HDLBP binds ER-targeted mRNAs by multivalent interactions to promote protein synthesis of transmembr

The biological role of RNA-binding proteins in the secretory pathway is not well established. Here, we describe that human HDLBP/Vigilin directly interacts with more than 80% of ER-localized mRNAs. PAR-CLIP analysis reveals that these transcripts represent high affinity HDLBP substrates and are specifically bound in their coding sequences (CDS), in contrast to CDS/3’UTR-bound cytosolic mRNAs. HDLBP crosslinks strongly to long CU-rich motifs, which frequently reside in CDS of ER-localized mRNAs and result in high affinity multivalent interactions. In addition to HDLBP-ncRNA interactome, quantification of HDLBP-proximal proteome confirms association with components of the translational apparatus and the signal recognition particle. Absence of HDLBP results in decreased translation efficiency of HDLBP target mRNAs, impaired protein synthesis and secretion in model cell lines, as well as decreased tumor growth in a lung cancer mouse model. These results highlight a general function for HDLBP in the translation of ER-localized mRNAs and its relevance for tumor progression.

In eukaryotic cells, the localization of functional protein products is largely determined by the site of their translation. While soluble proteins are translated in the cytosol, co-translational targeting to the endoplasmic reticulum (ER) enables newly synthesized proteins to enter the secretory pathway, resulting in their secretion or membrane integration^1,2,3,4. The canonical secretory pathway initiates in the cytosol with the synthesis of the hydrophobic targeting signal (signal peptide or transmembrane domain)⁵. Subsequent binding of the signal recognition particle (SRP) to the nascent peptide results in ribosome elongation arrest and formation of the ribosome nascent chain complexes (RNCs)⁶. This allows the re-localization of the cytosolic SRP-RNC to the ER membrane via the SRP receptor and translocation of the nascent peptide to the ER lumen⁷. In recent years, a non-canonical SRP-independent pathway was discovered in yeast⁸ along with the evidence for the recruitment of the SRP to the mRNA prior to ribosome engagement⁹ and SRP-independent ER targeting^10,11,12. This raises the possibility of the existence of yet unknown mechanisms for the recognition of membrane-bound mRNAs.

The potential role of regulatory elements in mRNA sequences for ribosome elongation arrest and nascent chain recognition are poorly understood. Several studies have identified elements within coding sequences (CDS) and 3’ untranslated regions (3′ UTRs) that may distinguish ER-bound from cytosolic mRNAs^{13,14,15,16,17}. However, trans-acting factors that may be responsible for the recognition of such elements are unknown. Recently, it was observed that a small subset of mRNAs that encode soluble proteins may also be localized and translated at the ER, indicating additional mechanisms that regulate the fate of a localized mRNA¹⁸. While there is evidence for subpools of ER-associated ribosomes that interact with pyruvate kinase in muscle¹⁹, comprehensive differences in the composition, assembly and active translation states of cytosolic and ER-bound ribosomes have not been identified^20,21,22. Furthermore, a previously unknown variant of the ribosome-dependent nonsense-mediated decay (NMD) pathway was discovered at the ER, hinting at an additional layer of regulation for ER-bound mRNAs²³. In summary, the translational fate of mRNAs encoding soluble and membrane proteins may be tightly regulated by trans-acting factors such as RNA-binding proteins, which could function beyond the canonical SRP-dependent model.

HDLBP (also known as VIGILIN) is a conserved and ubiquitously expressed RBP localized both to the cytosol and the ER membrane^24,25. It contains 15 hnRNPK-homology (KH) RNA-binding domains (RBDs)²⁶. KH domains are high-affinity RNA-recognition elements (RREs), most commonly tetranucleotides as observed for FMRP²⁷, SF1, HNRNPK, and others²⁸. Some may also recognize bipartite motifs, e.g., IGF2BP protein family^29,30,31,32. HDLBP and its yeast orthologue SCP160 have been found to contribute to many biological processes³³ such as translation^34,35 or protein aggregation³⁶, and have been linked to carcinogenesis^37,38. Recently, HDLBP has been shown to be required for replication of flaviviruses ZIKV and DENV, most likely by increasing the translation efficiency of viral proteins at the ER²⁴. HDLBP is also a promising target for cardiovascular research, since it promotes secretion of very-low-density lipoprotein (VLDL) and leads to less atherosclerotic plaques upon hepatic HDLBP knockdown in atherosclerosis prone Ldlr^−/− mice³⁵. It was proposed that HDLBP binds to CHHC or CHYC (H = A/C/U and Y = C/U) containing regions in a subset of mRNAs encoding for secreted proteins and enhances their translation. However, functional aspects of HDLBP binding to RNA and mechanistic events during translation remain uncertain.

Here, we assayed HDLBP binding sites in a transcriptome-wide manner by PAR-CLIP and discovered their potential function as selective sequence determinants of ER-bound mRNAs. HDLBP directly and specifically interacted with more than 80% of all ER-localized mRNAs and was primarily bound to long CU-rich motifs in their CDS, a unique feature which is much more frequently found in membrane-bound compared to cytosolic mRNAs. Biochemical, transcriptomic, and proteomic methods were used to evaluate the functional consequences of HDLBP absence on ER translational efficiency, protein synthesis, and secretion and highlighted its requirement for these biological processes. Finally, we expanded our findings to an in vivo system and evaluated the effect of the absence of HDLBP on tumor formation capacity.

HDLBP directly interacts with ER-targeted mRNAs

To functionally characterize the interactions of HDLBP with the localized transcriptome, we quantified cytosolic and membrane-bound mRNAs in HEK293 cells. Using a subcellular chemical fractionation approach³⁹, we obtained cytosol and membrane fractions, as evidenced by the presence or absence of compartment-specific protein markers (Supplementary Fig. 1a). To obtain localized mRNA profiles, we next performed mRNA-seq from whole-cell, cytosol, and membrane fractions (Supplementary Fig. 1b), resulting in highly reproducible mRNA abundance quantifications (Supplementary Fig. 1c). Due to the bimodal distribution of membrane-to-cytosol enrichment (Fig. 1a), we were able to classify over 7000 mRNAs according to their partitioning between the cytosol and membrane (Supplemental Data 1). To validate this result, we compared the enrichment values between mRNAs that encoded or lacked co-translational targeting signals. As expected, we found that mitochondrial DNA-encoded proteins, as well as signal peptide (SP) and transmembrane helices (TM) encoding mRNAs had much higher membrane enrichment when compared to mRNAs encoding post-translationally targeted tail-anchored proteins or nuclear DNA-encoded mitochondrial proteins (Fig. 1b), validating our approach.

a Quantification of membrane and cytosolic mRNA localization in HEK293 cells by sequencing. Histogram of log2-transformed membrane enrichment (ratio between the membrane and cytosolic read counts, mean of two replicates). Cutoffs of 0.5 and 1.5 (gray lines) were chosen to classify membrane-bound (n = 1157), cytosolic (n = 6139) and mRNAs with undefined localization (n = 193). mRNAs with a TPM of at least ten were included in this analysis. b Quality control of mRNA-seq-derived localization. Log2-transformed membrane enrichment was compared between groups of mRNAs classified according to presence of encoded targeting signals (TS): cytosol-localized mRNAs with no targeting signals (n = 5142), mRNAs encoding mitochondrial proteins excluding mtDNA-encoded mRNAs (n = 739), mRNAs coding for post-translationally targeted tail-anchored transmembrane proteins (n = 102), mitochondrial DNA expressed mRNAs encoding mitochondrial proteins (n = 13), membrane-localized mRNAs with no known TS (n = 107), mRNAs encoding signal peptide (SignalP) containing proteins (n = 214) or transmembrane helices (n = 724) or both (n = 282). mRNAs with a TPM of at least ten were included in this analysis. The lower and upper hinges of box plots correspond to the 25th and 75th percentiles, respectively. Upper and lower whiskers extend from the hinge to the largest or smallest value no further than the 1.5× interquartile range from the hinge, respectively. Center lines of box plots depict the median values. c Distribution of HDLBP crosslinks (T-C transitions) throughout the transcriptome as detected by PAR-CLIP in HEK293 cells. d Scatter plot of normalized HDLBP cross-linking signal in CDS versus mRNA expression level in non-fractionated cells. mRNAs were classified according to their membrane enrichment in (a). Inset above shows density distributions of HDLBP cross-linking signal. e HDLBP-bound and unbound mRNAs with respect to their localization. Absolute numbers and percentages of HDLBP-bound mRNAs of total HEK293 cytosolic and membrane-bound mRNAs are shown. f HDLBP PAR-CLIP binding enrichment (mean of two biological replicates) for different mRNA classes; cytosol-localized mRNAs with no TS (n = 1578), mRNAs encoding mitochondrial proteins excluding mtDNA-expressed mRNAs (n = 251), mRNAs coding for post-translationally targeted tail-anchored transmembrane proteins (n = 30), membrane-localized mRNAs with no known TS (n = 74), mRNAs encoding signal peptide (SignalP) containing proteins (n = 138) or transmembrane helices (n = 492) or both (n = 219). The lower and upper hinges of box plots correspond to the 25th and 75th percentiles, respectively. Upper and lower whiskers extend from the hinge to the largest or smallest value no further than the 1.5× interquartile range from the hinge, respectively. Center lines of box plots depict the median values. a–f Source data are provided as a Source Data file.

Full size image

Successful quantification of mRNA localization prompted us to determine the whole-cell HDLBP-bound transcriptome in HEK293 cells by PAR-CLIP (Supplementary Fig. 1d, e). The majority of FLAG/HA-HDLBP cross-linking signal, detected as T-C transitions, was present in coding sequences (CDS) and 3′ untranslated regions (3′ UTRs) of mRNAs (Supplementary Data 2, 3), as well as in rRNA and tRNAs (Fig. 1c). After expression-level normalization (Supplementary Fig. 1f, g), the PAR-CLIP T-C signal of membrane-bound mRNAs was normally distributed without mRNA abundance bias (Fig. 1d), suggesting high specificity of HDLBP interactions for this pool of mRNAs. Interestingly, high HDLBP PAR-CLIP enrichment in membrane-localized mRNAs was mostly due to high CDS interactions rather than 3′ UTR binding (Supplementary Fig. 1h). We estimated that more than 80% of the total membrane-bound mRNA pool was bound by HDLBP (Fig. 1e). The unbound membrane-localized mRNAs included 13 mitochondrial DNA-encoded genes (Supplementary Fig. 1i), which were not expected to be HDLBP targets since HDLBP is not localized to mitochondria²⁵. Correspondingly, HDLBP-bound mRNAs were highly enriched for SP and TM helix encoding transcripts, while a significantly weaker binding was observed for cytosolic mRNAs and those encoding mitochondrial proteins (Fig. 1f). In addition, we found that mRNAs encoding tail-anchored membrane proteins were not enriched, confirming that HDLBP binds to the endoplasmic reticulum (ER)-targeted mRNAs with high specificity.

Differential HDLBP binding to transcript regions of membrane-bound and cytosolic mRNAs

To determine the molecular characteristics of HDLBP–RNA interactions, we first examined the relative distribution of HDLBP cross-linking positions within mRNA transcript regions. High reproducibility of T-C positional counts was observed (Supplementary Fig. 2a), allowing us to construct high-resolution maps of HDLBP cross-linking profiles. We observed reciprocal binding of HDLBP to CDS and 3′ UTR regions depending on mRNA localization. HDLBP interacted predominantly with the CDS of membrane-bound mRNAs, whereas for cytosolic mRNAs the binding was more prominent in the 3′ UTR (Fig. 2a). To address the relationship between CDS, 3′ UTR binding, and mRNA localization, we calculated the ratio between length-normalized T-C transitions detected in CDS versus 3′ UTR and found that membrane-localized mRNAs had higher ratios compared to cytosolic transcripts (Fig. 2b). In addition, the contribution to the total T-C crosslink signal mostly originated from high CDS binding for membrane-bound mRNAs (Fig. 2c). Comparison between T-C transition patterns in membrane-bound mRNAs IGF2R and COL2A1, and cytosolic HNRNPUL1 and FTL transcripts (Fig. 2d and Supplementary Fig. 2b), supported the conclusions from the transcriptome-wide analysis. In summary, HDLBP target mRNAs are primarily membrane-localized and bound in the CDS, while in cytosolic mRNAs HDLBP binding is on average equal between CDS and 3′ UTR.

a Meta-transcript analysis of HDLBP cross-linking signal for cytosolic and membrane-bound mRNAs. T-C transition signal per nucleotide was normalized for library size (T-C per million). Transcripts with at least 5 T-C per million were included in this analysis. T-C per million was scaled to the maximum T-C signal per transcript. At each position, the mean scaled T-C signal was plotted. b Scatter plot of the ratio of HDLBP crosslinks in the CDS vs. 3′ UTR plotted against total crosslinks per mRNA for cytosolic and membrane-bound mRNAs. c mRNAs were split into three groups according to total HDLBP crosslinks per mRNA (n = 16: TC <0.3, membrane, n = 65: 0.3 < TC <1.39, membrane, n = 35:, TC >1.39, membrane, n = 279: TC <0.3, cytosolic, n = 379: 0.3 < TC <1.39, cytosolic, n = 144: TC >1.39, cytosolic) and the distributions of the CDS vs. 3′ UTR ratio of HDLBP crosslinks were plotted. The lower and upper hinges of box plots correspond to the 25th and 75th percentiles, respectively. Upper and lower whiskers extend from the hinge to the largest or smallest value no further than the 1.5× interquartile range from the hinge, respectively. Center lines of box plots depict the median values. d Browser representation of PAR-CLIP read coverage and T-C transitions in IGF2R (membrane-bound) and HNRNPUL1 (cytosolic) mRNAs. Reads were mapped to human mRNA sequences. 5′ UTR, CDS, and 3′ UTR regions are indicated. Transcript IDs are indicated. a–d Source data are provided as a Source Data file.

Full size image

RNA-recognition elements for HDLBP binding are more common in membrane-bound mRNAs

We next explored the primary sequence characteristics underlining the specificity of HDLBP for membrane-bound mRNAs. We ranked crosslinked seven-mers by their median crosslink signal per transcript (Supplementary Fig. 3a) and by the frequency of T-C crosslink positions relative to all crosslinked seven-mers (Fig. 3a). These two metrics were compared between CDS and 3′ UTR of mRNAs either localized to the membrane or the cytosol. We found that HDLBP is most frequently crosslinked to CU-containing seven-mers with a variable number of UUC/UC/CUU/CU repeats, located in the CDS of membrane-bound mRNAs (Fig. 3a, b and Supplementary Fig. 3a).

a Frequency of top ten HDLBP crosslinked seven-mers located either in 3′ UTR or CDS of membrane-bound and cytosolic mRNAs. b Sequence logo of top five HDLBP crosslinked seven-mers ranked according to their frequency among all detected crosslinked seven-mers. c Distribution of z-scores calculated from differences in the frequency of all possible k-mers within membrane-bound and cytosolic CDS and 3′ UTR sequences. For each k-mer length, this analysis was performed for the group of top 40 HDLBP crosslinked k-mers (left) and all other k-mers (right). d pvalues of pairwise Wilcoxon rank-sum test between z-scores obtained for top 40 bound HDLBP k-mers as described in (c). e HDLBP multivalency analysis in +40/−40 nt regions around cross-linking sites. To evaluate the T-C binding affinity as a function of multivalency of HDLBP binding sites, we binned the multivalency scores for the top ten enriched four-mers within the 40-nt regions into five categories (group sized from highest to lowest score groups, n = 994, n = 973, n = 673, n = 1036, n = 1308). The total normalized T-C transition signal over the +40/−40 nt regions was then plotted for all five categories. Lower and upper hinges of box plots correspond to the 25th and 75th percentiles, respectively. Upper and lower whiskers extend from the hinge to the largest or smallest value no further than the 1.5× interquartile range from the hinge, respectively. Center lines of box plots depict the median values. f Analysis of the percentage of total T-C transitions for every nucleotide position within the +40/−40 nt region for each multivalency bin. The two bins with the highest multivalency scores are shown and correspond to (e). gComparison of mean multivalency scores between differentially localized mRNAs in their CDS. A positive set (a four-mer group consisting of top ten HDLBP crosslinked four-mers, UUCU) and a negative set (AAGU) with no HDLBP enrichment. Occurrence of these four-mer groups were counted in 30-nt sliding windows and the mean score per transcript was computed. Mean distribution was then compared between different localized CDS by Wilcoxon rank-sum tests. h Apparent dissociation constants of recombinant GST-HDLBP fragments (constructs A though D) and full-length protein (FL), schematically shown, for different RNA oligonucleotides as determined by fluorescence anisotropy binding assays. Dissociation constants that were measured but could not be determined are indicated with “n.d.” and interaction that were not measured by “—“. i Fluorescence anisotropy binding assays for RNA oligonucleotides with different number of HDLBP binding four-mers (left, H40-44). GST-HDLBP construct B (KH5-9) was incubated with FAM-labeled RNA oligonucleotides, anisotropy measured and K_D determined from the binding curves (middle panel). For each oligonucleotide, three independent K_d values were determined. Significant differences in K_d values were evaluated using two-sided t-test and are indicated with asterisks (*P < 0.01, *P < 0.05). Data were presented as mean values ±  SD. a–i Source data are provided as a Source Data file.

Full size image

We asked if HDLBP binding is determined by the differential sequence composition of cytosolic and membrane-bound mRNAs. For this purpose, we determined the frequency of all possible seven-mers within whole 3′ UTR and CDS sequences and compared it to the frequency of HDLBP crosslinked seven-mers. While the top HDLBP crosslinked seven-mers were not among the most frequent seven-mers in the transcriptome, they showed a significantly higher occurrence (P = 2.7e-07) in the CDS of membrane-bound compared to cytosolic mRNAs (Supplementary Fig. 3b).

Since HDLBP is composed of 15 KH domains^26,33, we next addressed the possibility that it could recognize longer RREs⁴⁰. Therefore, the difference in the frequency between membrane-bound and cytosolic mRNAs was calculated for the 40 most highly crosslinked k-mers with a length between 4 and 12 nucleotides. Crosslinked k-mers were generally more frequent in sequences of membrane-bound mRNAs than in cytosolic mRNAs (Fig. 3c). In addition, the greatest difference between membrane-bound and cytosolic mRNAs was observed for the longest k-mers (10–12 nt) (Fig. 3c). Therefore, membrane-bound mRNAs contained a significantly higher number of longer high-affinity RREs for HDLBP binding than cytosolic mRNAs (Fig. 3d). On the other hand, the occurrence of crosslinked k-mers was comparable between CDS and 3′ UTR of cytosolic mRNAs (Supplementary Fig. 3c, d), and corresponded to the equal distribution of HDLBP crosslinks between these two transcript regions (Fig. 2c). In summary, the specific binding of HDLBP to the CDS of membrane-bound mRNAs can at least in part be explained by the differential k-mer composition of membrane-bound and cytosolic mRNAs.

High-affinity multivalent HDLBP interactions are more frequently formed in membrane-bound mRNAs

Due to the high number of KH domains, we next reasoned that HDLBP could recognize RREs interspaced with unbound nucleotides, giving rise to multivalent interactions. By counting the frequency of most frequently crosslinked four-mers in 40 nt regions around all detected crosslink positions (Supplementary Fig. 3f), we found that the regions with the highest four-mer frequency also showed the highest crosslink values within these regions (Fig. 3e). Therefore, the highly multivalent HDLBP binding sites also resulted in high-affinity interactions. In addition, the average positional crosslink signal within regions with the highest multivalency revealed that several positions upstream (−13 nt) and downstream (+16nt and +20nt) of the crosslink were specifically crosslinked (Fig. 3f). Therefore, high-affinity HDLBP sites contain 3–4 RNA elements positioned several nucleotides apart giving rise to binding sites of ~40 nt in length. Notably, multivalent interactions resulted in high-affinity interactions both in the CDS and 3′ UTR of membrane-bound and cytosolic mRNAs, respectively (Supplementary Fig. 3e).

To explore the potential of membrane-bound and cytosolic mRNAs to form multivalent interactions with HDLBP, we next calculated the occurrence of four-mers within 30-nt sliding windows of mRNA sequences. We observed that the local frequencies of HDLBP crosslinked four-mers were significantly higher in CDS of membrane-bound than in cytosolic mRNAs (Fig. 3g). This effect was absent for a group of unbound four-mers, confirming that membrane-bound CDS contain a high density of local HDLBP recognition elements that give rise to multivalent interactions.

To determine the RNA substrate specificity in more detail, we expressed and purified recombinant full-length HDLBP (FL) and protein variants (A–D) harboring different sets of KH domains (Fig. 3h and Supplementary Fig. 3i). Using a fluorescent anisotropy assay, we determined the apparent dissociation constants (Kd) for the respective protein–RNA substrate combinations, which are summarized in Fig. 3h. The full-length HDLBP bound with high-affinity to regions of TFRC mRNA (TFRC_1 and _2), which were selected based on our PAR-CLIP data. Mutations in one of the TFRC sites (TFRC_1_mut) reduced binding. Using artificial sequences, we could show that HDLBP interacted strongly with CCU- and CUU-oligomers, whereas affinity to a CAA oligomer was about tenfold reduced. In addition, HDLBP, albeit with reduced affinity, is bound to a region in the ES6 region of the 18 S rRNA. Interestingly, HDLBP construct B, comprising KH domains 5 through 9, bound with nearly similar affinities to the tested mRNA and 18S rRNA sequences as the full-length protein, suggesting that the central part of HDLBP is important for recognition and interaction with different RNA substrates. The 5 C-terminal KH domains HDLBP showed reduced binding to the TFRC mRNA region when compared to full-length HDLBP. For the N-terminal HDLBP fragment A (KH0-KH4) we were unable to identify a high-affinity RNA substrate.

To understand the impact of multivalent interactions on HDLBP binding affinity, we next performed the in vitro HDLBP binding assay using RNA molecules with different numbers of bound four-mer motifs within a longer (34 nt) sequence (Fig. 3i). We determined significantly lower Kd values for synthetic RNA molecules containing three HDLBP-bound four-mers (CUUC or UCUU) than those that harbored only 2. Therefore, HDLBP (KH5-9) has ~3–5-fold higher affinity for RNAs with higher multivalent potential. These findings strongly support our conclusions from PAR-CLIP analysis and confirm that HDLBP binds with high affinity to long RNA regions with multivalent interactions.

HDLBP interacts with the translational apparatus

Since HDLBP showed specific interactions with the CDS of membrane-bound mRNAs, we next explored the possibility that it plays a role in the translational regulation of ER-targeted mRNAs. HDLBP PAR-CLIP showed binding to 18S rRNA with two major binding sites located in expansion segment 6SB and helix 16 (Fig. 4a). These sites likely play a regulatory role in translation initiation and elongation^41,42due to their proximity to the binding sites of eukaryotic initiation and elongation factors (EIF4A, EIF4G1, and EEF2) (Supplementary Fig. 4a). Next, we examined HDLBP binding to the 7SL RNA component of the signal recognition particle (SRP), an RNP required for co-translational targeting of nascent polypeptide-associated complexes to the ER membrane. In 7SL RNA, we observed HDLBP contacts that were distinct from the binding patterns of other RBPs (IRE1, SSB, and MOV10) (Fig. S4B). HDLBP crosslinks were located in helices (5d–f, 6, 8a) of the large (S) domain and at the unpaired uridines of the small Alu region (Supplementary Fig. 4c). The percentage of SRP over total T-C crosslinks was comparable to other known or expected 7SL RNA interactors (IRE1 and SSB) and ~10-fold higher than for MOV10, which does not strongly bind 7SL RNA (Supplementary Fig. 4d). In summary, the HDLBP interacts with the 40S subunit of the ribosome at positions that are also close to the SRP contact sites (Fig. 4b).

a PAR-CLIP coverage and crosslinks detected in pre-rRNA regions. For comparison, results for IRE1 PAR-CLIP are included. Expansion segment positions are indicated in green. b Structures of the human 80S ribosome (PDB: 4V6X) and the SRP-ribosome complex (PDB: 3JAJ) were juxtaposed and HDLBP rRNA and 7SL RNA crosslinked nucleotides were mapped (indicated in red). c RNA immunoprecipitation was performed with FLAG/HA-HDLBP as bait. Co-precipitating RNAs were detected by qRT-PCR. Average fold enrichment (anti-FLAG vs. IgG control) from four replicates was calculated with error bars indicating standard deviation. Results are shown for 7SL RNA, IGF2R, YWHAZ, CD46, and ATP1A1 are shown, along with the mtDNA-encoded mRNA (MT-CO1) as a negative control. d BioID analysis of proteins in proximity to BirA-FLAG-HDLBP. The top 60 enriched proteins (LFQ(Dox) vs. LFQ(noDox) >3) were ranked according to mean LFQ (three replicates of Dox samples). The size of the dot corresponds to the enrichment value. e Gene Ontology enrichment analysis of 249 enriched BioID proteins. Adjusted p values for the top five enriched categories are shown. f FLAG/HA-HDLBP was co-immunoprecipitated with either anti-FLAG or IgG antibodies. Western analysis of input lysates (0.25%) and eluates (19%) was performed with antibodies as indicated. a–f Source data are provided as a Source Data file.

Full size image

In order to ensure that these contacts resulted from stable interactions, we confirmed various HDLBP RNA targets by RNA immunoprecipitation (RIP) experiments (Fig. 4c). Results corresponded to the conclusions of PAR-CLIP experiments with 7SL RNA showing a moderate enrichment, while several membrane-bound mRNAs (YWHAZ, ATP1A1, CD46, and IGF2R) showed high enrichment, whereas an mtDNA-encoded mRNA (MT-CO1) showed no enrichment. These results supported the conclusion that HDLBP forms stable interactions with RNA species identified by PAR-CLIP.

To validate the interaction with the translational apparatus by an orthogonal method, we next profiled HDLBP-proximal proteins using BioID⁴³, a proximity labeling assay in HEK293 cells expressing HDLBP fused to a promiscuous biotin ligase BirA* (Supplementary Fig. 1d). We confirmed by RIP that BirA*-FLAG-HDLBP was also bound to several transcripts that we previously detected to be bound by the FLAG/HA-tagged HDLBP (Supplementary Fig. 4f). Reproducible identification of HDLBP-proximal proteins (Supplementary Fig. 4e, g and Supplementary Data 4) overlapped ~50 % with the previously published BioID experiment (Supplementary Fig. 4h)⁴⁴ (Fig. 4d, e). We found that the top enriched proteins were involved in translation and included translation initiation factors (EIF4G1, EIF4B, EIF5, and EIF4E2) (Fig. 4d, e and Supplementary Fig. 4i), chaperones and chaperonins (HSPA1A, HSPA8, and CCT8), SRP components (SRP68) and ribosomal proteins (RPS3A and RPS10) of the small subunit (SSU). We validated the potential SSU interaction by orthogonal anti-FLAG co-immunoprecipitation experiments and found that the co-immunoprecipitates contained the SSU component of the RPS6 but very little amount of RPL7 (Fig. 4f). This finding agrees with specific 18S rRNA crosslinks detected by PAR-CLIP and HDLBP binding to an 18S rRNA sequence as shown in Fig. 3h. Taken together, these results showed that HDLBP interacts with the translational apparatus including the SRP and is localized in the proximity of ribosome-associated factors.

HDLBP promotes translation at the ER membrane

Based on previous findings^{45,46,47,48,49} and the specific interaction of HDLBP with the CDS of membrane-bound mRNAs and the ribosome observed in our study, we next addressed the function of HDLBP in ER-associated translation. We thus measured the process of active translation in the presence and absence of HDLBP by generating two CRISPR/Cas9 HDLBP knockout (KO) cell lines (Fig. 5a). The HEK293 HDLBP knockout cells showed no apparent growth defect and electron microscopy imaging of the ER revealed no morphology changes (data not shown). To quantify translation efficiency in the KO and wild-type (WT) conditions we generated ribosome profiling datasets, which showed high read periodicity and dominant in-frame P-site coverage in the CDS (Supplementary Fig. 5a, b). Since this dataset was obtained from non-fractionated cells, we asked if our experiment sufficiently captured the ribosome footprints on ER-bound mRNAs. As expected we observed low footprint density for SP and TM-containing mRNAs in the region downstream of the start codon, until the emergence of both targeting signals from the ribosome (Supplementary Fig. 5c), confirming that the data recovered ER-bound ribosome complexes with near-nucleotide resolution.

a Schematic overview of the ribosome profiling experiment in HEK293 parental and HDLBP KO cells. Western analysis shows the absence of HDLBP in KO cells. b Membrane-bound HDLBP target mRNAs were split into three similarly sized groups based on their cross-linking signal in the CDS (indicated in parentheses). Differences in translation efficiency (HDLBP KO vs. WT) were compared between groups. A two-sided Wilcoxon rank-sum test was used to test for significance. c pSILAC analysis of newly synthesized proteins in HEK293 parental and HDLBP KO cells. SILAC heavy vs. medium ratios (H/M) reflect changes in protein synthesis upon HDLBP KO and were quantified in membrane fractions. Proteins were split into three similarly sized groups based on their cross-linking signal in the CDS (indicated in parentheses). SILAC ratios were compared between groups using a two-sided Wilcoxon rank-sum test. d Proteins were split into three similarly sized groups based on their PAR-CLIP signal and membrane localization (indicated in parentheses). SILAC ratios were compared between groups using a two-sided Wilcoxon rank-sum test. eParental and HDLBP KO cells were transfected with secreted Gaussia luciferase (Gluc) construct. Gluc activity was quantified in the medium and normalized to the intracellular Firefly luciferase (Fluc) activity. Each of the four replicate experiments (r1–r4) was carried out with five technical replicates. Data were presented as mean values ± SD. f Parental and HDLBP KO cells were transfected secreted alkaline phosphatase (SEAP) construct. SEAP signal was quantified in medium and normalized to the intracellular Firefly luciferase (Fluc) activity. Each experiment was carried out in five technical replicates. Data were presented as mean values ± SD. g Western analysis of A549 cells stably transfected with construct overexpressing HDLBP (OE). h SEAP activity, relative to Fluc, was measured in HEK293 and A549 cells with different HDLBP protein levels (KO knockout, WT wild-type, HDLBP/OE overexpression). The experiment was carried out five times with at least five technical replicates. Data were presented as mean values ± SD. i Ribosome P-site coverage around targeting signals (signal peptides and transmembrane helices) was compared between HEK293 HDLBP knockout (KO) vs. parental (WT) cells. P-site coverage was scaled to the coverage in codons 20–40 of each mRNA. A rolling mean of 5 nt was used to smooth the profiles. Absolute numbers of analyzed mRNAs are given, median signal peptide and first transmembrane helix lengths are indicated with a vertical dotted line. a–i Source data are provided as a Source Data file.

Full size image

To address the differences in translation efficiency upon HDLBP KO we compared groups of membrane-bound mRNAs (Supplementary Data 5), which were classified according to the number of HDLBP crosslinks in the CDS. The highest decrease in translation efficiency was observed for mRNAs that had the largest number of HDLBP crosslinks in the CDS, suggesting that HDLBP interactions with membrane-bound mRNAs promoted translation (Fig. 5b) but did not affect ER mRNA localization (Supplementary Fig. 5d). We validated these findings by quantifying protein synthesis using pulsed stable isotope labeling with amino acids in cell culture (pSILAC)⁵⁰ in combination with subcellular fractionation (Supplementary Data 6). The absence of HDLBP generally resulted in a decrease in protein synthesis of proteins encoded by membrane-bound mRNAs (Fig. 5c) and the extent of decrease depended on the level of HDLBP cross-linking signal (Fig. 5d and Supplementary Fig. 5e). Therefore, HDLBP is required for efficient protein synthesis of its target mRNAs. Since the ER membrane is the primary site of translation of secretory proteins, we next asked if HDLBP influences their secretion. For this purpose, we expressed the secreted Gaussia luciferase (Gluc) and alkaline phosphatase (SEAP) in HEK293 cells in WT and KO conditions and quantified enzyme activity in the culture medium. Gluc and SEAP activity was significantly decreased upon HDLBP KO by 20–40%, showing that HDLBP depletion reduces the secretion of the two reporter proteins (Fig. 5e, f and Supplementary Fig. 5f).

Since the depletion of HDLBP reduced secretion, we next tested the impact of HDLBP overexpression. To this end we transiently transfected FLAG/HA-HDBLP into HEK293 and HEK293 HDLBP KO cells. We also stably overexpressed HDLBP in A549 cells using a piggybac transposon carrying HDLBP (Fig. 5g).

To test the effect on secretion, we transfected the SEAP reporter construct into HEK293 and A549 cells with differing HDLBP expression levels (Fig. 5h). As expected, KO of HDLBP in HEK293 and A549 cells reduced the secretion of SEAP, whereas overexpression of HDLBP in HDLBP knockout cells rescued SEAP secretion to the levels observed in wild-type cells. Interestingly, the re-introduction of HDLBP into HEK293 and A459 cells resulted in increased secretion of about 1.15- and 2-fold, respectively (Fig. 5h) confirming that HDLBP expression levels directly influence the extent of SEAP secretion.

We next asked if the absence of HDLBP affects ribosome occupancy on membrane-bound mRNAs. For this purpose, we displayed the footprint coverage around known targeting signals (Fig. 5i and Supplementary Fig. 5g) and found that the absence of HDLBP resulted in lower ribosome density immediately downstream of the region encoding SPs or the first TM domain. This suggests that HDLBP contributes to ribosome elongation arrest, which is required for co-translational targeting and efficient translocation of the nascent peptide.

HDLBP crosslinks to CU/UU-containing codons in mRNAs and tRNAs decoding these codons

We next interrogated our ribosome profiling datasets to quantify ribosome occupancy per codon in WT and KO conditions. While we found that upon HDLBP KO several codons (including UUC/Phe, UUU/Phe, CUC/Leu, and CUU/Leu) were on average slightly more occupied by the ribosome and the P-site and E-site, this increase was not statistically significant. Therefore, global ribosome stalling in the KO condition may be measurable but could be an indirect effect of HDLBP depletion. Analysis of normalized HDLBP PAR-CLIP signal in codons (Supplementary Fig. 6a), identified the same codons (UUC/Phe, CUC/Leu, and CUU/Leu) to be among the most highly bound.

In addition, we analyzed HDLBP binding to tRNAs by PAR-CLIP enrichment as well T-C transition specificity for each tRNA cross-linking position (Supplementary Data 7) and observed that four leucine isotype tRNAs were among the top 15 enriched tRNAs in the HDLBP-bound pool (Fig. 6b). Their cross-linking sites were located in variable and D loops (Fig. 6c and Supplementary Fig. 6b) and corresponded to HDLBP mRNA binding motifs (UCUUC). Interestingly, the codons decoded by these tRNAs tend to be more occupied in HDLBP-depleted cells (Fig. 6a), suggesting that HDLBP tRNA binding enables more efficient tRNA decoding likely by facilitating tRNA recycling or reduction of tRNA diffusion from the ribosome, as proposed previously³⁴.

a Difference in codon frequencies in the P-site (top) and E-site codons (bottom) in HDLBP KO vs. WT. Mean codon shift was calculated for four replicates (mean ± standard deviation is shown). bEnrichment of tRNAs in HDLBP PAR-CLIP and their binding sites. T-C transitions in tRNAs were normalized to total tRNA abundance and ranked from highest to the lowest value (left to right). For each T-C transition, we displayed its transition specificity (T-C transition vs. total read coverage). Mean values of two PAR-CLIP biological replicates ± SD are depicted. Total log2-transformed tRNA abundance and codon usage are also shown (top). c (Left) Browser representation of alignment to tRNA Leu-UAG. T-C transitions in the D-loop and V-region are indicated for the HDLBP PAR-CLIP dataset. The second track shows coverage in the total RNA sample. (Right) HDLBP crosslinked uridines are indicated with respect to secondary tRNA structure.

Full size image

Absence of HDLBP results in lower proliferation and tumor formation capacity

HDLBP function in ER translation and secretion could impact the production of mitogens, growth factors, receptors and extracellular matrix, and consequently greatly influence cell proliferation, differentiation, migration, and invasion. To test this, we established a CRISPR/Cas9-induced HDLBP KO in the lung adenocarcinoma (LUAD) derived cell line A549. In 2D cultured A549 cells, adhesive growth was markedly (by 40–50%) reduced by HDLBP deletion (Fig. 7a). Likewise, HDLBP KO interfered with the 2D migration of A549 cells, as indicated by severely reduced wound closure in scratch migration analyses (Fig. 7b). The surface expression of the transmembrane glycoprotein CD71 encoded by the HDLBP-bound TFRC mRNA (Fig. 3h) was reduced upon HDLBP KO (Fig. 7c and Supplementary Fig. 7a, b). In contrast, overexpression of HDLBP in A549 cells resulted in accelerated wound closure (Fig. 7d). Collectively, these findings suggested that HDLBP substantially influences the oncogenic properties of tumor cells. To test this in vivo, parental and HDLBP KO A549 cells were stably transduced with iRFP (infrared fluorescent protein) and viable cells were subcutaneously (s.c.) injected into athymic (FOXN1^nu/nu) nude mice to monitor tumor initiation and growth (Fig. 7e). Homogenous s.c. application of both cell populations was validated by non-invasive near-infrared imaging of iRFP immediately post-injection (Supplementary Fig. 7c, 0 days post-injection). Strikingly, tumor initiation and growth were severely reduced by HDLBP deletion (Fig. 7e and Supplementary Fig. 7c). In contrast to parental cells, which formed tumors in all eight analyzed animals, HDLBP KO cells formed palpable tumors in only three of eight animals (Supplementary Fig. 7c). In accordance, tumor volume and final tumor mass were significantly reduced by HDLBP deletion (Fig. 7f), indicating that HDLBP is essentially involved in tumor initiation and growth.

a Relative confluence, viability, and DNA content of A549 parental or HDLBP KO cells. Representative confluence masks are shown (left panel). Unpaired t-test was used to test for significance (***p < 0.001). b Wound healing of WT or HDLBP KO cells. The wound closure was monitored for 20 h upon damage (right panel). The wound density was calculated 20 h post damage (middle panel). Representative confluence masks are shown (left panel). Unpaired t-test (left) and two-way ANOVA (right) were used to test for significance (both ***p < 0.001). c CD71 surface expression of WT or HDLBP KO cells. The mean fluorescent intensity (MFI) of CD71 was determined by flow cytometry experiments. d Wound healing of A549 cells stably transfected with doxycycline-inducible HDLBP overexpression (see Fig. 5g). The wound density was calculated 20 h post damage (right panel). Representative confluence masks are shown (left panel). Unpaired t-test was used to test for significance (***p < 0.001). e Schematic of mouse injection experiment. WT or HDLBO KO A549 cells were injected into the right and left flanks of athymic nude mice. Tumor formation was followed in eight mice over 21 days post-injection. f Tumor volume (in mm³) was quantified 7, 14, and 21 days post-injection (left panel). Box plots of log2-transformed tumor volume formed from KO or WT cells are shown. An unpaired two-sided t-test was used for comparisons. The number of palpable tumors per group is given in parentheses. Tumor weight was quantified 21 days post-injection (right panel). Box plots of tumor weight (in mg) formed from KO or WT cells are shown. An unpaired two-sided t-test was used for comparisons. In case of the absence of a tumor, weight was plotted as 0. The lower and upper hinges of box plots correspond to the 25th and 75th percentiles, respectively. Upper and lower whiskers extend from the hinge to the largest or smallest value no further than the 1.5× interquartile range from the hinge, respectively. Center lines of box plots depict the median values. g Gene Ontology enrichment analysis of down and upregulated mRNAs in tumors from HDLBP KO cells. Adjusted p values for the top six enriched categories are shown. h Differences in fold change (HDLBP KO vs. WT tumors) were compared between membrane-bound and cytosolic mRNAs, as determined in the HEK293 RNA-seq fractionation experiment. A two-sided Wilcoxon rank-sum test was used to test for significance. a–d Data were presented as mean values ± SD. a–h Source data are provided as a Source Data file.

Full size image

To determine how HDLBP KO influenced gene expression, RNA abundance was monitored by RNA sequencing in final tumors (Supplementary Fig. 7d). In the absence of HDLBP, a high number of mRNAs was found to be up (n = 1039) and downregulated (n = 700), suggesting that HDLBP has a great impact on gene expression, presumably due to substantially impaired cell signaling and tumor-stroma cross-talk (Supplementary Fig. 7e and Supplementary Data 8). Strikingly, the protein products of mRNAs decreased by HDLBP KO were highly enriched for secreted proteins (e.g., collagens and matrix metalloproteases, MMP2) that function in biological processes such as “extracellular matrix organization”, “cell adhesion”, and “tissue development” (Fig. 7g). These findings supported the view, that HDLBP influences protein output, expression, and/or turnover of transcripts encoding factors essentially involved in modulating the composition of and cross-talk within the tumor-stroma landscape. To evaluate the direct control of deregulated mRNAs by HDLBP, we compared the fold changes between WT and KO xenograft tumors for subgroups of mRNAs classified according to their PAR-CLIP signal in HEK293 cells. We found that changes in mRNA expression of HDLBP target mRNAs (Supplementary Fig. 7f) and membrane-bound mRNAs (Fig. 7h) were significantly decreased in xenografts lacking HDLBP than for non-targets or cytosolic mRNAs. Although it remains to be addressed if HDLBP influences mRNA turnover by modulating ER-associated translation, these findings suggest that next to indirect regulation of gene expression, HDLBP also directly influences mRNA abundance, potentially by modulating ER-associated mRNA translation. In sum, the presented findings provide strong evidence that HDLBP is an important modulator of tumor progression influencing the expression of secreted factors, receptors, and extracellular matrix components involved in modulating tumor initiation and progression.

In this work, we characterized the function of HDLBP in the context of translation at the ER. While ~80% of membrane-bound mRNAs were found to be high-affinity substrates of HDLBP (Fig. 1e), they also showed predominant binding in the CDS (Figs. 1c, 2b). In contrast, only ~40% of all cytosolic mRNAs contained HDLBP binding sites but showed significantly lower affinity and were more randomly distributed between CDS and 3′ UTRs (Fig. 2b). Investigation of the primary sequences of ER-targeted and cytosolic mRNAs revealed that the HDLBP-bound CU-containing motifs were more frequent in membrane-bound mRNAs (Fig. 3a), which also showed high codon frequency for hydrophobic amino acids, such as Leu (CUC and CUU), Ile (AUC), Phe (UUC), and Val (GUC), commonly present in signal peptides and transmembrane helices⁵¹.

While the previously reported CHHC/CHYC binding motif³⁵ matched our findings, we further found evidence that the frequency of longer HDLBP-bound motifs up to 12 nt in length is higher than for shorter motifs in ER-bound than in cytosolic mRNAs (Fig. 3c, d). Therefore, HDLBP may have evolved to specifically recognize membrane-bound mRNAs by making use of their differential sequence composition. Multiple KH domains may allow HDLBP to recognize ER-bound mRNA through its interaction with long heterogeneous RREs and/or with multipartite motifs resulting in multivalent high-affinity binding regions, as observed for other RBPs^28,52,53,54. We thus quantified highly multivalent HDLBP sites and found that they correlated with high binding affinity, were ~40 nt long and contained on average 3–4 UC/CU-containing four-mers positioned several nucleotides apart. This may reflect the binding of 3–4 functional domain modules within the HDLBP protein structure to multiple RREs. Using in vitro binding assays (Fig. 3i), we confirmed these observations and conclude that multivalent interactions indeed result in the higher affinity of HDLBP to long RNA-binding regions. Further studies will provide insight into which domains or their combinations are responsible for the recognition of long multivalent sites and the formation of HDLBP-containing mRNPs. The recently identified C/U-repeat-containing SECReTE motif¹⁷ supports the existence of long functional sequences in yeast mRNA that are required for the secretion of encoded proteins and very likely represents the RNA-binding sites of the HDLBP yeast orthologue Scp160p.

In addition, we report that HDLBP is interacting with ER-associated mRNAs and promoting their translation resulting in increased secretion of the synthesized proteins (Fig. 5), which is in line with previous findings^34,35. Since we found high-affinity HDLBP binding sites not only in ER-targeted mRNAs but also in membrane-localized mRNAs with no known or predicted targeting signals (Fig. 1d), HDLBP may significantly contribute to the efficient translation of both SRP-dependent and -independent mRNAs^8,9,10,11,12. Profiling of SRP-dependent and -independent mRNAs in mammalian cells will contribute to this understanding in future studies.

Furthermore, we show that HDLBP absence results in decreased ribosome elongation arrest around targeting signals (Fig. 5g). This process is required for efficient targeting and translocation of nascent peptides to the ER lumen, as well as to prevent misfolding, aggregation, and ER-associated degradation⁵⁵. HDLBP-mediated local ribosome slowdown may thus promote targeting, translocation, and accurate folding of protein domains encoded by HDLBP-bound regions. This possibility is strongly supported by the detected h16 and ES6SB 40S HDLBP interactions, which are in proximity to the EEF2 and EIF4G2 binding sites⁴¹ and could influence ribosome elongation arrest. In addition, low-affinity HDLBP contacts detected in the 7SL RNA region, which is required for the delay in GTP hydrolysis during elongation arrest, and membrane targeting activity⁵⁶ also support this notion. Finally, we identified key chaperones and chaperonins in close proximity to HDLBP, supporting its role in protein folding, as previously suggested for its yeast orthologue³⁶.

Mechanistically, we propose that HDLBP interacts with the translational apparatus by binding to the mRNA and the small ribosomal subunit as well as to other ribosome-associated factors. High-affinity mRNA binding contributes to the formation of elongation-arrested cytosolic RNCs, allowing their efficient localization to the ER membrane, translocon handover, and/or enabling proper folding of the nascent peptide. Since HDLBP is most likely bound to the mRNA downstream from the elongating ribosome, it could be sequestered from the mRNA via ribosome collisions possibly via tRNA and/or ribosome ES6SB-dependent mechanism, which remains to be elucidated. We speculate that HDLBP is bound to the mRNA only during the primary round of translation after which it is removed. This step may allow the mRNA molecule to localize to the ER, where additional rounds of translation can be carried out in the presence of ER folding machinery and aggregation quality control. In the future, the sequence of HDLBP binding events to mRNA, rRNA, and tRNA during different stages of translation should be addressed.

In the absence of HDLBP, we also detected increased ribosome occupancy at P- and E-sites. Since this increase was modest and not statistically significant, it may be a consequence of the indirect effects of HDLBP KO. Nevertheless, several studies suggest that E-site may act as a sensor for ribosome elongation kinetics⁵⁷ and that E-site occupancy may directly influence translation fidelity and ribosome translocation^58,59, which could be another possible mechanism of how HDLBP promotes translation elongation.

Overall, our results point to a general function for HDLBP in ER translation. Although this is obviously important for every cell, we expect that this phenotype is most prominent in specialized secretory cell types (e.g., fibroblasts, pancreatic and immune cells). In support of this view, HDLBP mRNA expression, and thus likely HDLBP protein level is substantially elevated in secretory cells (Supplementary Fig. 7g), most prominently in fibroblasts, which are key to orchestrating the extracellular landscape by producing and remodeling the extracellular matrix. In addition, HDLBP misregulation gives rise to far-reaching consequences and modulation of disease phenotypes, such as impaired viral replication²⁴, atherosclerotic plaque formation³⁵, and autism⁶⁰. Finally, the results presented in this study highlight the striking involvement of HDLBP in lung tumor cells during tumor progression and suggest that therapeutic interventions targeting HDLBP may represent a previously unrecognized strategy for inhibiting lung tumor growth or other malignancies. In the future, further implications of the regulatory role of HDLBP for tumor biology need to be explored.