from the other complex. This suggests that hexamers function as key transitional intermediates in the biogenesis of both canonical and hybrid octamers. Consistent with this model, we identified a stable hexameric species in vitro comprising all BORC subunits except BORCS5 and BORCS7.

Our cryoEM results identified regions of high flexibility in two BORC subunits, KxD1 and Snapin. The increased flexibility suggests that these two subunits are loosely bound to the rest of the BORC complex and may be critical for adapting the complex’s structure to physiological changes or to allow formation of mixed assemblies. KxD1 was previously described to be dispensable for synaptic vesicle trafficking and phagolysosomal vesiculation in C. elegans (22, 23). Snapin was shown to bind dynein motors thereby regulating retrograde transport of lysosomes in neuronal cells (44). This process is required to facilitate autophagic clearance, including the promotion of amyloid ß precursor processing in neurons (45). Snapin appears to perform this function independently of BLOC-1/BORC incorporation, as no other subunit of either complex has been shown to play an equivalent role. This functional autonomy implies that Snapin functions through distinct structural features that are neither acquired nor dependent on its incorporation into these multiprotein complexes. In particular, Snapin-mediated retrograde lysosomal trafficking counteracts anterograde transport driven by BORC/Arl8 (14). This bidirectional regulation suggests that the stoichiometric incorporation of Snapin into BORC, BLOC-1, and the levels of non-complex-bound Snapin may serve as a molecular determinant controlling lysosomal positioning. We propose that cells precisely modulate the partitioning of Snapin between these complexes to spatially regulate lysosome distribution, thereby fine-tuning location-dependent lysosomal functions. Such a mechanism, likely mediated through phosphorylation of specific residues, would provide cells with dynamic control over lysosome positioning while maintaining the integrity of the core transport machinery (46, 47).

Interestingly, mutagenesis of the BORCS5 H99 residue caused the dissociation of the two BORC tetramers. We demonstrated that the purified hemicomplexes are stable in vivo under these conditions. This finding directly validates the in vitro evidence from Ge et al. showing that the two halves of BORC can exist as stable, separate entities (21). Due to its side-chain pKa of ~6.0, histidine can exist in a positively charged, neutral, or negatively charged state, depending on the local pH. This property makes it a critical residue for pH sensing (48, 49). An increase in pH favors deprotonation, shifting histidine to a neutral or negatively charged state—a condition that can be mimicked by mutating it to glutamic acid (H99E). In our experiments, the BORCS5 H99E mutation prevented growth factor- and amino acid-dependent anterograde lysosomal movement and caused BORC dissociation into two hemicomplexes. In contrast, the BORCS5 H99A mutation, which introduces a nonionizable alanine residue, also impaired anterograde transport. Based on these results, we propose that cytosolic acidification promotes the protonation of BORCS5 H99. This protonated state would facilitate electrostatic attraction and promote hydrogen bond formation between BORCS5 H99 and the spatially aligned D41 residue of Snapin. We also found this pairing to be conserved across species. While this model has not been formally tested, it provides a direct mechanistic link between cytosolic pH and the stability of the anterograde lysosomal transport machinery.

Together, the structural work from our study and Ge et al. (21) provides a foundational understanding of the BORC complex assembly process. This architectural framework is further solidified by the recent confirmation of a similar elongated, rod-shaped BLOC-1 structure by Wang et al. (13). Our study goes beyond structure to validate the function of BORC assembly, determine the impact of disease-causing mutations on its mechanics, and identify regulatory mechanisms, such as Snapin phosphorylation and the pH-sensitive ionization of BORCS5 H99. Furthermore, our findings challenge the prevailing paradigm of BORC and BLOC-1 as independent entities by providing evidence of several hybrid complexes. We demonstrate that one such hybrid assembly regulates the rapid recycling of transferrin from early endosomes through its interaction with EARP/GARP components. The full repertoire and distinct physiological roles of other noncanonical BORC/BLOC-1 assemblies, as well as the signals that govern their formation and potential cell- or tissue-specificity, are compelling subjects for future investigation. We propose that the capacity for functional specialization through mixed complex formation is a fundamental and adaptable regulatory layer in the eukaryotic endolysosomal system.

Defects in the canonical BORC and BLOC-1 complexes have been linked to neurodevelopmental (17) and hypopigmentation diseases (50). Our finding of hybrid assemblies indicates that these subunits have an even broader functional repertoire, which could extend to other aspects of the endolysosomal system. This has profound implications for human disease. For example, BORC plays a pivotal role in the peripheral scattering of endolysosomes, which is a key driver of cancer cell invasiveness and is associated with poor patient prognosis (51–53). Concurrently, the critical function of BORC in neuronal lysosome positioning raises questions about its potential role in aging and neurodegeneration. Under this expanded paradigm, thoroughly discriminating the physiological functions performed by canonical complexes versus specific hybrid assemblies will be critical. The efficacy and specificity of future pharmacological interventions, whether for cancer, neurological conditions, or pigmentary disorders (20), will depend on our ability to target the correct complex. Therefore, unraveling the unique biology of these mixed assemblies is essential for designing the best therapeutic strategy for each disease and is not just a mechanistic refinement.

Materials and Methods

Constructs for Recombinant Expression of C. elegans BORC and BLOC-1 components.

The E. coli expression construct was designed on the basis of the pET System (T7 control) and adapted for polycistronic expression. In brief, the pET23 backbone carries a single synthetic gene, representing a polycistronic mRNA encoding eight open reading frames (ORFs) (SI Appendix, Fig. S2A). All ORFs are flanked with individual translational enhancer signals (ε) and ribosomal binding sites (SD). Intervening sequences have been varied as much as possible to avoid recombination events. In addition, the individual ORFs were codon optimized for E.coli expression. We removed the lipidation motif and included SUMO and His₆ at the N terminus of BORCS5, the anchor protein of the BORC complex (14). Finally, we included a C terminal TEV cleavage site followed by a linker and tandem STREP-tag on BORCS6. This strategy was based on our experience on the purification of BORC from human cells (38). Of note, the assigned ORFs order is identical to that of the pST39 vector containing BORC previously published by Pu et al. (14). The C. elegans BORC construct was transformed into BL21DE3, sequence verified, and stored as a glycerol stock. For the generation of constructs expressing only six out of the eight BORC subunits, the order of the subunits present and their respective regulatory regions was maintained. The hexameric versions were transformed into C41, sequence verified, and stored as glycerol stocks.

Protein Expression and Purification of C. elegans BORC.

The glycerol stock was used to inoculate 5 mL Luria Broth (LB) supplemented with Ampicillin to an end concentration of 100 µg/mL. The culture was left to grow for 6 to 7 h at 37 °C, 180 rpm, on an Innova 44 shaker. For the overnight culture, 50 µl of the initial suspension were used to inoculate 50 mL of LB supplemented with Ampicillin. Cultivation conditions were kept as stated before. The following day, the culture volume was expanded again via a 1:100 dilution in fresh LB supplemented with Ampicillin. The culture was maintained at 37 °C, 180 rpm until OD₆₀₀ reached 0.45 to 0.5. The large-scale culture was subsequently induced with 500 mM IPTG and left to grow for another 4 to 6 h.

In order to control for any recombination events, the preculture and the large-scale amplification were checked prior to protein purification. In brief, 1 mL of the respective bacteria suspensions was removed, centrifuged, and stored at –20 °C. Upon extraction, DNA was digested with BamHI for 3 h at 37 °C. Cultivation was deemed successful if the pattern of the bands in the agarose gel matched the theoretical sizes predicted.

The octameric C. elegans BORC complex was purified by Strep-based affinity chromatography followed by size exclusion chromatography. The cell pellet was resuspended in lysis buffer (50 mM HEPES pH 7.5, 150 mM NaCl, 0.5 mM TCEP) supplemented with 0.5% Tween20, 1 mg/mL Lysozyme, 10 µg/mL DNAse I 10 µg/mL aprotinin, 1 µg/mL pepstatin, 10 µg/mL leupeptin, and 1 mM pefabloc and left to thaw rocking at room temperature for 15 min. Subsequently, the lysate was transferred to ice and incubated for another 20 min to fully complete lysis. The lysate was sonicated and cleared by centrifugation at 20,000×g for 30 min and filtered through a 0.22 µm nitrocellulose membrane. Cleared lysate was loaded to a Strep-Tactin® column, preequilibrated in lysis buffer. Upon loading of the sample, the column was washed with 10CV of lysis buffer, followed by 8CV of lysis buffer supplemented with 2 mM ATP and 2mM Mg₂Cl₂ and finally another 2CV of lysis buffer. Proteins were eluted in lysis buffer supplemented with 2.5 mM D-desthiobiotin. Fractions containing the complex were pooled and subjected to dialysis in 10 mM HEPES pH 7.5, 150 mM NaCl, 0.5 mM TCEP, o/n at 4 °C. When appropriate, SUMO and TEV proteases (SIGMA) were added to the eluate during the dialysis procedure. Finally, the sample was further optimized by SEC using a Superdex 200 increase 10/300 column (GE Healthcare) equilibrated in dialysis buffer. Selected fractions were pooled and concentration was adjusted to 1 mg/mL using a 30 kDa cut-off concentrator.

Grid Preparation.

On a Leica EM GP (4 °C, 75% humidity), 3 µL of the sample was applied to a holey carbon R2/2 Cu200 grid (Quantifoil), which was plasma-cleaned using an SCD005 sputter coater (Bal-Tec, 60 s, 25 mA). The grids were blotted for 2 s (Whatman grade 1) using the Leica EM GP automated blotting feature, and plunge-frozen into liquid ethane.

CryoEM Screening and Data Collection.

CryoEM grids were screened on a 200 kV Glacios TEM (Thermo Fisher) at the Vienna Biocenter Core Facilities (VBCF) EM facility. Grids with good ice quality and BORC particles were selected for data collection on a 300 kV Titan Krios G3i TEM (Thermo Fisher) at Institute of Science and Technology Austria (ISTA). This instrument was equipped with an X-FEG electron source, a BioQuantum energy filter (Gatan) with a 20 eV slit width, and a K3 direct electron detector (Gatan). Data collection was performed with EPU, using 40 frames per exposure, a pixel size of 1.07 Ų/px, a total dose of 50 e⁻/Ų, and a defocus range of −1.0 to −3.0 µm.

CryoEM Image Processing.

The BORC dataset was processed using CryoSPARC v3.2. Initially, the movies were preprocessed using Patch Motion Correction and the CTF of the motion-corrected micrographs were estimated using Patch CTF Estimation. An initial small particle set was manually picked and submitted for 2D classification. 2D class averages with high contrast were used for Particle Template Picking. The identified particles were extracted with a box size of 450 px (481.5 Å) and were Fourier cropped to 300 px (1.605 Ų/px). After 4 iterations of 2D class cleaning ab initio Reconstruction was performed and used as input for homogeneous refinement. The refined map was postprocessed with DeepEMhancer.

Additional Material and Methods.

The description of additional material and methods is provided in SI. It includes information regarding protein expression and purification of BORC subcomplexes, cross-linking mass spectrometry analysis (XL-MS) of C. elegans BORC, two-dimensional gel electrophoresis, negative staining of purified BORC, cell culture and generation of cell lines, cell lysis and western blotting, affinity purification, mass spectrometry analysis of purified complexes, cross-linking mass spectrometry, immunofluorescence microscopy, and the transferrin recycling assay.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

The mass spectrometry proteomics data were deposited to the ProteomeXchange Consortium, (http://Proteomecentral.proteomexchange.org) via the Proteomics identification database partner repository (PRIDE) with the dataset identifier (54). The Cryo-EM density map was deposited in the Electron Microscopy data bank with the accession nr. EMD-56129 (55). Structural coordinates were deposited in the PDB with accession code PDB 9TQB. Publicly available at wwPDB deposition (http://deposit.wwpdb.org) (56). Remaining study data are included in the article and/or SI Appendix.