N terminal wobble base usage determines ribosome loading and thus protein expression fate

N terminal wobble base usage determines ribosome loading and thus protein expression fate | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article N terminal wobble base usage determines ribosome loading and thus protein expression fate Huimin Yu, Ruizhao Jiang, Yukun Zheng, Yuwen Wei, Zhirao Zhang, and 8 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8581843/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted You are reading this latest preprint version Abstract Heterologous expression of proteins and enzymes in prokaryotic hosts such as Escherichia coli and Rhodococcus ruber is central to biomanufacturing and biotechnology, yet a substantial fraction of proteins still fail to express for reasons that remain unknown. To address this long-standing problem, we focused on translation initiation, particularly the N-terminal coding-start region. By constructing large-scale 5′-synonymous codon libraries, we found that mRNA secondary structure alone cannot account for expression-fate of proteins across hosts. Instead, a previously unrecognized factor, designated as N-terminal-specified Incompatible Codons (NICs), were identified that decisively correlate protein-non-expression. Library mining revealed all of the NICs in E. coli , spanning 13 of the 20 amino acids, such as GTC for Val, AAG for Lys, ACC for Thr and CAG for Gln. Further studies showed that these NICs are independent of codon rarity, tRNA abundance, and mRNA folding, but correlated to the wobble-base usages. Ribosome loading experiments demonstrated that NICs act as kinetic barriers that prevent ribosome loading thereby resulting in failed translation from initiation and early elongation to efficient elongation. We further performed deep DIA-based LC–MS/MS analysis on intracellular soluble proteins and quantified protein abundance, obtaining a quantitative snapshot of the endogenous proteome. The top 500 most abundant proteins in E. coli were selected and the N-terminal optimal codons (NOC) were summarized. By integrating both NIC avoidance and NOC preference, we designed a new two-section codon-usage strategy for heterologous protein overexpression in prokaryotes, including a one-to-one (one-amino-acid, one-codon) codon table for 48 bp N-terminus and a mixed table (optimal codon utilization + rare codon substitution) for the subsequent sequence. Using this novel approach, 8 previously non-expressed proteins such as lipase, laccase and cysteine hydrolase, were reversely overexpressed in E. coli . An on-line codon-design tool RiboLoad Codon Optimizer , has been available http://47.86.169.8/ , serving for numerous proteins/enzymes overexpression with E. coli and R. ruber hosts. Together, these findings establish a mechanism-grounded codon-usage framework for overcoming translational bottlenecks and enabling efficient heterologous protein expression in prokaryotic hosts. Biological sciences/Molecular biology/Translation Biological sciences/Biotechnology/Expression systems Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Introduction Heterologous protein expression is foundational to modern biomedicine, diagnostics and industrial biocatalysis: it enables therapeutic proteins and vaccines, affords high-specificity analytical reagents, and powers enzyme-driven manufacturing at scale. Prokaryotic hosts such as Escherichia coli remain the workhorses for recombinant production owing to their fast growth, low cost, genetic tractability, and straightforward scale-up [ 1 , 2 ]. In recent years, the importance of heterologous expression has been further amplified by the rapid rise of sequence-designed proteins, driven by advances in protein engineering, directed evolution, and AI-assisted sequence generation. These approaches are dramatically accelerating the design–build–test cycle for enzymes and functional proteins, enabling rapid creation of variants tailored for applications in biomanufacturing, synthetic biology, environmental remediation, and healthcare[ 3 – 5 ]. However, the functional realization of these designed sequences critically depends on their successful expression in heterologous hosts. Yet, the frequent and often unexplained failure or extremely low yield of target proteins remains a fundamental and pervasive bottleneck that has long challenged the field[ 6 – 8 ]. A study by the Northeast Structural Genomics (NESG) platform indicates that 28% of heterologous proteins could not be expressed at all in Escherichia coli [ 9 ]. To address this challenge, numerous strategies have been developed, including vector and host optimization, promoter engineering, ribosome binding site design, and the use of fusion tags [ 10 – 13 ], which we have also been reviewed in detail recently[ 14 ]. Nevertheless, a substantial proportion of proteins—particularly those derived from eukaryotes or extremophiles—still fail to reach detectable expression levels in E. coli and other prokaryotic hosts, highlighting a profound gap in our current understanding of sequence-dependent control of gene expression. When improvements in vectors or hosts fail, the obstacle often resides in the gene sequence itself. Classical sequence-level strategies—most prominently, “rare-codon optimization” across the open reading frame—have enabled high expression for many targets in E. coli [ 10 , 15 – 17 ]. However, a non-trivial subset of heterologous proteins still fail to express for reasons that are not captured by traditional rarity-focused or genome-wide codon-usage heuristics[ 9 , 18 ]. These approaches are largely agnostic to the fine-scale organization of codons at the 5’ end, where translation is initiated and early elongation is established. In bacteria, translation initiation is the dominant kinetic checkpoint that sets protein output; features concentrated in the first ~ 30–50 nt of the coding sequence strongly modulate ribosome loading, start-site accessibility, and the handoff into early elongation [ 19 , 20 ]. Consistent with this, multiple studies have demonstrated that sequence features within the 5’ coding region exert a decisive influence on expression efficiency, and that rational optimization of this region can markedly alter the expression levels of target proteins[ 9 , 21 , 22 ]. The sequence features of the translation initiation region are widely recognized as a key determinant of protein synthesis efficiency, with codon choice at the 5’ end being particularly critical. Accumulating evidence suggests that the 5’-terminal region of gene may act as a rate-limiting element during translation initiation. Previous studies have shown that the first 18 bp of the open reading frame correspond to the ribosome-protected fragment of the 70S complex [ 23 ], and that the sequence composition of this region directly affects the efficiency of translation initiation[ 9 , 22 ]. In addition, an extended downstream window of approximately 30 bp has also been proposed to serve as a critical translation regulatory segment [ 24 ]. Nevertheless, the generalizable rules governing this region remain poorly defined, and the precise contribution of individual codons to translation initiation and early elongation has yet to be systematically elucidated. Recent work has further emphasized 5’-end mRNA structure as a putative core determinant of this region: stable hairpins formed immediately downstream of the start codon are thought to impede ribosome binding and scanning, thereby suppressing translation initiation[ 25 – 31 ]. For instance, Goodman et al. demonstrated in a large-scale study of sfGFP variants that reducing 5’-end mRNA secondary structure markedly increased protein yield[ 26 ]. Is true that it’s the 5’-end mRNA-secondary structure determines the expression fate of diverse proteins? Is there any other unknown factor that plays a critical role? Here, we systematically interrogate the 5′ coding region using large-scale synonymous libraries and uncover a previously unrecognized, codon-level determinant of translation initiation. We show that specific N-terminal codons impair ribosome loading and dictate the expression outcome of heterologous proteins, enabling a new, mechanism-grounded framework for rational gene design in prokaryotes. Results N-terminus mRNA structure is not the dominant factor for protein expression As mentioned above, poor expression of heterologous proteins has been largely attributed to the highly-structured mRNA at the N-terminus. Indeed, several studies have reported a significant correlation between mRNA folding energy at the N-terminus and protein expression level[25]. To test the accuracy of this conclusion, we constructed a synonymous mutation library encompassing the first 48 bp of the red fluorescent protein (mCherry) gene, thereby generating an N-terminal synonymous library (>10⁷ variants). We randomly chose 320 members for complete sequencing and fluorescence measurement to directly link sequence diversity (generating different mRNA secondary structures) to expression outcomes in terms of fluorescence. Using this dataset, we calculated the mRNA folding energy of each variant and analyzed its correlation with fluorescence intensity. As shown in Fig. 1A, 20 representative 48 bp N-terminal synonymous mutant DNAs generated 20 diverse structures of mRNA (first 48 nt coding region) with different folding energy scores, expressing mCherry with varied fluorescence intensities. Correlating the results of mRNA folding Minimum Free Energy (MFE) with fluorescence intensities, we summarized how mRNA structures affected the mCherry expression in E.coli and R. ruber host, respectively, as shown in Fig. 1B and 1C. In E. coli , mCherry expression levels were negatively correlated with mRNA structural stability (r=0.31, p<10 -4 ), seemingly consistent with the “mRNA structural hindrance” model. Strikingly, however, when the same mCherry-library was expressed in another prokaryotic chassis, R. ruber , a high-GC-content host, an entirely opposite trend emerged: expression levels were positively correlated with mRNA stability (r=-0.2007, p<10 -3 ); that is, more stable transcripts yielded higher protein expression. This finding fundamentally contradicted the conventional mRNA structure-inhibition theory, indicating that translation initiation efficiency might not be explained solely by mRNA-secondary-structure effects. Taken together, our experimental results revealed a host-dependent reversal that cannot be reconciled by a single “structure-is-bad” or “structure-is-good” rule. Identification of N-terminal-specified Incompatible Codons (NICs) causing non-expression of proteins Now that the mRNA-secondary structure is not the dominant factor governing the expression of proteins, what is the hidden factors behind? To address this, we introduced the N-terminal-48bp- synonymous mutation library for 10 amino acids in the previous section (mCherry 16 aa including Met, Val, Ser, Lys, Gly, Glu, Asp, Asn, Ala and Ile) and bridged the specific codon-fluorescence correlation. As shown in Fig. 2A, we surprisingly identified the existence of some specific codons for amino acids, which significantly reduced the expression of mCherry. For instance, at the second position (Valine), the four synonymous codons (GTT, GTC, GTA, GTG) resulted in significantly different protein expression levels. A post-hoc test revealed that mutants with the GTC codon exhibited markedly lower expression than those with the other three synonymous codons. To validate this finding, we analyzed the codon distribution within the lowest 5% expressing population. Of the 14 mutants in this group, 9 utilized the GTC codon at this position—a frequency far exceeding random expectation. Similar phenomena were also exhibited for Lys, Glu, and Asn (Fig. S2). We further selected 4 other functionally diverse model proteins: green fluorescent protein (GFP), nattokinase (NK), serine hydroxymethyl transferase (Bsu) from Bacillus subtilis , and thaumatin (Tha) from Thaumatococcus danielli with 16 aa termius as shown in Fig. 2B, respectively, to observe the N-terminus-codon effects. To enable the high-throughput screening and analyses, the mCherry reporter was fused to the C-terminus of NK, Bsu and thaumatin via a flexible (G₄S)₃ linker (Fig. 2A). Sequencing and characterization of over 4,000 clones yielded 1,525 validated mutants in total, establishing a large-scale 5′-end variant dataset that directly connects the codon-usage with fluorescence across these five systems (Fig. 2A, 2B-2K, and Fig. S3-S7). For each 16 aa-fragment, we correlated the codon usage with fluorescence, and through hypothesis testing, identified the low-expression codons (defined by p < 0.01) for each amino acid. For example, AAG for Lys in mCherry and GFP, ACC for Thr in GFP and Tha, CAG for Gln in Bsu, and GTG for Val in the NK library. Further, these low-expression-correlated codons exhibit positional specificity. For instance, in a low-expression variant containing 12 AAG codons at different positions, we found that only substituting the three N-terminus AAGs into AAA raised fluorescence from 5,741 to 18,756. In contrast, substituting 9 downstream AAGs while retaining the two N-terminus AAGs, the fluorescence didn’t improve (5,979). That is, the inhibitory effect of AAG is confined to the 5’ initiation region and does not extend into subsequent elongation. To further confirm the dominant role of the N-terminus, we performed domain-swapping experiments. Synonymous libraries were constructed by replacing 48 bp windows at different positions (codons 1-16, 17-32, and 118-132) in both high-expression and low-expression sequences. The outcome was consistent. The N-terminus without low-expression codons ensured high expression regardless of downstream codon composition. Conversely, another N-terminus containing 6 low-expression-correlated codons resulted in persistently low protein-expression levels (Fig.2L). Another unexpected finding is that these low-expression-correlated codons in E.coli are not rare ones. The aforementioned GTC codon, for example, is a common non-rare codon with a genomic usage frequency of 21%. Moreover, archetypal high-frequency codons, such as ACC for Threonine (Thr) (22.8%), were also identified as low-expression-leading codons. Conversely, some classic rare codons, such as ATA for Isoleucine (Ile) (5.5%) and GGA for Glycine (Gly) (12%), did not exert a significant negative effect on expression. This clearly demonstrates that the inhibitory effect of these specific codons is independent of conventional genomic codon usage frequency(Fig. 2N). Therefore, we finally defined these special low-expression-correlated codons as N-terminal-specified Incompatible Codons (NICs), which existed in 13 out of 20 natural amino acids. These NICs display two defining properties: (i) they are N-terminus-specific, exerting strong effects only near the start of the coding sequence; and (ii) they maybe are not rare codons in the conventional sense. Again, we further collected the poor-expressed mutants and analyzed the 48 bp N-terminus, it’s found that all of the poor-expressed targets contained 5-11 low-expression codons (Fig. 2M). Across the five libraries (1,525 validated mutants), the bottom 2% low-expression sequences contained a total of 582 NIC occurrences within the 1,200 N-terminal positions of the 75 low-expression variants. Validation of the NIC’s negative effects on protein expression We next conducted systematic experimental validation for the negative effects of NICs. We specifically selected two of the weakest variants, mCherry-A and mCherry-B, each carrying 11 NICs. Through individually replacing each NIC with a synonymous non-NIC, we observed that the fluorescence intensity increased by approximately 1.9～3 folds upon each single NIC repair. Substitution of all NICs resulted in more than a six-fold fluorescence increase (from 5,574 to 34,239 for mCherry-A, and from 5,741 to 34,051 for mCherry-B) (Fig.3A). Extending this strategy to the variant library with the lowest fluorescence intensity (mCherry-C-H), we consistently observed a substantial recovery in fluorescence, increasing from the bottom 5% to the top 20% (Fig. 3B). These results demonstrated that the cumulative accumulation of NICs is the major cause of protein-non-expression, and their in-total replacement can completely reverse the failure expression. Calculation of the 5′-end mRNA folding free energy revealed a more nuanced relationship between local sequence changes, mRNA structure, and expression output. Specifically, compared with the low-expression parental sequences, substitution of the second-position Val codon from GTC to GTA led to a pronounced decrease in the absolute folding free energy of the first 48 nt coding region of the mRNA, indicative of reduced structural stability (A-1 vs. A; B-1 vs. B). This structural relaxation was accompanied by an approximately two-fold increase in fluorescence intensity. In contrast, when NICs at other downstream positions were substituted, the folding free energy showed no significant reduction; notably, in the B-3 variant—where the fourth-position Lys codon AAG was replaced with AAA—the folding free energy was even significantly increased relative to the parental B sequence. Despite this apparent increase in mRNA structural stability, fluorescence intensity was nevertheless markedly enhanced. (Fig. 3A). Moreover, neither classical predictions using the Zuker algorithm nor advanced models based on deep learning attributed the decisive effect of NICs to structural changes (Fig.S8 and Fig. S9). This means that the NIC effect has no significant relationship with the mRNA secondary structure. Further studies showed that the inhibitory effects of part of NICs exhibit a degree of sequence context dependency. For example, for Val, GTC acted as the strongest unfit codon to mCherry, whereas GTG and GTT for Val displayed a stronger inhibitory effect in the Bsu and GFP libraries. This suggests that the impact of some NICs showed protein-specific features. Accordingly, codons that consistently exhibited strong inhibitory effects across at least one N-terminal context were operationally classified as NICs, and finally we obtained a NIC table, as shown in Fig. 3C, in which all of the NICs are highlighted with red background boxes and white text. The NIC Effect is Not Caused by tRNA Scarcity Following the identification of NICs, a central question naturally arose: what underlies the NIC effect? In the classical rate-limiting model of translation, protein non-expression or low expression is typically attributed to an insufficient supply of specific tRNAs. To test whether this traditional mechanism accounts for the effect of NICs, we systematically measured intracellular tRNA abundance in E. coli and compared the data with the identified NICs. Unlike early approaches that inferred tRNA availability from genomic copy numbers to calculate the tRNA adaptation index (tAI), we used high-throughput sequencing to directly quantify intracellular tRNA levels, thereby avoiding discrepancies between copy number and actual supply (Fig. 4A)[32, 33]. We first examined the relationship between codons and the abundance of their strictly matched cognate tRNAs. The analysis showed no direct association between whether a codon is a NIC and the intracellular concentration of its cognate tRNA. Several findings contradicted the tRNA- insufficient supply hypothesis. For example, aspartate codon GAC was identified as a strong NIC in multiple libraries, yet its cognate tRNA-GTC is present at relatively high abundance. In contrast, its synonymous non-NIC codon GAT lacks a cognate tRNA in the E. coli genome and is entirely decoded by wobble pairing through tRNA-GTC. Similarly, the phenylalanine codon TTC, classified as a NIC, corresponds to a tRNA present at much higher abundance than that of its synonymous non-NIC codon TTT. These counterintuitive examples directly challenge the view that insufficient tRNA supply is the primary cause of NIC effect (Fig.4B). Because wobble pairing is pervasive in bacterial translation, analyses based solely on cognate tRNA abundance may fail to capture the full decoding landscape. We therefore performed a second-level analysis in which the abundances of all tRNAs capable of decoding a given codon—including those engaging in wobble interactions—were summed to estimate its total decoding capacity. This expanded analysis again failed to support the tRNA-abundance hypothesis. The inhibitory effect of NICs showed no correlation with either total decoding capacity or the mode of decoding, i.e., whether a codon is read exclusively through wobble pairing or by a cognate tRNA. For example, the glutamate codon GAG, classified as a NIC, is decoded entirely via wobble by tRNA-TTC, whereas the aspartate codon GAC, also a NIC, is decoded by its cognate tRNA. Despite these fundamentally different decoding modes, both codons exhibited comparable inhibitory effects on expression. Conversely, strong counter-evidence was observed for valine: the two NIC codons GTT and GTC displayed significantly higher total decoding tRNA abundance than the synonymous non-NIC codon GTA (Fig. 4C). Taken together, analyses of both cognate tRNA abundance and total decoding capacity consistently indicate that the inhibitory effect of NICs does not arise from limited aminoacyl-tRNA supply. This conclusion rules out the most common rate-limiting mechanism in translation and strongly suggests that the NIC effect originates from deeper molecular events during translation. NIC effect is caused by blocking of the translation initiation and early elongation A new reasonable deduction for the molecular basis of NIC is the altered dynamics of the translational machinery itself. To investigate this mechanism, we turned to ribosome-level analysis and performed polysome profiling, which separates mRNAs according to the number of bound ribosomes and provides a snapshot of their translational state. As model systems, we again employed mCherry-A (harboring 11 NICs and exhibiting a fluorescence intensity of only 5,574) and mCherry-A7 (with these NICs removed, showing a markedly higher fluorescence of 34,239) (Fig. 5A). Consistent with our previous observations, the overall polysome profiles of the two strains showed striking differences at both the early fermentation stage (3 h) and the mid-log phase (6 h), as shown in Fig. 5D and Fig. 5F, respectively. In mCherry-A7, a pronounced enrichment in heavy polysome fractions was detected at both time points, indicating that a large proportion of mRNAs underwent efficient and concurrent translation by multiple ribosomes. In contrast, mCherry-A displayed a clear shift toward lighter fractions, suggesting that most of its transcripts were stalled at the initiation stage or remained bound by only a single ribosome. To quantify the translational dynamics of the target mRNA, we measured the distribution of mCherry transcripts across gradient fractions by qPCR. At 3 h, the mCherry-A strain showed that 46.0% (3.95% free + 42.1% initiation) of transcripts accumulated before or at early initiation, with another 33.8% trapped in early elongation and only 20.0% reaching stable polysomes. In contrast, mCherry-A7 exhibited a dramatically different pattern, with 68.8% of mRNAs residing in polysome fractions, and only 27.4% (8.52% free + 18.91% initiation) stalled prior to productive elongation (Fig. 5E). A similar but even more pronounced divergence was observed at 6 h. In mCherry-A, the majority of transcripts (69.9% = 21.43% free + 48.49% initiation) remained restricted to pre-initiation or early-initiation states, and merely 11.5% entered stable elongation. Meanwhile, in mCherry-A7, over 81.9% of transcripts accumulated in polysomes, with minimal retention in pre-initiation (14.4%) or early elongation (3.6%) fractions (Fig. 5G). These results provide direct molecular evidence for the mechanism of NIC repression. Consistent with this notion, preliminary docking analyses suggested that high-expression N-terminal sequences may exhibit improved compatibility with 16S rRNA at the mRNA 5′ end [34] (Fig. S11).The presence of NICs creates barriers at initiation or early elongation, sharply reducing ribosome entry and processive scanning along the coding sequence. As a consequence, many transcripts remain in low-efficiency states, leading to reduced protein output. This mechanism explains the translational bottleneck caused by NICs and integrates with the findings from the preceding analyses. To further validate that NICs repress translation by directly impairing ribosome recruitment and initiation complex stability, we next quantified the binding affinity between E. coli ribosomes and mRNAs using surface plasmon resonance (SPR, Biacore). This approach enabled direct measurement of ribosome–mRNA interactions at the molecular level, independent of downstream elongation effects. Four representative RNA constructs were designed to disentangle the contributions of the Shine–Dalgarno (SD) sequence and the early coding region: (i) SD sequence fused to the first 18 bp of mCherry-A, (ii) SD sequence fused to the first 18 bp of mCherry-A7 (with NICs replaced), (iii) the first 48 nt of mCherry-A without an SD sequence, and (iv) the first 48 nt of mCherry-A7 without an SD sequence. These constructs allowed us to specifically probe whether NICs affect ribosome binding in an SD-dependent or SD-independent manner. Strikingly, the SD–mCherry-A (18 bp) RNA exhibited markedly weaker ribosome binding, with an equilibrium dissociation constant (KD) of 3.55 × 10⁻⁹ M. This affinity was comparable to that observed for both SD-lacking constructs—mCherry-A (48 bp) and mCherry-A7(48 bp)—indicating inefficient ribosome engagement when NICs were present in the immediate post-start region. In sharp contrast, the SD–mCherry-A7(18 bp) RNA displayed substantially stronger ribosome association, with a KD of 9.88 × 10⁻¹¹ M, representing over a 30-fold increase in binding affinity relative to its NIC-containing counterpart(Fig. 5H and 5I). These results demonstrate that NICs profoundly weaken ribosome–mRNA binding even in the presence of an intact SD sequence, whereas removal of NICs restores high-affinity ribosome association. Importantly, the similar affinities observed for SD-free RNAs indicate that the enhanced binding of mCherry-A-7 arises specifically from cooperative interactions between the ribosome and the early coding region rather than from nonspecific RNA–ribosome contacts. Together with the polysome profiling data, these findings provide convergent molecular evidence that NICs impose a translational bottleneck by destabilizing ribosome loading and early initiation complexes. By reducing ribosome entry efficiency and preventing productive transition into elongation, NICs trap mRNAs in low-occupancy states, thereby sharply limiting protein output. This mechanism offers a direct physical explanation for NIC-mediated translational repression and integrates seamlessly with our ribosome-level and expression-level analyses. AT/GC3-driven host differences in NICs To further assess the universality and host dependence of the NIC phenomenon, we applied the same “library construction–NIC screening” workflow to Rhodococcus ruber . Remarkably, NICs were also readily detectable in this high-GC Gram-positive actinobacterium, and individual inhibitory codons exerted strong suppressive effects on recombinant protein expression. These findings demonstrate that the NIC mechanism is not an idiosyncrasy of E. coli but instead represents a broadly conserved layer of sequence-encoded translational control across phylogenetically distant prokaryotes. Despite this universality, the inhibitory landscapes of NICs diverged sharply between hosts at the level of wobble base usage. In E. coli , the Val codon GTC was among the strongest inhibitors of mCherry expression, whereas in R. ruber all four Val codons behaved equivalently and showed no detectable repression. The same pattern was observed for Asn, where E. coli favored AAT, while R. ruber exhibited a clear preference for AAC (Fig. S10). These host-specific patterns were also evident at the global level of NIC composition. Among all 20 NICs identified in E. coli , 17 were GC3-ending codons, whereas in R. ruber , 8 of the 9 NICs discovered in our current screening set were AT3-ending. This GC3-to-AT3 polarity switch represents a fundamental rewiring of N-terminal wobble base usage between the two hosts, rather than a simple shift in codon frequency. This polarity shift in third-position preference provides an explanatory framework for the opposite “mRNA structure–expression” correlations observed in the two systems. In E. coli , the dominant GC3 endings tend to stabilize local RNA structures, which correlates with the classical “more stable → lower expression” pattern. In contrast, in R. ruber , inhibitory elements are predominantly AT3-type codons, which usually destabilize local structures; as a result, a “less stable → lower expression” trend emerges—mirroring previous observations in actinobacterial expression systems[25, 26, 35]. Together, these findings reveal that the inhibitory effects of NICs cannot be attributed to RNA secondary structure alone. Instead, they reflect host-specific molecular rules governing translation initiation and early elongation, which are encoded at the level of third-position nucleotide identity. We therefore propose that NICs constitute a universal yet evolutionarily diversified translational control module, whose functional divergence arises from distinct wobble-position decoding constraints that shape the early ribosomal engagement landscape, ultimately determining the expression fate of heterologous proteins. Mining N-terminal optimal codons and selecting downstream-favored codons from naturally-overexpressed proteins Building on our prior identification of NICs and their inhibitory effects on ribosome loading, we hypothesized that naturally overexpressed proteins should exhibit strong selective pressure against such inhibitory codons, particularly within the translation initiation region. To test this hypothesis, we performed deep DIA-based LC–MS/MS analysis on intracellular soluble proteins and quantified protein abundance using intensity-based absolute quantification (iBAQ). This approach provided a comprehensive and quantitative snapshot of the endogenous proteome across a wide dynamic range. Based on iBAQ intensities, proteins were ranked by abundance, and the top 1,500 most abundant proteins were selected as a high-confidence pool representing the highly expressed endogenous proteome. To relate protein-level abundance to underlying sequence features, the identified protein groups were mapped to their corresponding genes in the reference genome of the host strain. For each of the top 1,500 abundant proteins, the matched coding sequence (CDS) was retrieved, thereby establishing a direct link between experimentally measured protein abundance and its underlying gene sequence. From this high-abundance pool, we further focused our analysis on the top 500 most abundant proteins in order to minimize potential noise from mid-abundance species and to concentrate on the most robustly expressed endogenous genes. Codon-usage analysis was first focused on the N-terminal region (the first 16 codons, 48 bp), where NICs exert their strongest effects on translation initiation. As expected, many NICs, such as GGG (Gly) and TCG (Ser), were strongly depleted at the 5′ ends of these highly expressed genes. Notably, however, NIC exclusion was not absolute. A subset of NICs still appeared at measurable frequencies, suggesting that their inhibitory effects can be partially tolerated depending on local sequence context or protein-specific constraints (Fig. S14). To further quantify this selective pressure, we compared codon usage at the N terminus with that across full-length CDSs. Several NICs, including AAC (Asn) and ACC (Thr), remained among the most frequently used synonymous codons at the N terminus but were nonetheless significantly underrepresented relative to their genome-wide frequencies (e.g., AAC: 57% at the N terminus vs. 72% genome-wide). These codons therefore appear to be selectively constrained—but not strictly excluded—within the N-terminal region (Fig. 6D). In contrast, a distinct set of codons exhibited pronounced enrichment specifically at the N terminus. For example, ACA (Thr) increased from 6.6% in full-length CDSs to 18.3% at the N terminus, while CAA (Gln) increased from 25.1% to 49.5%. These observations indicate that the N-terminal coding region of highly expressed genes is shaped not only by avoidance of inhibitory codons but also by positive selection for host-preferred “N-terminal-favored” codons that likely facilitate efficient ribosome engagement and early elongation. Having characterized codon selection at the 5′ end, we next examined whether codon usage within the downstream coding region also correlates with expression level. Analysis of full-length CDSs across different abundance tiers (top 500, top 1,500, and genome-wide background) revealed that most codons displayed highly consistent usage across all groups. However, four codons—AAC (Asn), CAC (His), TAC (Tyr), and TTC (Phe)—showed a clear monotonic enrichment with increasing protein abundance, suggesting preferential usage during efficient elongation of naturally overexpressed proteins (Fig. S13). Together, these analyses reveal a compartmentalized codon-selection landscape in highly expressed genes: the N-terminal region is subject to strong, region-specific selective pressures, while the downstream region is largely governed by global codon bias with limited enrichment of a small subset of elongation-favored codons. Codon-usage-new-criteria for overexpression of heterologous proteins in E. coli Although NICs were generally depleted within the N-terminal 48 bp of highly expressed proteins, a more striking pattern emerged when we further focused on the extreme N-terminal core region immediately following initiation. Specifically, within the first 18 bp of the most abundant proteins (top 100), certain codons exhibited extremely low occurrence frequencies (below 0.05), indicating a pronounced level of depletion at an exceptionally short sequence scale. This observation suggests the existence of “core-region incompatible codons” that exert a disproportionate inhibitory effect on translation during the earliest elongation steps (Fig. S15). Integrating this finding with our systematic analyses of NICs and N-terminal codon preferences, we propose a new codon-usage criterion for efficient heterologous protein overexpression in E. coli , based on a two-section design strategy (Fig. 7A). Section 1 encompasses the N-terminal 48 bp region, which governs ribosome loading and early elongation efficiency. Codon selection within this region should strictly follow an N-terminal Optimal Codon (NOC) scheme, in which each amino acid is encoded by a single preferred codon. These optimal codons are defined by excluding both NICs and core-region incompatible codons, followed by selection of the synonymous codons most strongly enriched or dominant at the N termini of highly expressed host proteins. Section 2 corresponds to the downstream coding region beyond the first 48 bp. In this region, codon design can follow a more conventional strategy, preferentially employing codons enriched in highly expressed proteins (e.g., AAC, CAC, TAC, and TTC) while avoiding known rare codons, thereby maintaining efficient elongation without introducing unnecessary ribosomal pausing. An on-line codon-design tool RiboLoad Codon Optimizer , has been available http://47.86.169.8/. To validate the efficacy and universality of this new criteria, we systematically tested a diverse panel of refractory proteins from various origins that are notoriously difficult to express in E. coli . These included archaeal molecular chaperones (Pfdb, Tms14AD, and Tms14ED), human death effector domain proteins and suppressors of cytokine signaling, Pleurotus ostreatus laccase, Yarrowia lipolytica lipase, and Bacillus subtilis cysteine hydrolase (Fig. 7B-7I). Across all tested proteins, the codon design strategy proposed in this study produced robust and consistent improvements in expression. For Tms14AD, Tms14ED, the suppressor of cytokine signaling, laccase, and cysteine hydrolase, neither the native sequences nor those optimized using mainstream commercial codon-optimization tools yielded detectable protein bands on SDS–PAGE. In contrast, application of our new codon-usage criterion resulted in a marked increase in expression levels. Notably, Tms14AD and Tms14ED were expressed in soluble form, whereas the remaining proteins accumulated at high levels as inclusion bodies. Moreover, for Pfdb and the lipase—two proteins that exhibited modest but limited improvements following commercial codon optimization—our approach achieved substantially greater enhancement, with expression levels increasing by more than an order of magnitude relative to commercial optimization. Importantly, for laccase, lipase, and cysteine hydrolase, replacement of only the core incompatible codons within the N-terminal 18 bp region was sufficient to dramatically elevate expression, underscoring the disproportionate regulatory impact of this minimal sequence window (Fig.S16). Collectively, these results demonstrate that the proposed codon-usage criterion not only rescues expression systems that completely fail under conventional optimization strategies, but also delivers order-of-magnitude gains even on top of existing optimizations, highlighting its broad applicability and mechanistic significance Discussion Efficient expression of heterologous proteins remains a central and still largely unsolved challenge in modern biotechnology. Despite decades of optimization, a striking number of heterologous proteins—across hosts and systems—remain non-expressing or express at undetectable levels, with no clear mechanistic explanation. Current optimization strategies are largely built on two classical hypotheses. The first is the “rare codon” model, which attributes low expression to delays caused by limited supply of rare tRNAs. The second is the “mRNA secondary structure” model, which posits that G/C-rich sequences in the 5’ region form stable hairpins that block ribosome binding and initiation. Based on these principles, researchers have developed codon optimization tools and codon usage tables such as CAI and tAI, which are widely applied in practice. However, real-world performance is far more complex. In the same host, E. coli , some model proteins can be expressed at high levels, whereas others fail to express at all. This inconsistency highlights the limitations of existing codon tables and optimization rules, which cannot fully explain or resolve the “non-expression” problem. What, then, is the true cause of this discrepancy? Does an unrecognized rule exist that determines whether a protein can be translated successfully? To address this core scientific question, we constructed systematic 5’ synonymous mutation libraries and uncovered a previously underappreciated class of regulatory elements—NICs. These codons exert strong inhibitory effects in the translation initiation region and basically determine whether a target protein can be expressed. By sequencing and quantifying more than 4,000 independent clones, we established 1,525 precise genotype–phenotype relationships. Unlike FACS-seq and other mixed-library approaches, this strategy eliminated noise from bulk sequencing and allowed single-codon contributions to be resolved[ 9 , 22 ]. Several lines of evidence indicate that the NIC effect is fundamentally distinct from traditional codon-based models. First, NICs do not correlate with genomic codon frequency: multiple common codons, such as ACC for Thr and GTC for Val, acted as strong inhibitors when positioned at the 5′ end, whereas canonical rare codons, including ATA for Ile, often showed no detectable negative effect. Second, direct quantification of intracellular tRNA abundances demonstrated that NIC-mediated repression cannot be explained by limited aminoacyl-tRNA supply. In several cases, NICs exhibited equal or even higher total decoding capacity than their synonymous high-expression counterparts, contradicting the classical “tRNA scarcity” hypothesis and suggesting that decoding availability per se is not the limiting factor. Although mRNA secondary structure is widely recognized as a global determinant of translation efficiency, consistent with prior large-scale studies[ 25 , 26 ], our rescue experiments revealed that replacing only a small number of NICs at the N terminus could increase protein expression by more than sevenfold, with minimal changes in local folding free energy. This decoupling between expression recovery and mRNA structural stability indicates that NICs do not act primarily by occluding ribosome access through stable RNA structures. Instead, their inhibitory effect appears to arise downstream of ribosome binding, at a functional stage of translation initiation or early elongation that is not captured by static structure-based metrics. Cross-species validation further revealed the mechanistic depth of the NIC phenomenon. In Rhodococcus ruber , a high-GC actinobacterium, the NIC repertoire differed almost completely from that of E. coli . Several strong NICs in R. ruber , including Ser-AGT, Asp-GAT, and Asn-AAT, were A/T-ending codons, in contrast to the GC3-dominated NICs observed in E. coli . This inversion provides a coherent explanation for the opposite correlations between 5′-end mRNA structural stability and expression observed in the two hosts, but cannot be rationalized by RNA folding alone. We propose that these host-specific NIC profiles reflect long-term co-evolution between each organism’s translational machinery—including ribosomal RNA, ribosomal proteins, and initiation and elongation factors—and its genomic codon usage patterns. In a given host, certain codons may impose intrinsic kinetic or conformational barriers during the earliest decoding events, thereby reducing the probability that ribosomes successfully transition from initiation into processive elongation. Because translation initiation is the dominant rate-limiting step in bacterial protein synthesis, even modest increases in abortive initiation or early dissociation events can be strongly amplified, creating a decisive bottleneck that downstream sequence optimization cannot overcome. Polysome profiling provided direct molecular evidence for this model, showing that mRNAs containing NICs accumulate predominantly in initiation or monosome-associated fractions and fail to engage efficiently in polysome elongation, whereas NIC-repaired transcripts readily enter productive translation. At a deeper mechanistic level, our observations are also consistent with previous studies showing that the bacterial 30S subunit recognizes mRNA primarily through 16S rRNA and the ribosomal protein S1, both of which play central roles in mRNA recruitment, positioning, and early decoding. Protein S1, in particular, has been implicated in preferential binding to single-stranded, A/U-rich regions and in facilitating the loading of structured or weakly accessible mRNAs onto the ribosome. Importantly, both the sequence composition of 16S rRNA and the abundance, structure, and binding preferences of S1 vary substantially across bacterial species. In this context, the host-specific inversion of N-terminal GC/AT preferences observed here may reflect intrinsic differences in how distinct ribosomes engage with nascent coding sequences immediately downstream of the start codon. Rather than acting solely through global mRNA folding energetics, early synonymous codons may modulate ribosome recruitment, stabilization, or conformational transitions during the formation of a productive initiation complex in a host-dependent manner. This perspective provides a plausible molecular basis for why the same 5′-end sequence features can exert opposite effects on expression in different organisms[ 34 , 36 , 37 ]. Building on these mechanistic insights, we propose a two-section design strategy in which the N-terminal 48 bp is optimized using an N-terminal Optimal Codon (NOC) chart that excludes both NICs and core-region incompatible codons, while the downstream region is designed more conservatively using codons enriched in highly expressed host proteins. Validation across a diverse panel of refractory proteins demonstrated that this criterion consistently rescued previously non-expressing targets and achieved order-of-magnitude improvements over conventional codon-optimization approaches, in several cases requiring replacement of only the core incompatible codons within the first 18 bp. Together, these results highlight the disproportionate regulatory impact of a minimal N-terminal sequence window and establish a mechanism-grounded codon-usage rule for heterologous protein overexpression. Despite these advances, the precise molecular mechanism of NICs remains to be resolved. Although polysome profiling located the incompatible at initiation and early elongation, the specific codon–ribosome interactions underlying this barrier remain a “black box.” Future studies using in vitro translation systems and cryo-EM could capture transient ribosome–mRNA structures at atomic resolution and reveal the conformational states that cause stalling[ 38 ]. Beyond mechanistic resolution, the broader significance of the NIC phenomenon lies in its implications for the rapidly expanding field of sequence-designed proteins. Advances in rational protein engineering, directed evolution, and AI-assisted sequence generation are dramatically accelerating the creation of novel enzymes and functional proteins. However, the practical realization of these designs remains fundamentally constrained by heterologous expression failure, which often arises independently of protein folding or catalytic potential. Our findings suggest that for a substantial fraction of designed sequences, expression fate is already determined within a minimal N-terminal coding window, long before downstream sequence features can exert any effect. In this context, NICs provide a mechanistic explanation for why many otherwise promising designs fail at the first experimental hurdle. Rather than being an incidental nuisance, non-expression emerges as a predictable, sequence-encoded outcome governed by host-specific decoding constraints at the earliest stages of translation. Recognizing and eliminating NICs therefore shifts heterologous expression from an empirical, trial-and-error process toward a rule-based and mechanistically grounded design problem. Looking forward, extending NIC screening to a broader range of prokaryotic and eukaryotic hosts—including yeast and mammalian systems—will test the generality of this principle and may uncover additional layers of host-dependent translational regulation. More broadly, our work reframes heterologous protein expression as a problem of early translational decision-making, in which a small number of synonymous codons at the N terminus exert a disproportionate regulatory influence over expression fate. By identifying and mechanistically contextualizing these incompatible codons, this study establishes a new conceptual and practical framework for bridging the gap between sequence design and functional protein production—an essential step for the next generation of enzyme engineering, synthetic biology, and AI-driven protein discovery. Conclusion In this study, we systematically identified and validated a previously unrecognized class of translational regulatory elements—N-terminal-specific Incompatible Codons (NICs)—that decisively determine the expression fate of heterologous proteins in prokaryotic hosts. Through large-scale synonymous library screening, ribosome-level analyses, and cross-host validation, we demonstrate that the inhibitory effects of NICs are not explained by codon rarity, tRNA supply, or mRNA secondary structure. Instead, NICs act by destabilizing ribosome loading and the transition from translation initiation and early elongation to efficient translation, thereby creating a stringent kinetic barrier that prevents productive translation. By integrating NIC analysis with proteome-guided codon-usage mining, we further distilled these mechanistic insights into a new codon-usage criterion for heterologous protein overexpression. This criterion emphasizes the functional compartmentalization of the translation initiation region, identifying a minimal core window within the first 18 bp that is under extreme negative selection in highly expressed genes, and defining an expanded N-terminal design region of 48 bp governed by host-specific optimal codon usage. Rational remodeling of this core region at single-codon resolution was sufficient to rescue multiple previously non-expressing proteins and to achieve order-of-magnitude improvements over conventional codon-optimization strategies. Collectively, this work reveals a previously hidden layer of sequence-encoded translational control that operates upstream of classical elongation-based models and establishes a mechanism-grounded design dogma for gene optimization in prokaryotes. By shifting the focus from global codon bias to decisive N-terminal bottlenecks, our findings provide both conceptual insight into translation regulation and a broadly applicable framework for overcoming long-standing barriers in heterologous protein production. Declarations Acknowledgement This work was supported by the National Key R&D Program of China (2025YFA0921800), National Natural Science Foundation of China (No. U24A6011), Beijing Municipal Science & Technology Commission No. Z231100003723015 and 202504841063. Author contributions Ruizhao Jiang: Writing–original draft, Methodology, Investigation, Data curation, Conceptualization. Yukun Zheng : Methodology, Data curation, Conceptualization. Yuwen Wei: Methodology, Data curation, Conceptualization. Mingye Song: Figure illustration, Data curation. Shunshun Yang: Figure illustration, Data curation. Ke Wang: Methodology, Data curation. Yilong Zhou : Methodology, Data curation, Validation. Yan Du: Methodology, Data curation. Fulong Li: Methodology. Youxiang Liang: Methodology, Data curation. Miaomiao Wang: Methodology, Data curation. Huimin Yu: Writing, review & editing, Supervision, Funding acquisition, Conceptualization. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. References De Marco, A., Recent advances in recombinant production of soluble proteins in E. coli. Microbial cell factories, 2025. 24 (1): p. 21. France, S.P., R.D. Lewis, and C.A. Martinez, The evolving nature of biocatalysis in pharmaceutical research and development. JACS Au, 2023. 3 (3): p. 715-735. Kim, D., et al., Computational design of metallohydrolases. Nature, 2025: p. 1-8. Huang, P.-S., S.E. Boyken, and D. Baker, The coming of age of de novo protein design. Nature, 2016. 537 (7620): p. 320-327. Chen, Y., et al., Glycolysis-compatible urethanases for polyurethane recycling. Science, 2025. 390 (6772): p. 503-509. Lozano Terol, G., et al., Impact of the expression system on recombinant protein production in Escherichia coli BL21. Frontiers in microbiology, 2021. 12 : p. 682001. Kondo, T. and S. Yumura, Strategies for enhancing gene expression in Escherichia coli. Applied Microbiology and Biotechnology, 2020. 104 (9): p. 3825-3834. Price, W.N., et al., Large-scale experimental studies show unexpected amino acid effects on protein expression and solubility in vivo in E. coli. Microbial Informatics and Experimentation, 2011. 1 : p. 1-20. Boël, G., et al., Codon influence on protein expression in E. coli correlates with mRNA levels. Nature, 2016. 529 (7586): p. 358-363. Schmidt, M., et al., Maximizing Heterologous Expression of Engineered Type I Polyketide Synthases: Investigating Codon Optimization Strategies. ACS Synthetic Biology, 2023. Fusco, F., et al., Influence of the mRNA initial region on protein production: a case study using recombinant detoxified pneumolysin as a model. Frontiers in Bioengineering and Biotechnology, 2024. 11 : p. 1304965. Tolia, N.H. and L. Joshua-Tor, Strategies for protein coexpression in Escherichia coli. Nature methods, 2006. 3 (1): p. 55-64. Ojima-Kato, T., S. Nagai, and H. Nakano, N-terminal SKIK peptide tag markedly improves expression of difficult-to-express proteins in Escherichia coli and Saccharomyces cerevisiae. Journal of bioscience and bioengineering, 2017. 123 (5): p. 540-546. Jiang, R., et al., Strategies to overcome the challenges of low or no expression of heterologous proteins in Escherichia coli. Biotechnology Advances, 2024. 75 : p. 108417. Gustafsson, C., S. Govindarajan, and J. Minshull, Codon bias and heterologous protein expression. Trends in biotechnology, 2004. 22 (7): p. 346-353. Ikemura, T., Codon usage and tRNA content in unicellular and multicellular organisms. Molecular biology and evolution, 1985. 2 (1): p. 13-34. Quax, T.E., et al., Codon bias as a means to fine-tune gene expression. Molecular cell, 2015. 59 (2): p. 149-161. Bhandari, B.K., et al., Analysis of 11,430 recombinant protein production experiments reveals that protein yield is tunable by synonymous codon changes of translation initiation sites. PLoS computational biology, 2021. 17 (10): p. e1009461. Gualerzi, C.O. and C.L. Pon, Initiation of mRNA translation in bacteria: structural and dynamic aspects. Cellular and Molecular Life Sciences, 2015. 72 (22): p. 4341-4367. McCarthy, J.E. and C. Gualerzi, Translational control of prokaryotic gene expression. Trends in Genetics, 1990. 6 : p. 78-85. Kim, D.J., et al., DeepTESR: a deep learning framework to predict the degree of translational elongation short ramp for gene expression control. ACS Synthetic Biology, 2022. 11 (5): p. 1719-1726. Nieuwkoop, T., et al., Revealing determinants of translation efficiency via whole-gene codon randomization and machine learning. Nucleic acids research, 2023. 51 (5): p. 2363-2376. Duval, M., et al., Escherichia coli ribosomal protein S1 unfolds structured mRNAs onto the ribosome for active translation initiation. PLoS biology, 2013. 11 (12): p. e1001731. Tuller, T. and H. Zur, Multiple roles of the coding sequence 5′ end in gene expression regulation. Nucleic acids research, 2015. 43 (1): p. 13-28. Kudla, G., et al., Coding-sequence determinants of gene expression in Escherichia coli. science, 2009. 324 (5924): p. 255-258. Goodman, D.B., G.M. Church, and S. Kosuri, Causes and effects of N-terminal codon bias in bacterial genes. Science, 2013. 342 (6157): p. 475-479. Griswold, K.E., et al., Effects of codon usage versus putative 5′-mRNA structure on the expression of Fusarium solani cutinase in the Escherichia coli cytoplasm. Protein expression and purification, 2003. 27 (1): p. 134-142. Eyre-Walker, A. and M. Bulmer, Reduced synonymous substitution rate at the start of enterobacterial genes. Nucleic acids research, 1993. 21 (19): p. 4599-4603. Tuller, T., et al., An evolutionarily conserved mechanism for controlling the efficiency of protein translation. Cell, 2010. 141 (2): p. 344-354. Shah, P., et al., Rate-limiting steps in yeast protein translation. Cell, 2013. 153 (7): p. 1589-1601. Bentele, K., et al., Efficient translation initiation dictates codon usage at gene start. Molecular systems biology, 2013. 9 (1): p. 675. Zheng, G., et al., Efficient and quantitative high-throughput tRNA sequencing. Nature methods, 2015. 12 (9): p. 835-837. Carbone, A., A. Zinovyev, and F. Képes, Codon adaptation index as a measure of dominating codon bias. Bioinformatics, 2003. 19 (16): p. 2005-2015. Gonzalez de Valdivia, E.I. and L.A. Isaksson, A codon window in mRNA downstream of the initiation codon where NGG codons give strongly reduced gene expression in Escherichia coli. Nucleic acids research, 2004. 32 (17): p. 5198-5205. Castillo-Méndez, M., et al., Adenine-containing codons enhance protein synthesis by promoting mRNA binding to ribosomal 30S subunits provided that specific tRNAs are not exhausted. Biochimie, 2012. 94 (3): p. 662-672. Etchegaray, J.-P. and M. Inouye, Translational enhancement by an element downstream of the initiation codon in Escherichia coli. Journal of Biological Chemistry, 1999. 274 (15): p. 10079-10085. Vimberg, V., et al., Translation initiation region sequence preferences in Escherichia coli. BMC molecular biology, 2007. 8 (1): p. 100. Wang, C., et al., Structural basis of transcription-translation coupling. Science, 2020. 369 (6509): p. 1359-1365. Additional Declarations There is NO Competing Interest. Supplementary Files SI.docx SUPPLEMENTARY INFORMATION rs.pdf Reporting Summary Cite Share Download PDF Status: Under Review Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8581843","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":604630432,"identity":"d5eac5ba-c3ae-41c2-9bdf-acd69d537596","order_by":0,"name":"Huimin Yu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABBUlEQVRIiWNgGAWjYFACHgaGD0CSD8ROACOIIF4tjDOAJBsDMwlamEEKwFoYiNFicCP36GbbHDsZNgb+wx8e7qjNMzh+gPHB2zYGeXOcWvLSbuduSwY5jE0i8czxYoMzCcyGc9sYDHc24NKSYwbUwgzWwpDYdixxw4EENmneNoYEgwN4tFhuqwdpYf4A1nL+AftvgloYtx0Gh5hEYltN4oYbCWzM+LRInnljdrN323EeNmZmM6CWA4kzbzxslpxzTsJwAw4tfMdzzG783FZtz8/e+Pjjz7a6xL7zyQc/vCmzkcdliwJcHBIth4GYsQFISGBXDwTyDaj8OpwqR8EoGAWjYOQCABX3W6dyXDm9AAAAAElFTkSuQmCC","orcid":"","institution":"Tsinghua University","correspondingAuthor":true,"prefix":"","firstName":"Huimin","middleName":"","lastName":"Yu","suffix":""},{"id":604630433,"identity":"bb7e95a1-b2b8-45f2-8f06-b12828782fa4","order_by":1,"name":"Ruizhao Jiang","email":"","orcid":"","institution":"Tsinghua University","correspondingAuthor":false,"prefix":"","firstName":"Ruizhao","middleName":"","lastName":"Jiang","suffix":""},{"id":604630434,"identity":"fb39e98f-e584-4edf-9eeb-282d12d31110","order_by":2,"name":"Yukun Zheng","email":"","orcid":"https://orcid.org/0000-0003-0096-0979","institution":"Tsinghua University","correspondingAuthor":false,"prefix":"","firstName":"Yukun","middleName":"","lastName":"Zheng","suffix":""},{"id":604630435,"identity":"89d02b79-2713-4888-84cf-097ca035532a","order_by":3,"name":"Yuwen Wei","email":"","orcid":"","institution":"Tsinghua University","correspondingAuthor":false,"prefix":"","firstName":"Yuwen","middleName":"","lastName":"Wei","suffix":""},{"id":604630436,"identity":"bfc24402-a6c1-425a-908e-31697533d350","order_by":4,"name":"Zhirao Zhang","email":"","orcid":"","institution":"Tsinghua University","correspondingAuthor":false,"prefix":"","firstName":"Zhirao","middleName":"","lastName":"Zhang","suffix":""},{"id":604630437,"identity":"206ebc6f-f4a8-45f8-83e5-06d3b939729c","order_by":5,"name":"Mingye Song","email":"","orcid":"https://orcid.org/0009-0003-6399-4321","institution":"Tsinghua University","correspondingAuthor":false,"prefix":"","firstName":"Mingye","middleName":"","lastName":"Song","suffix":""},{"id":604630438,"identity":"5c06b7fb-5a09-46f0-bcd9-52774d50b8d6","order_by":6,"name":"Shunshun Yang","email":"","orcid":"","institution":"Tsinghua University","correspondingAuthor":false,"prefix":"","firstName":"Shunshun","middleName":"","lastName":"Yang","suffix":""},{"id":604630439,"identity":"dc9b64ee-f597-40ce-bedb-f53d405888ad","order_by":7,"name":"Ke Wang","email":"","orcid":"","institution":"Tsinghua University","correspondingAuthor":false,"prefix":"","firstName":"Ke","middleName":"","lastName":"Wang","suffix":""},{"id":604630440,"identity":"d291eec1-7de5-41b6-8b32-aa73a2d5344a","order_by":8,"name":"Yan Du","email":"","orcid":"","institution":"Tsinghua University","correspondingAuthor":false,"prefix":"","firstName":"Yan","middleName":"","lastName":"Du","suffix":""},{"id":604630441,"identity":"246dcb04-e73c-4dd3-9389-439b2e84eee7","order_by":9,"name":"Yilong Zhou","email":"","orcid":"","institution":"Tsinghua University","correspondingAuthor":false,"prefix":"","firstName":"Yilong","middleName":"","lastName":"Zhou","suffix":""},{"id":604630442,"identity":"854f8357-8b1c-4c09-93a4-441d94428357","order_by":10,"name":"Fulong Li","email":"","orcid":"","institution":"Beijing Evolyzer Co.,Ltd","correspondingAuthor":false,"prefix":"","firstName":"Fulong","middleName":"","lastName":"Li","suffix":""},{"id":604630443,"identity":"47766c68-520f-4e4b-8a3f-c876a270a937","order_by":11,"name":"Youxiang Liang","email":"","orcid":"","institution":"Beijing Evolyzer Co.,Ltd","correspondingAuthor":false,"prefix":"","firstName":"Youxiang","middleName":"","lastName":"Liang","suffix":""},{"id":604630444,"identity":"99d704ee-f985-4b72-a389-ccb8b9151185","order_by":12,"name":"Miaomiao Wang","email":"","orcid":"","institution":"Beijing Evolyzer Co.,Ltd","correspondingAuthor":false,"prefix":"","firstName":"Miaomiao","middleName":"","lastName":"Wang","suffix":""}],"badges":[],"createdAt":"2026-01-12 12:46:22","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8581843/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8581843/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":105706090,"identity":"a1b3ffe7-4eb7-4801-a231-bee835bd4b39","added_by":"auto","created_at":"2026-03-30 07:05:20","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":584940,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSynonymous mutation library of mCherry and correlation between mRNA structures and expression level across hosts.\u0026nbsp;\u003c/strong\u003eA. Overview of mCherry 5′-end library design and representative high- and low-expression variants and their structures across hosts. In \u003cem\u003eE. coli\u003c/em\u003e, highly expressed variants generally exhibit low mRNA structural stability within the first 48 nt coding region, whereas in \u003cem\u003eR. ruber\u003c/em\u003e, high-expression sequences are typically associated with strongly structured 5′- mRNA secondary structures with low Minimum Free Energies (MFEs), The MFEs were computed using the ViennaRNA package (http://rna.tbi.univie.ac.at). B. Correlation analysis between 5′-end mRNA folding free energy (MFE) and fluorescence intensity. A positive correlation was observed (r = 0.3139, p \u0026lt; 10⁻⁴), indicating that weaker 5′-end mRNA structures are associated with higher expression. C. Correlation analysis between 5′-end MFE and fluorescence intensity. A negative correlation was observed (r = −0.2007, p \u0026lt; 10⁻³), indicating that weaker 5′-end mRNA structures are associated with lower expression.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8581843/v1/bc8503a4e97bea2638f3803b.png"},{"id":105728994,"identity":"6233e6df-50f6-4eb7-b6ec-21a97a92d9c8","added_by":"auto","created_at":"2026-03-30 11:13:12","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":178947,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eConstruction of diverse model-protein libraries and identification of NICs.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(A) \u003c/strong\u003eLibrary designs of additional model proteins covering all 20 amino acids and codon-grouped fluorescence analysis of the mCherry library at position 2 (Val). The GTC codon showed significantly lower expression than the other Val codons (one-way ANOVA, p \u0026lt; 0.05).\u003cstrong\u003e (B)\u003c/strong\u003e Fluorescence images of the GFP 5′-end library. \u003cstrong\u003e(C)\u003c/strong\u003e Position-4 Lys analysis in the GFP library, where AAG exhibited significantly lower expression than AAA (t-test, p \u0026lt; 0.01).\u003cstrong\u003e (D)\u003c/strong\u003e Fluorescence images of the Tha–mCherry library.\u003cstrong\u003e (E\u003c/strong\u003e Position-4 Thr analysis in the Tha–mCherry library, where ACC showed significantly lower expression than the other Thr codons (one-way ANOVA, p \u0026lt; 0.01).\u003cstrong\u003e (F)\u003c/strong\u003e Position-5 Phe analysis in the Tha–mCherry library, where TTC showed significantly lower expression than TTT (t-test, p \u0026lt; 0.01).\u003cstrong\u003e (G)\u003c/strong\u003e Fluorescence images of the Bsu–mCherry library.\u003cstrong\u003e (H)\u003c/strong\u003e Position-11 Gln analysis in the Bsu–mCherry library, where CAG exhibited significantly lower expression than CAA (t-test, p \u0026lt; 0.01).\u003cstrong\u003e (I)\u003c/strong\u003e Position-15 Ala analysis in the Bsu–mCherry library, where GCC showed significantly lower expression than the remaining Ala codons (one-way ANOVA, p \u0026lt; 0.01). \u003cstrong\u003e(J)\u003c/strong\u003e Fluorescence images of the NK–mCherry library.\u003cstrong\u003e (K)\u003c/strong\u003e Position-2 Ala analysis in the NK–mCherry library, where GCG showed significantly lower expression than the other Ala codons (one-way ANOVA, p \u0026lt; 0.01).\u003cstrong\u003e (L)\u003c/strong\u003e Expression comparison of synonymous mutation libraries constructed at downstream regions (positions 17–31, 118–132, and 222–236) using high-expression versus low-expression 5′-end scaffolds. For all three regions, synonymous variants built on high-expression N-terminal scaffolds showed significantly higher fluorescence than those built on low-expression scaffolds (t-test, p \u0026lt; 0.01). \u003cstrong\u003e(M)\u003c/strong\u003e Association between NIC count and expression level in the mCherry library. Variants containing more NICs showed progressively lower fluorescence. \u003cstrong\u003e(N) \u003c/strong\u003eGenomic frequency analysis of identified NICs. NICs discovered in the libraries are not rare codons at the genome level.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8581843/v1/b44fd7f7d6d774d8b81f7ae0.png"},{"id":105729303,"identity":"19852e38-7b57-46a7-a5d2-56f370359db2","added_by":"auto","created_at":"2026-03-30 11:14:20","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":171830,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eInhibitory effects of NICs on expression and the complete NIC catalog.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(A)\u003c/strong\u003e Fluorescence and MFE changes of two low-expression variants (mCherry-A and mCherry-B) after replacing their NICs. Despite minimal changes in 5′-end folding stability, NIC repair resulted in substantial increases in fluorescence intensity. \u003cstrong\u003e(B)\u003c/strong\u003e Expression recovery of additional low-fluorescence variants after replacing all NICs within the first 16 codons. mCherry C-H represents the remaining sequences within the lowest 5% expression group of the library. Removal of NICs consistently elevated fluorescence across different sequence backgrounds. Notably, for the low-expression variant mCherry-H, the MFE of the first 48 nt coding region was −5.91 kcal/mol, whereas after codon optimization the corresponding high-expression variant exhibited an even more stable 5′-end structure, with an increased MFE of −6.99 kcal/mol. \u003cstrong\u003e(C)\u003c/strong\u003e Complete table of identified NICs. All NICs are highlighted with red background boxes and white text.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-8581843/v1/565450b24094089552bf251a.png"},{"id":105706095,"identity":"60630f84-1a70-414a-8075-70eaed9fe762","added_by":"auto","created_at":"2026-03-30 07:05:20","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":437122,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003etRNA supply analysis for different codons.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(A)\u003c/strong\u003e Workflow of intracellular tRNA sequencing. \u003cstrong\u003e(B)\u003c/strong\u003e Abundance of perfectly decoding tRNAs for each codon, with NICs indicated by red arrows. The host genome encodes only 39 of the 61 possible tRNA species. It shows no clear correlation with the abundance of their corresponding perfectly matching tRNAs. Samples were collected after 3 h of fermentation at 16 °C. \u003cstrong\u003e(C)\u003c/strong\u003e Total tRNA supply capable of decoding each codon, including perfectly decoding tRNAs (light green) and wobble-decoding tRNAs (light red). This analysis further confirms that NICs are not associated with a low supply of either specific or total decoding tRNAs. NICs are highlighted with red arrows.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-8581843/v1/8db7e702adaaac86061a9109.png"},{"id":105728992,"identity":"b2951a76-8a8b-43f2-b1c5-df01b55547e4","added_by":"auto","created_at":"2026-03-30 11:13:12","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":313171,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eTranslational differences between a low-expression NIC-rich sequence and its NIC-repaired high-expression counterpart.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(A)\u003c/strong\u003e Schematic of the polysome profiling assay. A low-fluorescence variant (Sequence-A) and its high-fluorescence counterpart (Sequence-A7, with NICs repaired) were harvested at the same growth phase. \u003cstrong\u003e(B)\u003c/strong\u003e Fluorescence trajectories of Sequence-A and Sequence-A7 during fermentation, with sampling points indicated by arrows. \u003cstrong\u003e(C)\u003c/strong\u003e Workflow of ribosome (polysome) profiling. Cell lysates were subjected to sucrose-gradient ultracentrifugation to fractionate mRNAs based on their ribosome occupancy and translational states. \u003cstrong\u003e(D)\u003c/strong\u003e Global polysome profiles of Sequence-A and Sequence-A7 after 3 h of fermentation. Sequence-A displayed substantially higher proportions of mRNA associated with 50S/70S particles, whereas polysome-associated mRNA was markedly reduced compared with Sequence-A7. \u003cstrong\u003e(E)\u003c/strong\u003e Distribution of mCherry mRNA across different translational stages at 3 h. In Sequence-A, the majority of transcripts accumulated in initiation or early-elongation fractions, whereas in Sequence-A7 most were engaged in efficient elongating. \u003cstrong\u003e(F)\u003c/strong\u003e Global polysome profiles after 6 h of fermentation. It shows trends similar to those observed at 3 h. \u003cstrong\u003e(G)\u003c/strong\u003e Distribution of mCherry mRNA across translational stages at 6 h. Sequence-A showed increased proportions of free and initiating transcripts, while Sequence-A7 exhibited an even higher proportion of efficient elongating.\u003cstrong\u003e (H)\u003c/strong\u003e Binding and dissociation kinetics of ribosomes interacting with the SD-containing mCherry-A N-terminal sequence. Sensorgrams show the association and dissociation curves of ribosomes at increasing concentrations binding to the first 18 bp of the mCherry-A sequence containing a Shine–Dalgarno (SD) motif. The kinetic parameters were obtained by fitting the data to a 1:1 binding model, yielding an association rate constant ka=2.82×10\u003csup\u003e6\u003c/sup\u003e M\u003csup\u003e-1\u003c/sup\u003es\u003csup\u003e-1\u003c/sup\u003e, a dissociation rate constant kd=1.00×10\u003csup\u003e-2\u003c/sup\u003e s\u003csup\u003e-1\u003c/sup\u003e, and an equilibrium dissociation constant KD=3.55×10\u003csup\u003e-9\u003c/sup\u003e M. (I) Binding and dissociation kinetics of ribosomes interacting with the SD-containing mCherry-A-7 N-terminal sequence. Sensorgrams show the association and dissociation curves of ribosomes at increasing concentrations binding to the first 18 bp of the mCherry-A sequence containing a Shine–Dalgarno (SD) motif. The kinetic parameters were obtained by fitting the data to a 1:1 binding model, yielding an association rate constant ka=7.6×10\u003csup\u003e7\u003c/sup\u003e M\u003csup\u003e-1\u003c/sup\u003es\u003csup\u003e-1\u003c/sup\u003e, a dissociation rate constant kd=7.51×10\u003csup\u003e-3\u003c/sup\u003e s\u003csup\u003e-1\u003c/sup\u003e, and an equilibrium dissociation constant KD=9.88×10\u003csup\u003e-11\u003c/sup\u003e M. The colored curves represent the raw sensorgrams recorded by the Biacore system, whereas the black curves indicate the corresponding kinetic fitting.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-8581843/v1/1b8812a6f5aa5ca6f92196a9.png"},{"id":105706091,"identity":"8879f037-fd31-49e0-81e4-8d0389f739ae","added_by":"auto","created_at":"2026-03-30 07:05:20","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":430982,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eA Proteomics-Guided Forward Design Strategy for Optimizing 5’-End Codon Usage to Enhance Heterologous Protein Expression.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(A)\u003c/strong\u003e Workflow for the proteomics-guided strategy. The top 1,500 most abundant intracellular proteins were identified through host cell proteomics, and their corresponding DNA sequences were retrieved from the genome.\u003cstrong\u003e (B) \u003c/strong\u003ePanoramic view of mass spectrometry intensities for digested peptides. \u003cstrong\u003e(C)\u003c/strong\u003eWorkflow for the high-abundance protein sequence mapping The high-abundance proteins identified through proteomics were first mapped to their full-length amino acid sequences. Subsequently, genomic mapping was performed to obtain the corresponding nucleotide sequence \u003cstrong\u003e(C)\u003c/strong\u003e Usage counts of each codon within the 5’ ends 48 bp of the top 500 high-abundance proteins. NICs are highlighted with red arrows. \u003cstrong\u003e(D)\u003c/strong\u003e Comparison of codon usage frequencies between the 5’ end and the full-length sequence. Codons with elevated usage frequency at the 5’ end are marked with red vertical lines, while codons with high usage frequency across the full-length sequence are marked with black vertical lines. NICs are highlighted with red arrows.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-8581843/v1/8e7ca863a5ccf802975d163f.png"},{"id":105706093,"identity":"57c85cbf-8c53-4183-9268-4116f5cbcc02","added_by":"auto","created_at":"2026-03-30 07:05:20","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":705327,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe new codon-usage criterion for efficient heterologous protein overexpression in \u003c/strong\u003e\u003cem\u003e\u003cstrong\u003eE. coli \u003c/strong\u003e\u003c/em\u003e\u003cstrong\u003eand its effects.\u003c/strong\u003e(A) Schematic illustration of the new codon-usage criterion for heterologous gene design.The design process follows a two-section strategy. Within the N-terminal 48 bp region, codons are selected according to the N-terminal Optimal Codon (NOC) scheme, in which each amino acid is encoded by a single preferred codon. For the downstream coding region, codon usage is designed by preferentially employing codons enriched in highly expressed host proteins (e.g., AAC, CAC, TAC, and TTC) while avoiding known rare codons, thereby maintaining efficient translational elongation. (B–J) SDS–PAGE analysis of different target proteins before and after codon optimization. (B) Molecular chaperone Pfdb. (C) Molecular chaperone Tms14ED. (D) Death effector domain proteins. (E) Suppressors of cytokine signaling. (F) Laccase. (G) Molecular chaperone Tms14AD. (H) Lipase. (I) Cysteine hydrolase. For each target protein, 1 corresponds to the native (wild-type) sequence, 2 to the sequence optimized using a commercial codon-optimization tool, and 3 to the sequence designed using the new codon-usage criterion proposed in this study. S, soluble (supernatant) fraction; P, insoluble (pellet) fraction.\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-8581843/v1/e131e7fb22cc33886285f3b7.png"},{"id":105903934,"identity":"b792f00f-4dc8-4800-bf65-f5848b4e0600","added_by":"auto","created_at":"2026-04-01 09:58:45","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3898230,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8581843/v1/cf22e1f6-1b3c-4dae-9b8e-94ed9d8bf12f.pdf"},{"id":105728755,"identity":"53eff10a-993d-48d0-b217-82f758787931","added_by":"auto","created_at":"2026-03-30 11:12:36","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":2862033,"visible":true,"origin":"","legend":"SUPPLEMENTARY INFORMATION","description":"","filename":"SI.docx","url":"https://assets-eu.researchsquare.com/files/rs-8581843/v1/6039eb59d68a92587d92299a.docx"},{"id":105706099,"identity":"54e25584-d3d8-4d96-8d3d-1693043e2897","added_by":"auto","created_at":"2026-03-30 07:05:20","extension":"pdf","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":72201,"visible":true,"origin":"","legend":"Reporting Summary","description":"","filename":"rs.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8581843/v1/775040cc458be3d61f0537bb.pdf"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"N terminal wobble base usage determines ribosome loading and thus protein expression fate","fulltext":[{"header":"Introduction","content":"\u003cp\u003eHeterologous protein expression is foundational to modern biomedicine, diagnostics and industrial biocatalysis: it enables therapeutic proteins and vaccines, affords high-specificity analytical reagents, and powers enzyme-driven manufacturing at scale. Prokaryotic hosts such as \u003cem\u003eEscherichia coli\u003c/em\u003e remain the workhorses for recombinant production owing to their fast growth, low cost, genetic tractability, and straightforward scale-up [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. In recent years, the importance of heterologous expression has been further amplified by the rapid rise of sequence-designed proteins, driven by advances in protein engineering, directed evolution, and AI-assisted sequence generation. These approaches are dramatically accelerating the design\u0026ndash;build\u0026ndash;test cycle for enzymes and functional proteins, enabling rapid creation of variants tailored for applications in biomanufacturing, synthetic biology, environmental remediation, and healthcare[\u003cspan additionalcitationids=\"CR4\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. However, the functional realization of these designed sequences critically depends on their successful expression in heterologous hosts. Yet, the frequent and often unexplained failure or extremely low yield of target proteins remains a fundamental and pervasive bottleneck that has long challenged the field[\u003cspan additionalcitationids=\"CR7\" citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. A study by the Northeast Structural Genomics (NESG) platform indicates that 28% of heterologous proteins could not be expressed at all in \u003cem\u003eEscherichia coli\u003c/em\u003e[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. To address this challenge, numerous strategies have been developed, including vector and host optimization, promoter engineering, ribosome binding site design, and the use of fusion tags [\u003cspan additionalcitationids=\"CR11 CR12\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e], which we have also been reviewed in detail recently[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Nevertheless, a substantial proportion of proteins\u0026mdash;particularly those derived from eukaryotes or extremophiles\u0026mdash;still fail to reach detectable expression levels in \u003cem\u003eE. coli\u003c/em\u003e and other prokaryotic hosts, highlighting a profound gap in our current understanding of sequence-dependent control of gene expression.\u003c/p\u003e \u003cp\u003eWhen improvements in vectors or hosts fail, the obstacle often resides in the gene sequence itself. Classical sequence-level strategies\u0026mdash;most prominently, \u0026ldquo;rare-codon optimization\u0026rdquo; across the open reading frame\u0026mdash;have enabled high expression for many targets in \u003cem\u003eE. coli\u003c/em\u003e[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan additionalcitationids=\"CR16\" citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. However, a non-trivial subset of heterologous proteins still fail to express for reasons that are not captured by traditional rarity-focused or genome-wide codon-usage heuristics[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. These approaches are largely agnostic to the fine-scale organization of codons at the 5\u0026rsquo; end, where translation is initiated and early elongation is established.\u003c/p\u003e \u003cp\u003eIn bacteria, translation initiation is the dominant kinetic checkpoint that sets protein output; features concentrated in the first\u0026thinsp;~\u0026thinsp;30\u0026ndash;50 nt of the coding sequence strongly modulate ribosome loading, start-site accessibility, and the handoff into early elongation [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Consistent with this, multiple studies have demonstrated that sequence features within the 5\u0026rsquo; coding region exert a decisive influence on expression efficiency, and that rational optimization of this region can markedly alter the expression levels of target proteins[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe sequence features of the translation initiation region are widely recognized as a key determinant of protein synthesis efficiency, with codon choice at the 5\u0026rsquo; end being particularly critical. Accumulating evidence suggests that the 5\u0026rsquo;-terminal region of gene may act as a rate-limiting element during translation initiation. Previous studies have shown that the first 18 bp of the open reading frame correspond to the ribosome-protected fragment of the 70S complex [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e], and that the sequence composition of this region directly affects the efficiency of translation initiation[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. In addition, an extended downstream window of approximately 30 bp has also been proposed to serve as a critical translation regulatory segment [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. Nevertheless, the generalizable rules governing this region remain poorly defined, and the precise contribution of individual codons to translation initiation and early elongation has yet to be systematically elucidated.\u003c/p\u003e \u003cp\u003eRecent work has further emphasized 5\u0026rsquo;-end mRNA structure as a putative core determinant of this region: stable hairpins formed immediately downstream of the start codon are thought to impede ribosome binding and scanning, thereby suppressing translation initiation[\u003cspan additionalcitationids=\"CR26 CR27 CR28 CR29 CR30\" citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. For instance, Goodman et al. demonstrated in a large-scale study of sfGFP variants that reducing 5\u0026rsquo;-end mRNA secondary structure markedly increased protein yield[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIs true that it\u0026rsquo;s the 5\u0026rsquo;-end mRNA-secondary structure determines the expression fate of diverse proteins? Is there any other unknown factor that plays a critical role? Here, we systematically interrogate the 5\u0026prime; coding region using large-scale synonymous libraries and uncover a previously unrecognized, codon-level determinant of translation initiation. We show that specific N-terminal codons impair ribosome loading and dictate the expression outcome of heterologous proteins, enabling a new, mechanism-grounded framework for rational gene design in prokaryotes.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003eN-terminus mRNA structure is not the dominant factor for protein expression\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAs mentioned above, poor expression of heterologous proteins has been largely attributed to the highly-structured mRNA at the N-terminus. Indeed, several studies have reported a significant correlation between mRNA folding energy at the N-terminus and protein expression level[25]. To test the accuracy of this conclusion, we constructed a synonymous mutation library encompassing the first 48 bp of the red fluorescent protein (mCherry) gene, thereby generating an N-terminal synonymous library (\u0026gt;10⁷ variants). We randomly chose 320 members for complete sequencing and fluorescence measurement to directly link sequence diversity (generating different mRNA secondary structures) to expression outcomes in terms of fluorescence. Using this dataset, we calculated the mRNA folding energy of each variant and analyzed its correlation with fluorescence intensity. As shown in Fig. 1A, 20 representative 48 bp N-terminal synonymous mutant DNAs generated 20 diverse structures of mRNA (first 48 nt coding region) with different folding energy scores, expressing mCherry with varied fluorescence intensities. Correlating the results of mRNA folding Minimum Free Energy (MFE) with fluorescence intensities, we summarized how mRNA structures affected the mCherry expression in \u003cem\u003eE.coli\u003c/em\u003e and \u003cem\u003eR. ruber\u003c/em\u003e host, respectively, as shown in Fig. 1B and 1C.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn \u003cem\u003eE. coli\u003c/em\u003e, mCherry expression levels were negatively correlated with mRNA structural stability (r=0.31, p\u0026lt;10\u003csup\u003e-4\u003c/sup\u003e), seemingly consistent with the\u0026nbsp;“mRNA structural hindrance”\u0026nbsp;model. Strikingly, however, when the same mCherry-library was expressed in another prokaryotic chassis, \u003cem\u003eR. ruber\u003c/em\u003e, a high-GC-content host, an entirely opposite trend emerged: expression levels were positively correlated with mRNA stability (r=-0.2007, p\u0026lt;10\u003csup\u003e-3\u003c/sup\u003e); that is, more stable transcripts yielded higher protein expression. This finding fundamentally contradicted the conventional mRNA structure-inhibition theory, indicating that translation initiation efficiency might not be explained solely by mRNA-secondary-structure effects.\u003c/p\u003e\n\u003cp\u003eTaken together, our experimental results revealed a host-dependent reversal that cannot be reconciled by a single\u0026nbsp;“structure-is-bad”\u0026nbsp;or\u0026nbsp;“structure-is-good”\u0026nbsp;rule.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eIdentification of N-terminal-specified Incompatible Codons (NICs) causing non-expression of proteins\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNow that the mRNA-secondary structure is not the dominant factor governing the expression of proteins, what is the hidden factors behind? To address this, we introduced the N-terminal-48bp- synonymous mutation library for\u0026nbsp;10 amino acids in the previous section (mCherry 16 aa including Met, Val, Ser, Lys, Gly, Glu, Asp, Asn, Ala and Ile) and bridged the specific codon-fluorescence correlation. As shown in Fig. 2A, we surprisingly identified the existence of some specific codons for amino acids, which significantly reduced the expression of mCherry. For instance, at the second position (Valine), the four synonymous codons (GTT, GTC, GTA, GTG) resulted in significantly different protein expression levels. A post-hoc test revealed that mutants with the GTC codon exhibited markedly lower expression than those with the other three synonymous codons.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo validate this finding, we analyzed the codon distribution within the lowest 5% expressing population. Of the 14 mutants in this group, 9 utilized the GTC codon at this position—a frequency far exceeding random expectation. Similar phenomena were also exhibited for Lys, Glu, and Asn (Fig. S2).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe further selected 4 other functionally diverse model proteins: green fluorescent protein (GFP), nattokinase (NK), serine hydroxymethyl transferase (Bsu) from \u003cem\u003eBacillus subtilis\u003c/em\u003e, and thaumatin (Tha) from \u003cem\u003eThaumatococcus danielli\u003c/em\u003e with 16 aa termius as shown in Fig. 2B, respectively, to observe the N-terminus-codon effects. To enable the high-throughput screening and analyses, the mCherry reporter was fused to the C-terminus of NK, Bsu and thaumatin via a flexible (G₄S)₃ linker (Fig. 2A). Sequencing and characterization of over 4,000 clones yielded 1,525 validated mutants in total, establishing a large-scale 5′-end variant dataset that directly connects the codon-usage with fluorescence across these five systems (Fig. 2A, 2B-2K, and Fig. S3-S7). For each 16 aa-fragment, we correlated the codon usage with fluorescence, and through hypothesis testing, identified the low-expression codons (defined by p \u0026lt; 0.01) for each amino acid. For example, AAG for Lys in mCherry and GFP, ACC for Thr in GFP and Tha, CAG for Gln in Bsu, and GTG for Val in the NK library.\u003c/p\u003e\n\u003cp\u003eFurther, these low-expression-correlated codons exhibit positional specificity. For instance, in a low-expression variant containing 12 AAG codons at different positions, we found that only substituting the three N-terminus AAGs into AAA raised fluorescence from 5,741 to 18,756. In contrast, substituting 9 downstream AAGs while retaining the two N-terminus AAGs, the fluorescence didn’t improve (5,979). That is, the inhibitory effect of AAG is confined to the 5’ initiation region and does not extend into subsequent elongation.\u003c/p\u003e\n\u003cp\u003eTo further confirm the dominant role of the N-terminus, we performed domain-swapping experiments. Synonymous libraries were constructed by replacing 48 bp windows at different positions (codons 1-16, 17-32, and 118-132)\u0026nbsp;in both high-expression and low-expression sequences. The outcome was consistent. The N-terminus without low-expression codons ensured high expression regardless of downstream codon composition. Conversely, another N-terminus containing 6 low-expression-correlated codons resulted in persistently low protein-expression levels (Fig.2L).\u003c/p\u003e\n\u003cp\u003eAnother unexpected finding is that these low-expression-correlated codons in \u003cem\u003eE.coli\u003c/em\u003e are not rare ones. The aforementioned GTC codon, for example, is a common non-rare codon with a genomic usage frequency of 21%. Moreover, archetypal high-frequency codons, such as ACC for Threonine (Thr) (22.8%), were also identified as low-expression-leading codons. Conversely, some classic rare codons, such as ATA for Isoleucine (Ile) (5.5%) and GGA for Glycine (Gly) (12%), did not exert a significant negative effect on expression. This clearly demonstrates that the inhibitory effect of these specific codons is independent of conventional genomic codon usage frequency(Fig. 2N).\u003c/p\u003e\n\u003cp\u003eTherefore, we finally defined these special low-expression-correlated codons as N-terminal-specified Incompatible Codons (NICs), which existed in 13 out of 20 natural amino acids. These NICs display two defining properties: (i) they are N-terminus-specific, exerting strong effects only near the start of the coding sequence; and (ii) they maybe are not rare codons in the conventional sense.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAgain, we further collected the poor-expressed mutants and analyzed the 48 bp N-terminus, it’s found that all of the poor-expressed targets contained 5-11 low-expression codons (Fig. 2M). Across the five libraries (1,525 validated mutants), the bottom 2% low-expression sequences contained a total of 582 NIC occurrences within the 1,200 N-terminal positions of the 75 low-expression variants.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eValidation of the NIC’s negative effects on protein expression\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe next conducted systematic experimental validation for the negative effects of NICs. We specifically selected two of the weakest variants, mCherry-A and mCherry-B, each carrying 11 NICs. Through individually replacing each NIC with a synonymous non-NIC, we observed that the fluorescence intensity increased by approximately 1.9～3 folds upon each single NIC repair. Substitution of all NICs resulted in more than a six-fold fluorescence increase (from 5,574 to 34,239 for mCherry-A, and from 5,741 to 34,051 for mCherry-B) (Fig.3A). Extending this strategy to the variant library with the lowest fluorescence intensity (mCherry-C-H), we consistently observed a substantial recovery in fluorescence, increasing from the bottom 5% to the top 20% (Fig. 3B). These results demonstrated that the cumulative accumulation of NICs is the major cause of protein-non-expression, and their in-total replacement can completely reverse the failure expression.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eCalculation of the 5′-end mRNA folding free energy revealed a more nuanced relationship between local sequence changes, mRNA structure, and expression output. Specifically, compared with the low-expression parental sequences, substitution of the second-position Val codon from GTC to GTA led to a pronounced decrease in the absolute folding free energy of the first 48 nt coding region of the mRNA, indicative of reduced structural stability (A-1 vs. A; B-1 vs. B). This structural relaxation was accompanied by an approximately two-fold increase in fluorescence intensity. In contrast, when NICs at other downstream positions were substituted, the folding free energy showed no significant reduction; notably, in the B-3 variant—where the fourth-position Lys codon AAG was replaced with AAA—the folding free energy was even significantly increased relative to the parental B sequence. Despite this apparent increase in mRNA structural stability, fluorescence intensity was nevertheless markedly enhanced. (Fig. 3A). Moreover, neither classical predictions using the Zuker algorithm nor advanced models based on deep learning attributed the decisive effect of NICs to structural changes (Fig.S8 and Fig. S9). This means that the NIC effect has no significant relationship with the mRNA secondary structure.\u003c/p\u003e\n\u003cp\u003eFurther studies showed that the inhibitory effects of part of NICs exhibit a degree of sequence context dependency. For example, for Val, GTC acted as the strongest unfit codon to mCherry, whereas GTG and GTT for Val displayed a stronger inhibitory effect in the Bsu and GFP libraries. This suggests that the impact of some NICs showed protein-specific features. Accordingly, codons that consistently exhibited strong inhibitory effects across at least one N-terminal context were operationally classified as NICs, and finally we obtained a NIC table, as shown in Fig. 3C, in which all of the NICs are highlighted with red background boxes and white text. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eThe NIC Effect is Not Caused by tRNA Scarcity\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFollowing the identification of NICs, a central question naturally arose: what underlies the NIC effect? In the classical rate-limiting model of translation, protein non-expression or low expression is typically attributed to an insufficient supply of specific tRNAs. To test whether this traditional mechanism accounts for the effect of NICs, we systematically measured intracellular tRNA abundance in \u003cem\u003eE. coli\u003c/em\u003e and compared the data with the identified NICs. Unlike early approaches that inferred tRNA availability from genomic copy numbers to calculate the tRNA adaptation index (tAI), we used high-throughput sequencing to directly quantify intracellular tRNA levels, thereby avoiding discrepancies between copy number and actual supply (Fig. 4A)[32, 33].\u003c/p\u003e\n\u003cp\u003eWe first examined the relationship between codons and the abundance of their strictly matched cognate tRNAs. The analysis showed no direct association between whether a codon is a NIC and the intracellular concentration of its cognate tRNA. Several findings contradicted the tRNA- insufficient supply hypothesis. For example, aspartate codon GAC was identified as a strong NIC in multiple libraries, yet its cognate tRNA-GTC is present at relatively high abundance. In contrast, its synonymous non-NIC codon GAT lacks a cognate tRNA in the \u003cem\u003eE. coli\u003c/em\u003e genome and is entirely decoded by wobble pairing through tRNA-GTC. Similarly, the phenylalanine codon TTC, classified as a NIC, corresponds to a tRNA present at much higher abundance than that of its synonymous non-NIC codon TTT. These counterintuitive examples directly challenge the view that insufficient tRNA supply is the primary cause of NIC effect (Fig.4B).\u003c/p\u003e\n\u003cp\u003eBecause wobble pairing is pervasive in bacterial translation, analyses based solely on cognate tRNA abundance may fail to capture the full decoding landscape. We therefore performed a second-level analysis in which the abundances of all tRNAs capable of decoding a given codon—including those engaging in wobble interactions—were summed to estimate its total decoding capacity. This expanded analysis again failed to support the tRNA-abundance hypothesis. The inhibitory effect of NICs showed no correlation with either total decoding capacity or the mode of decoding, i.e., whether a codon is read exclusively through wobble pairing or by a cognate tRNA. For example, the glutamate codon GAG, classified as a NIC, is decoded entirely via wobble by tRNA-TTC, whereas the aspartate codon GAC, also a NIC, is decoded by its cognate tRNA. Despite these fundamentally different decoding modes, both codons exhibited comparable inhibitory effects on expression. Conversely, strong counter-evidence was observed for valine: the two NIC codons GTT and GTC displayed significantly higher total decoding tRNA abundance than the synonymous non-NIC codon GTA (Fig. 4C).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTaken together, analyses of both cognate tRNA abundance and total decoding capacity consistently indicate that the inhibitory effect of NICs does not arise from limited aminoacyl-tRNA supply. This conclusion rules out the most common rate-limiting mechanism in translation and strongly suggests that the NIC effect originates from deeper molecular events during translation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNIC effect is caused by blocking of the translation initiation and early elongation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA new reasonable deduction for the molecular basis of NIC is the altered dynamics of the translational machinery itself. To investigate this mechanism, we turned to ribosome-level analysis and performed polysome profiling, which separates mRNAs according to the number of bound ribosomes and provides a snapshot of their translational state. As model systems, we again employed mCherry-A (harboring 11 NICs and exhibiting a fluorescence intensity of only 5,574) and mCherry-A7 (with these NICs removed, showing a markedly higher fluorescence of 34,239) (Fig. 5A).\u003c/p\u003e\n\u003cp\u003eConsistent with our previous observations, the overall polysome profiles of the two strains showed striking differences at both the early fermentation stage (3 h) and the mid-log phase (6 h), as shown in Fig. 5D and Fig. 5F, respectively. In mCherry-A7, a pronounced enrichment in heavy polysome fractions was detected at both time points, indicating that a large proportion of mRNAs underwent efficient and concurrent translation by multiple ribosomes. In contrast, mCherry-A displayed a clear shift toward lighter fractions, suggesting that most of its transcripts were stalled at the initiation stage or remained bound by only a single ribosome.\u003c/p\u003e\n\u003cp\u003eTo quantify the translational dynamics of the target mRNA, we measured the distribution of mCherry transcripts across gradient fractions by qPCR. At 3 h, the mCherry-A strain showed that 46.0% (3.95% free + 42.1% initiation) of transcripts accumulated before or at early initiation, with another 33.8% trapped in early elongation and only 20.0% reaching stable polysomes. In contrast, mCherry-A7 exhibited a dramatically different pattern, with 68.8% of mRNAs residing in polysome fractions, and only 27.4% (8.52% free + 18.91% initiation) stalled prior to productive elongation (Fig. 5E).\u003c/p\u003e\n\u003cp\u003eA similar but even more pronounced divergence was observed at 6 h. In mCherry-A, the majority of transcripts (69.9% = 21.43% free + 48.49% initiation) remained restricted to pre-initiation or early-initiation states, and merely 11.5% entered stable elongation. Meanwhile, in mCherry-A7, over 81.9% of transcripts accumulated in polysomes, with minimal retention in pre-initiation (14.4%) or early elongation (3.6%) fractions (Fig. 5G).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThese results provide direct molecular evidence for the mechanism of NIC repression. Consistent with this notion, preliminary docking analyses suggested that high-expression N-terminal sequences may exhibit improved compatibility with 16S rRNA at the mRNA 5′\u0026nbsp;end [34] (Fig. S11).The presence of NICs creates barriers at initiation or early elongation, sharply reducing ribosome entry and processive scanning along the coding sequence. As a consequence, many transcripts remain in low-efficiency states, leading to reduced protein output. This mechanism explains the translational bottleneck caused by NICs and integrates with the findings from the preceding analyses.\u003c/p\u003e\n\u003cp\u003eTo further validate that NICs repress translation by directly impairing ribosome recruitment and initiation complex stability, we next quantified the binding affinity between\u003cem\u003e\u0026nbsp;E. coli\u003c/em\u003e ribosomes and mRNAs using surface plasmon resonance (SPR, Biacore). This approach enabled direct measurement of ribosome–mRNA interactions at the molecular level, independent of downstream elongation effects.\u003c/p\u003e\n\u003cp\u003eFour representative RNA constructs were designed to disentangle the contributions of the Shine–Dalgarno (SD) sequence and the early coding region: (i) SD sequence fused to the first 18 bp of mCherry-A, (ii) SD sequence fused to the first 18 bp of mCherry-A7 (with NICs replaced), (iii) the first 48 nt of mCherry-A without an SD sequence, and (iv) the first 48 nt of mCherry-A7 without an SD sequence. These constructs allowed us to specifically probe whether NICs affect ribosome binding in an SD-dependent or SD-independent manner.\u003c/p\u003e\n\u003cp\u003eStrikingly, the SD–mCherry-A (18 bp) RNA exhibited markedly weaker ribosome binding, with an equilibrium dissociation constant (KD) of 3.55 × 10⁻⁹ M. This affinity was comparable to that observed for both SD-lacking constructs—mCherry-A (48 bp) and mCherry-A7(48 bp)—indicating inefficient ribosome engagement when NICs were present in the immediate post-start region. In sharp contrast, the SD–mCherry-A7(18 bp) RNA displayed substantially stronger ribosome association, with a KD of 9.88 × 10⁻¹¹ M, representing over a 30-fold increase in binding affinity relative to its NIC-containing counterpart(Fig. 5H and 5I).\u003c/p\u003e\n\u003cp\u003eThese results demonstrate that NICs profoundly weaken ribosome–mRNA binding even in the presence of an intact SD sequence, whereas removal of NICs restores high-affinity ribosome association. Importantly, the similar affinities observed for SD-free RNAs indicate that the enhanced binding of mCherry-A-7 arises specifically from cooperative interactions between the ribosome and the early coding region rather than from nonspecific RNA–ribosome contacts.\u003c/p\u003e\n\u003cp\u003eTogether with the polysome profiling data, these findings provide convergent molecular evidence that NICs impose a translational bottleneck by destabilizing ribosome loading and early initiation complexes. By reducing ribosome entry efficiency and preventing productive transition into elongation, NICs trap mRNAs in low-occupancy states, thereby sharply limiting protein output. This mechanism offers a direct physical explanation for NIC-mediated translational repression and integrates seamlessly with our ribosome-level and expression-level analyses.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAT/GC3-driven host differences in NICs\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo further assess the universality and host dependence of the NIC phenomenon, we applied the same “library construction–NIC screening” workflow to \u003cem\u003eRhodococcus ruber\u003c/em\u003e. Remarkably, NICs were also readily detectable in this high-GC Gram-positive actinobacterium, and individual inhibitory codons exerted strong suppressive effects on recombinant protein expression. These findings demonstrate that the NIC mechanism is not an idiosyncrasy of \u003cem\u003eE. coli\u003c/em\u003e but instead represents a broadly conserved layer of sequence-encoded translational control across phylogenetically distant prokaryotes.\u003c/p\u003e\n\u003cp\u003eDespite this universality, the inhibitory landscapes of NICs diverged sharply between hosts at the level of wobble base usage. In \u003cem\u003eE. coli\u003c/em\u003e, the Val codon GTC was among the strongest inhibitors of mCherry expression, whereas in \u003cem\u003eR. ruber\u003c/em\u003e all four Val codons behaved equivalently and showed no detectable repression. The same pattern was observed for Asn, where \u003cem\u003eE. coli\u003c/em\u003e favored AAT, while \u003cem\u003eR. ruber\u003c/em\u003e exhibited a clear preference for AAC (Fig. S10).\u003c/p\u003e\n\u003cp\u003eThese host-specific patterns were also evident at the global level of NIC composition. Among all 20 NICs identified in \u003cem\u003eE. coli\u003c/em\u003e, 17 were GC3-ending codons, whereas in \u003cem\u003eR. ruber\u003c/em\u003e, 8 of the 9 NICs discovered in our current screening set were AT3-ending. This GC3-to-AT3 polarity switch represents a fundamental rewiring of N-terminal wobble base usage between the two hosts, rather than a simple shift in codon frequency.\u003c/p\u003e\n\u003cp\u003eThis polarity shift in third-position preference provides an explanatory framework for the opposite “mRNA structure–expression” correlations observed in the two systems. In \u003cem\u003eE. coli\u003c/em\u003e, the dominant GC3 endings tend to stabilize local RNA structures, which correlates with the classical “more stable → lower expression” pattern. In contrast, in \u003cem\u003eR. ruber\u003c/em\u003e, inhibitory elements are predominantly AT3-type codons, which usually destabilize local structures; as a result, a “less stable → lower expression” trend emerges—mirroring previous observations in actinobacterial expression systems[25, 26, 35].\u003c/p\u003e\n\u003cp\u003eTogether, these findings reveal that the inhibitory effects of NICs cannot be attributed to RNA secondary structure alone. Instead, they reflect host-specific molecular rules governing translation initiation and early elongation, which are encoded at the level of third-position nucleotide identity. We therefore propose that NICs constitute a universal yet evolutionarily diversified translational control module, whose functional divergence arises from distinct wobble-position decoding constraints that shape the early ribosomal engagement landscape, ultimately determining the expression fate of heterologous proteins.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMining N-terminal optimal codons and selecting downstream-favored codons from naturally-overexpressed proteins\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eBuilding on our prior identification of NICs and their inhibitory effects on ribosome loading, we hypothesized that naturally overexpressed proteins should exhibit strong selective pressure against such inhibitory codons, particularly within the translation initiation region. To test this hypothesis, we performed deep DIA-based LC–MS/MS analysis on intracellular soluble proteins and quantified protein abundance using intensity-based absolute quantification (iBAQ). This approach provided a comprehensive and quantitative snapshot of the endogenous proteome across a wide dynamic range. Based on iBAQ intensities, proteins were ranked by abundance, and the top 1,500 most abundant proteins were selected as a high-confidence pool representing the highly expressed endogenous proteome.\u003c/p\u003e\n\u003cp\u003eTo relate protein-level abundance to underlying sequence features, the identified protein groups were mapped to their corresponding genes in the reference genome of the host strain. For each of the top 1,500 abundant proteins, the matched coding sequence (CDS) was retrieved, thereby establishing a direct link between experimentally measured protein abundance and its underlying gene sequence. From this high-abundance pool, we further focused our analysis on the top 500 most abundant proteins in order to minimize potential noise from mid-abundance species and to concentrate on the most robustly expressed endogenous genes.\u003c/p\u003e\n\u003cp\u003eCodon-usage analysis was first focused on the N-terminal region (the first 16 codons, 48 bp), where NICs exert their strongest effects on translation initiation. As expected, many NICs, such as GGG (Gly) and TCG (Ser), were strongly depleted at the 5′ ends of these highly expressed genes. Notably, however, NIC exclusion was not absolute. A subset of NICs still appeared at measurable frequencies, suggesting that their inhibitory effects can be partially tolerated depending on local sequence context or protein-specific constraints (Fig. S14).\u003c/p\u003e\n\u003cp\u003eTo further quantify this selective pressure, we compared codon usage at the N terminus with that across full-length CDSs. Several NICs, including AAC (Asn) and ACC (Thr), remained among the most frequently used synonymous codons at the N terminus but were nonetheless significantly underrepresented relative to their genome-wide frequencies (e.g., AAC: 57% at the N terminus vs. 72% genome-wide). These codons therefore appear to be selectively constrained—but not strictly excluded—within the N-terminal region (Fig. 6D).\u003c/p\u003e\n\u003cp\u003eIn contrast, a distinct set of codons exhibited pronounced enrichment specifically at the N terminus. For example, ACA (Thr) increased from 6.6% in full-length CDSs to 18.3% at the N terminus, while CAA (Gln) increased from 25.1% to 49.5%. These observations indicate that the N-terminal coding region of highly expressed genes is shaped not only by avoidance of inhibitory codons but also by positive selection for host-preferred “N-terminal-favored” codons that likely facilitate efficient ribosome engagement and early elongation.\u003c/p\u003e\n\u003cp\u003eHaving characterized codon selection at the 5′ end, we next examined whether codon usage within the downstream coding region also correlates with expression level. Analysis of full-length CDSs across different abundance tiers (top 500, top 1,500, and genome-wide background) revealed that most codons displayed highly consistent usage across all groups. However, four codons—AAC (Asn), CAC (His), TAC (Tyr), and TTC (Phe)—showed a clear monotonic enrichment with increasing protein abundance, suggesting preferential usage during efficient elongation of naturally overexpressed proteins (Fig. S13).\u003c/p\u003e\n\u003cp\u003eTogether, these analyses reveal a compartmentalized codon-selection landscape in highly expressed genes: the N-terminal region is subject to strong, region-specific selective pressures, while the downstream region is largely governed by global codon bias with limited enrichment of a small subset of elongation-favored codons.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCodon-usage-new-criteria for overexpression of heterologous proteins in \u003cem\u003eE. coli\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAlthough NICs were generally depleted within the N-terminal 48 bp of highly expressed proteins, a more striking pattern emerged when we further focused on the extreme N-terminal core region immediately following initiation. Specifically, within the first 18 bp of the most abundant proteins (top 100), certain codons exhibited extremely low occurrence frequencies (below 0.05), indicating a pronounced level of depletion at an exceptionally short sequence scale. This observation suggests the existence of “core-region incompatible codons” that exert a disproportionate inhibitory effect on translation during the earliest elongation steps (Fig. S15).\u003c/p\u003e\n\u003cp\u003eIntegrating this finding with our systematic analyses of NICs and N-terminal codon preferences, we propose a new codon-usage criterion for efficient heterologous protein overexpression in \u003cem\u003eE. coli\u003c/em\u003e, based on a two-section design strategy (Fig. 7A).\u003c/p\u003e\n\u003cp\u003eSection 1 encompasses the N-terminal 48 bp region, which governs ribosome loading and early elongation efficiency. Codon selection within this region should strictly follow an N-terminal Optimal Codon (NOC) scheme, in which each amino acid is encoded by a single preferred codon. These optimal codons are defined by excluding both NICs and core-region incompatible codons, followed by selection of the synonymous codons most strongly enriched or dominant at the N termini of highly expressed host proteins.\u003c/p\u003e\n\u003cp\u003eSection 2 corresponds to the downstream coding region beyond the first 48 bp. In this region, codon design can follow a more conventional strategy, preferentially employing codons enriched in highly expressed proteins (e.g., AAC, CAC, TAC, and TTC) while avoiding known rare codons, thereby maintaining efficient elongation without introducing unnecessary ribosomal pausing. An on-line codon-design tool \u003cstrong\u003eRiboLoad Codon Optimizer\u003c/strong\u003e, has been available http://47.86.169.8/.\u003c/p\u003e\n\u003cp\u003eTo validate the efficacy and universality of this new criteria, we systematically tested a diverse panel of refractory proteins from various origins that are notoriously difficult to express in\u003cem\u003e\u0026nbsp;E. coli\u003c/em\u003e. These included archaeal molecular chaperones (Pfdb, Tms14AD, and Tms14ED), human death effector domain proteins and suppressors of cytokine signaling, \u003cem\u003ePleurotus ostreatus\u003c/em\u003e laccase, \u003cem\u003eYarrowia lipolytica\u003c/em\u003e lipase, and \u003cem\u003eBacillus subtilis\u003c/em\u003e cysteine hydrolase (Fig. 7B-7I).\u003c/p\u003e\n\u003cp\u003eAcross all tested proteins, the codon design strategy proposed in this study produced robust and consistent improvements in expression. For Tms14AD, Tms14ED, the suppressor of cytokine signaling, laccase, and cysteine hydrolase, neither the native sequences nor those optimized using mainstream commercial codon-optimization tools yielded detectable protein bands on SDS–PAGE. In contrast, application of our new codon-usage criterion resulted in a marked increase in expression levels. Notably, Tms14AD and Tms14ED were expressed in soluble form, whereas the remaining proteins accumulated at high levels as inclusion bodies.\u003c/p\u003e\n\u003cp\u003eMoreover, for Pfdb and the lipase—two proteins that exhibited modest but limited improvements following commercial codon optimization—our approach achieved substantially greater enhancement, with expression levels increasing by more than an order of magnitude relative to commercial optimization. Importantly, for laccase, lipase, and cysteine hydrolase, replacement of only the core incompatible codons within the N-terminal 18 bp region was sufficient to dramatically elevate expression, underscoring the disproportionate regulatory impact of this minimal sequence window (Fig.S16).\u003c/p\u003e\n\u003cp\u003eCollectively, these results demonstrate that the proposed codon-usage criterion not only rescues expression systems that completely fail under conventional optimization strategies, but also delivers order-of-magnitude gains even on top of existing optimizations, highlighting its broad applicability and mechanistic significance\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eEfficient expression of heterologous proteins remains a central and still largely unsolved challenge in modern biotechnology. Despite decades of optimization, a striking number of heterologous proteins\u0026mdash;across hosts and systems\u0026mdash;remain non-expressing or express at undetectable levels, with no clear mechanistic explanation. Current optimization strategies are largely built on two classical hypotheses. The first is the \u0026ldquo;rare codon\u0026rdquo; model, which attributes low expression to delays caused by limited supply of rare tRNAs. The second is the \u0026ldquo;mRNA secondary structure\u0026rdquo; model, which posits that G/C-rich sequences in the 5\u0026rsquo; region form stable hairpins that block ribosome binding and initiation. Based on these principles, researchers have developed codon optimization tools and codon usage tables such as CAI and tAI, which are widely applied in practice. However, real-world performance is far more complex. In the same host, \u003cem\u003eE. coli\u003c/em\u003e, some model proteins can be expressed at high levels, whereas others fail to express at all. This inconsistency highlights the limitations of existing codon tables and optimization rules, which cannot fully explain or resolve the \u0026ldquo;non-expression\u0026rdquo; problem. What, then, is the true cause of this discrepancy? Does an unrecognized rule exist that determines whether a protein can be translated successfully?\u003c/p\u003e \u003cp\u003eTo address this core scientific question, we constructed systematic 5\u0026rsquo; synonymous mutation libraries and uncovered a previously underappreciated class of regulatory elements\u0026mdash;NICs. These codons exert strong inhibitory effects in the translation initiation region and basically determine whether a target protein can be expressed. By sequencing and quantifying more than 4,000 independent clones, we established 1,525 precise genotype\u0026ndash;phenotype relationships. Unlike FACS-seq and other mixed-library approaches, this strategy eliminated noise from bulk sequencing and allowed single-codon contributions to be resolved[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eSeveral lines of evidence indicate that the NIC effect is fundamentally distinct from traditional codon-based models. First, NICs do not correlate with genomic codon frequency: multiple common codons, such as ACC for Thr and GTC for Val, acted as strong inhibitors when positioned at the 5\u0026prime; end, whereas canonical rare codons, including ATA for Ile, often showed no detectable negative effect. Second, direct quantification of intracellular tRNA abundances demonstrated that NIC-mediated repression cannot be explained by limited aminoacyl-tRNA supply. In several cases, NICs exhibited equal or even higher total decoding capacity than their synonymous high-expression counterparts, contradicting the classical \u0026ldquo;tRNA scarcity\u0026rdquo; hypothesis and suggesting that decoding availability per se is not the limiting factor.\u003c/p\u003e \u003cp\u003eAlthough mRNA secondary structure is widely recognized as a global determinant of translation efficiency, consistent with prior large-scale studies[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], our rescue experiments revealed that replacing only a small number of NICs at the N terminus could increase protein expression by more than sevenfold, with minimal changes in local folding free energy. This decoupling between expression recovery and mRNA structural stability indicates that NICs do not act primarily by occluding ribosome access through stable RNA structures. Instead, their inhibitory effect appears to arise downstream of ribosome binding, at a functional stage of translation initiation or early elongation that is not captured by static structure-based metrics.\u003c/p\u003e \u003cp\u003eCross-species validation further revealed the mechanistic depth of the NIC phenomenon. In \u003cem\u003eRhodococcus ruber\u003c/em\u003e, a high-GC actinobacterium, the NIC repertoire differed almost completely from that of \u003cem\u003eE. coli\u003c/em\u003e. Several strong NICs in \u003cem\u003eR. ruber\u003c/em\u003e, including Ser-AGT, Asp-GAT, and Asn-AAT, were A/T-ending codons, in contrast to the GC3-dominated NICs observed in \u003cem\u003eE. coli\u003c/em\u003e. This inversion provides a coherent explanation for the opposite correlations between 5\u0026prime;-end mRNA structural stability and expression observed in the two hosts, but cannot be rationalized by RNA folding alone. We propose that these host-specific NIC profiles reflect long-term co-evolution between each organism\u0026rsquo;s translational machinery\u0026mdash;including ribosomal RNA, ribosomal proteins, and initiation and elongation factors\u0026mdash;and its genomic codon usage patterns. In a given host, certain codons may impose intrinsic kinetic or conformational barriers during the earliest decoding events, thereby reducing the probability that ribosomes successfully transition from initiation into processive elongation. Because translation initiation is the dominant rate-limiting step in bacterial protein synthesis, even modest increases in abortive initiation or early dissociation events can be strongly amplified, creating a decisive bottleneck that downstream sequence optimization cannot overcome. Polysome profiling provided direct molecular evidence for this model, showing that mRNAs containing NICs accumulate predominantly in initiation or monosome-associated fractions and fail to engage efficiently in polysome elongation, whereas NIC-repaired transcripts readily enter productive translation.\u003c/p\u003e \u003cp\u003eAt a deeper mechanistic level, our observations are also consistent with previous studies showing that the bacterial 30S subunit recognizes mRNA primarily through 16S rRNA and the ribosomal protein S1, both of which play central roles in mRNA recruitment, positioning, and early decoding. Protein S1, in particular, has been implicated in preferential binding to single-stranded, A/U-rich regions and in facilitating the loading of structured or weakly accessible mRNAs onto the ribosome. Importantly, both the sequence composition of 16S rRNA and the abundance, structure, and binding preferences of S1 vary substantially across bacterial species.\u003c/p\u003e \u003cp\u003eIn this context, the host-specific inversion of N-terminal GC/AT preferences observed here may reflect intrinsic differences in how distinct ribosomes engage with nascent coding sequences immediately downstream of the start codon. Rather than acting solely through global mRNA folding energetics, early synonymous codons may modulate ribosome recruitment, stabilization, or conformational transitions during the formation of a productive initiation complex in a host-dependent manner. This perspective provides a plausible molecular basis for why the same 5\u0026prime;-end sequence features can exert opposite effects on expression in different organisms[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eBuilding on these mechanistic insights, we propose a two-section design strategy in which the N-terminal 48 bp is optimized using an N-terminal Optimal Codon (NOC) chart that excludes both NICs and core-region incompatible codons, while the downstream region is designed more conservatively using codons enriched in highly expressed host proteins. Validation across a diverse panel of refractory proteins demonstrated that this criterion consistently rescued previously non-expressing targets and achieved order-of-magnitude improvements over conventional codon-optimization approaches, in several cases requiring replacement of only the core incompatible codons within the first 18 bp. Together, these results highlight the disproportionate regulatory impact of a minimal N-terminal sequence window and establish a mechanism-grounded codon-usage rule for heterologous protein overexpression.\u003c/p\u003e \u003cp\u003eDespite these advances, the precise molecular mechanism of NICs remains to be resolved. Although polysome profiling located the incompatible at initiation and early elongation, the specific codon\u0026ndash;ribosome interactions underlying this barrier remain a \u0026ldquo;black box.\u0026rdquo; Future studies using in vitro translation systems and cryo-EM could capture transient ribosome\u0026ndash;mRNA structures at atomic resolution and reveal the conformational states that cause stalling[\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. Beyond mechanistic resolution, the broader significance of the NIC phenomenon lies in its implications for the rapidly expanding field of sequence-designed proteins. Advances in rational protein engineering, directed evolution, and AI-assisted sequence generation are dramatically accelerating the creation of novel enzymes and functional proteins. However, the practical realization of these designs remains fundamentally constrained by heterologous expression failure, which often arises independently of protein folding or catalytic potential. Our findings suggest that for a substantial fraction of designed sequences, expression fate is already determined within a minimal N-terminal coding window, long before downstream sequence features can exert any effect.\u003c/p\u003e \u003cp\u003eIn this context, NICs provide a mechanistic explanation for why many otherwise promising designs fail at the first experimental hurdle. Rather than being an incidental nuisance, non-expression emerges as a predictable, sequence-encoded outcome governed by host-specific decoding constraints at the earliest stages of translation. Recognizing and eliminating NICs therefore shifts heterologous expression from an empirical, trial-and-error process toward a rule-based and mechanistically grounded design problem.\u003c/p\u003e \u003cp\u003eLooking forward, extending NIC screening to a broader range of prokaryotic and eukaryotic hosts\u0026mdash;including yeast and mammalian systems\u0026mdash;will test the generality of this principle and may uncover additional layers of host-dependent translational regulation. More broadly, our work reframes heterologous protein expression as a problem of early translational decision-making, in which a small number of synonymous codons at the N terminus exert a disproportionate regulatory influence over expression fate. By identifying and mechanistically contextualizing these incompatible codons, this study establishes a new conceptual and practical framework for bridging the gap between sequence design and functional protein production\u0026mdash;an essential step for the next generation of enzyme engineering, synthetic biology, and AI-driven protein discovery.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn this study, we systematically identified and validated a previously unrecognized class of translational regulatory elements\u0026mdash;N-terminal-specific Incompatible Codons (NICs)\u0026mdash;that decisively determine the expression fate of heterologous proteins in prokaryotic hosts. Through large-scale synonymous library screening, ribosome-level analyses, and cross-host validation, we demonstrate that the inhibitory effects of NICs are not explained by codon rarity, tRNA supply, or mRNA secondary structure. Instead, NICs act by destabilizing ribosome loading and the transition from translation initiation and early elongation to efficient translation, thereby creating a stringent kinetic barrier that prevents productive translation.\u003c/p\u003e \u003cp\u003eBy integrating NIC analysis with proteome-guided codon-usage mining, we further distilled these mechanistic insights into a new codon-usage criterion for heterologous protein overexpression. This criterion emphasizes the functional compartmentalization of the translation initiation region, identifying a minimal core window within the first 18 bp that is under extreme negative selection in highly expressed genes, and defining an expanded N-terminal design region of 48 bp governed by host-specific optimal codon usage. Rational remodeling of this core region at single-codon resolution was sufficient to rescue multiple previously non-expressing proteins and to achieve order-of-magnitude improvements over conventional codon-optimization strategies.\u003c/p\u003e \u003cp\u003eCollectively, this work reveals a previously hidden layer of sequence-encoded translational control that operates upstream of classical elongation-based models and establishes a mechanism-grounded design dogma for gene optimization in prokaryotes. By shifting the focus from global codon bias to decisive N-terminal bottlenecks, our findings provide both conceptual insight into translation regulation and a broadly applicable framework for overcoming long-standing barriers in heterologous protein production.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by the National Key R\u0026amp;D Program of China (2025YFA0921800), National Natural Science Foundation of China (No. U24A6011), Beijing Municipal Science \u0026amp; Technology Commission No. Z231100003723015 and 202504841063.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRuizhao Jiang:\u003c/strong\u003e Writing–original draft, Methodology, Investigation, Data curation, Conceptualization. \u003cstrong\u003eYukun Zheng\u003c/strong\u003e: Methodology, Data curation, Conceptualization. \u003cstrong\u003eYuwen Wei:\u0026nbsp;\u003c/strong\u003eMethodology, Data curation, Conceptualization. \u003cstrong\u003eMingye Song:\u0026nbsp;\u003c/strong\u003eFigure illustration, Data curation. \u003cstrong\u003eShunshun Yang:\u0026nbsp;\u003c/strong\u003eFigure illustration, Data curation. \u003cstrong\u003eKe Wang:\u0026nbsp;\u003c/strong\u003eMethodology, Data curation. \u003cstrong\u003eYilong Zhou\u003c/strong\u003e:\u0026nbsp;Methodology, Data curation, Validation. \u003cstrong\u003eYan Du:\u0026nbsp;\u003c/strong\u003eMethodology, Data curation. \u003cstrong\u003eFulong Li:\u0026nbsp;\u003c/strong\u003eMethodology. \u003cstrong\u003eYouxiang Liang:\u0026nbsp;\u003c/strong\u003eMethodology, Data curation. \u003cstrong\u003eMiaomiao Wang:\u0026nbsp;\u003c/strong\u003eMethodology, Data curation. \u003cstrong\u003eHuimin Yu:\u003c/strong\u003e Writing, review \u0026amp; editing, Supervision, Funding acquisition, Conceptualization.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDeclaration of competing interest\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eDe Marco, A., \u003cem\u003eRecent advances in recombinant production of soluble proteins in E. coli.\u003c/em\u003e Microbial cell factories, 2025. \u003cstrong\u003e24\u003c/strong\u003e(1): p. 21.\u003c/li\u003e\n\u003cli\u003eFrance, S.P., R.D. Lewis, and C.A. Martinez, \u003cem\u003eThe evolving nature of biocatalysis in pharmaceutical research and development.\u003c/em\u003e JACS Au, 2023. \u003cstrong\u003e3\u003c/strong\u003e(3): p. 715-735.\u003c/li\u003e\n\u003cli\u003eKim, D., et al., \u003cem\u003eComputational design of metallohydrolases.\u003c/em\u003e Nature, 2025: p. 1-8.\u003c/li\u003e\n\u003cli\u003eHuang, P.-S., S.E. Boyken, and D. Baker, \u003cem\u003eThe coming of age of de novo protein design.\u003c/em\u003e Nature, 2016. \u003cstrong\u003e537\u003c/strong\u003e(7620): p. 320-327.\u003c/li\u003e\n\u003cli\u003eChen, Y., et al., \u003cem\u003eGlycolysis-compatible urethanases for polyurethane recycling.\u003c/em\u003e Science, 2025. \u003cstrong\u003e390\u003c/strong\u003e(6772): p. 503-509.\u003c/li\u003e\n\u003cli\u003eLozano Terol, G., et al., \u003cem\u003eImpact of the expression system on recombinant protein production in Escherichia coli BL21.\u003c/em\u003e Frontiers in microbiology, 2021. \u003cstrong\u003e12\u003c/strong\u003e: p. 682001.\u003c/li\u003e\n\u003cli\u003eKondo, T. and S. Yumura, \u003cem\u003eStrategies for enhancing gene expression in Escherichia coli.\u003c/em\u003e Applied Microbiology and Biotechnology, 2020. \u003cstrong\u003e104\u003c/strong\u003e(9): p. 3825-3834.\u003c/li\u003e\n\u003cli\u003ePrice, W.N., et al., \u003cem\u003eLarge-scale experimental studies show unexpected amino acid effects on protein expression and solubility in vivo in E. coli.\u003c/em\u003e Microbial Informatics and Experimentation, 2011. \u003cstrong\u003e1\u003c/strong\u003e: p. 1-20.\u003c/li\u003e\n\u003cli\u003eBo\u0026euml;l, G., et al., \u003cem\u003eCodon influence on protein expression in E. coli correlates with mRNA levels.\u003c/em\u003e Nature, 2016. \u003cstrong\u003e529\u003c/strong\u003e(7586): p. 358-363.\u003c/li\u003e\n\u003cli\u003eSchmidt, M., et al., \u003cem\u003eMaximizing Heterologous Expression of Engineered Type I Polyketide Synthases: Investigating Codon Optimization Strategies.\u003c/em\u003e ACS Synthetic Biology, 2023.\u003c/li\u003e\n\u003cli\u003eFusco, F., et al., \u003cem\u003eInfluence of the mRNA initial region on protein production: a case study using recombinant detoxified pneumolysin as a model.\u003c/em\u003e Frontiers in Bioengineering and Biotechnology, 2024. \u003cstrong\u003e11\u003c/strong\u003e: p. 1304965.\u003c/li\u003e\n\u003cli\u003eTolia, N.H. and L. Joshua-Tor, \u003cem\u003eStrategies for protein coexpression in Escherichia coli.\u003c/em\u003e Nature methods, 2006. \u003cstrong\u003e3\u003c/strong\u003e(1): p. 55-64.\u003c/li\u003e\n\u003cli\u003eOjima-Kato, T., S. Nagai, and H. Nakano, \u003cem\u003eN-terminal SKIK peptide tag markedly improves expression of difficult-to-express proteins in Escherichia coli and Saccharomyces cerevisiae.\u003c/em\u003e Journal of bioscience and bioengineering, 2017. \u003cstrong\u003e123\u003c/strong\u003e(5): p. 540-546.\u003c/li\u003e\n\u003cli\u003eJiang, R., et al., \u003cem\u003eStrategies to overcome the challenges of low or no expression of heterologous proteins in Escherichia coli.\u003c/em\u003e Biotechnology Advances, 2024. \u003cstrong\u003e75\u003c/strong\u003e: p. 108417.\u003c/li\u003e\n\u003cli\u003eGustafsson, C., S. Govindarajan, and J. Minshull, \u003cem\u003eCodon bias and heterologous protein expression.\u003c/em\u003e Trends in biotechnology, 2004. \u003cstrong\u003e22\u003c/strong\u003e(7): p. 346-353.\u003c/li\u003e\n\u003cli\u003eIkemura, T., \u003cem\u003eCodon usage and tRNA content in unicellular and multicellular organisms.\u003c/em\u003e Molecular biology and evolution, 1985. \u003cstrong\u003e2\u003c/strong\u003e(1): p. 13-34.\u003c/li\u003e\n\u003cli\u003eQuax, T.E., et al., \u003cem\u003eCodon bias as a means to fine-tune gene expression.\u003c/em\u003e Molecular cell, 2015. \u003cstrong\u003e59\u003c/strong\u003e(2): p. 149-161.\u003c/li\u003e\n\u003cli\u003eBhandari, B.K., et al., \u003cem\u003eAnalysis of 11,430 recombinant protein production experiments reveals that protein yield is tunable by synonymous codon changes of translation initiation sites.\u003c/em\u003e PLoS computational biology, 2021. \u003cstrong\u003e17\u003c/strong\u003e(10): p. e1009461.\u003c/li\u003e\n\u003cli\u003eGualerzi, C.O. and C.L. Pon, \u003cem\u003eInitiation of mRNA translation in bacteria: structural and dynamic aspects.\u003c/em\u003e Cellular and Molecular Life Sciences, 2015. \u003cstrong\u003e72\u003c/strong\u003e(22): p. 4341-4367.\u003c/li\u003e\n\u003cli\u003eMcCarthy, J.E. and C. Gualerzi, \u003cem\u003eTranslational control of prokaryotic gene expression.\u003c/em\u003e Trends in Genetics, 1990. \u003cstrong\u003e6\u003c/strong\u003e: p. 78-85.\u003c/li\u003e\n\u003cli\u003eKim, D.J., et al., \u003cem\u003eDeepTESR: a deep learning framework to predict the degree of translational elongation short ramp for gene expression control.\u003c/em\u003e ACS Synthetic Biology, 2022. \u003cstrong\u003e11\u003c/strong\u003e(5): p. 1719-1726.\u003c/li\u003e\n\u003cli\u003eNieuwkoop, T., et al., \u003cem\u003eRevealing determinants of translation efficiency via whole-gene codon randomization and machine learning.\u003c/em\u003e Nucleic acids research, 2023. \u003cstrong\u003e51\u003c/strong\u003e(5): p. 2363-2376.\u003c/li\u003e\n\u003cli\u003eDuval, M., et al., \u003cem\u003eEscherichia coli ribosomal protein S1 unfolds structured mRNAs onto the ribosome for active translation initiation.\u003c/em\u003e PLoS biology, 2013. \u003cstrong\u003e11\u003c/strong\u003e(12): p. e1001731.\u003c/li\u003e\n\u003cli\u003eTuller, T. and H. Zur, \u003cem\u003eMultiple roles of the coding sequence 5\u0026prime; end in gene expression regulation.\u003c/em\u003e Nucleic acids research, 2015. \u003cstrong\u003e43\u003c/strong\u003e(1): p. 13-28.\u003c/li\u003e\n\u003cli\u003eKudla, G., et al., \u003cem\u003eCoding-sequence determinants of gene expression in Escherichia coli.\u003c/em\u003e science, 2009. \u003cstrong\u003e324\u003c/strong\u003e(5924): p. 255-258.\u003c/li\u003e\n\u003cli\u003eGoodman, D.B., G.M. Church, and S. Kosuri, \u003cem\u003eCauses and effects of N-terminal codon bias in bacterial genes.\u003c/em\u003e Science, 2013. \u003cstrong\u003e342\u003c/strong\u003e(6157): p. 475-479.\u003c/li\u003e\n\u003cli\u003eGriswold, K.E., et al., \u003cem\u003eEffects of codon usage versus putative 5\u0026prime;-mRNA structure on the expression of Fusarium solani cutinase in the Escherichia coli cytoplasm.\u003c/em\u003e Protein expression and purification, 2003. \u003cstrong\u003e27\u003c/strong\u003e(1): p. 134-142.\u003c/li\u003e\n\u003cli\u003eEyre-Walker, A. and M. Bulmer, \u003cem\u003eReduced synonymous substitution rate at the start of enterobacterial genes.\u003c/em\u003e Nucleic acids research, 1993. \u003cstrong\u003e21\u003c/strong\u003e(19): p. 4599-4603.\u003c/li\u003e\n\u003cli\u003eTuller, T., et al., \u003cem\u003eAn evolutionarily conserved mechanism for controlling the efficiency of protein translation.\u003c/em\u003e Cell, 2010. \u003cstrong\u003e141\u003c/strong\u003e(2): p. 344-354.\u003c/li\u003e\n\u003cli\u003eShah, P., et al., \u003cem\u003eRate-limiting steps in yeast protein translation.\u003c/em\u003e Cell, 2013. \u003cstrong\u003e153\u003c/strong\u003e(7): p. 1589-1601.\u003c/li\u003e\n\u003cli\u003eBentele, K., et al., \u003cem\u003eEfficient translation initiation dictates codon usage at gene start.\u003c/em\u003e Molecular systems biology, 2013. \u003cstrong\u003e9\u003c/strong\u003e(1): p. 675.\u003c/li\u003e\n\u003cli\u003eZheng, G., et al., \u003cem\u003eEfficient and quantitative high-throughput tRNA sequencing.\u003c/em\u003e Nature methods, 2015. \u003cstrong\u003e12\u003c/strong\u003e(9): p. 835-837.\u003c/li\u003e\n\u003cli\u003eCarbone, A., A. Zinovyev, and F. K\u0026eacute;pes, \u003cem\u003eCodon adaptation index as a measure of dominating codon bias.\u003c/em\u003e Bioinformatics, 2003. \u003cstrong\u003e19\u003c/strong\u003e(16): p. 2005-2015.\u003c/li\u003e\n\u003cli\u003eGonzalez de Valdivia, E.I. and L.A. Isaksson, \u003cem\u003eA codon window in mRNA downstream of the initiation codon where NGG codons give strongly reduced gene expression in Escherichia coli.\u003c/em\u003e Nucleic acids research, 2004. \u003cstrong\u003e32\u003c/strong\u003e(17): p. 5198-5205.\u003c/li\u003e\n\u003cli\u003eCastillo-M\u0026eacute;ndez, M., et al., \u003cem\u003eAdenine-containing codons enhance protein synthesis by promoting mRNA binding to ribosomal 30S subunits provided that specific tRNAs are not exhausted.\u003c/em\u003e Biochimie, 2012. \u003cstrong\u003e94\u003c/strong\u003e(3): p. 662-672.\u003c/li\u003e\n\u003cli\u003eEtchegaray, J.-P. and M. Inouye, \u003cem\u003eTranslational enhancement by an element downstream of the initiation codon in Escherichia coli.\u003c/em\u003e Journal of Biological Chemistry, 1999. \u003cstrong\u003e274\u003c/strong\u003e(15): p. 10079-10085.\u003c/li\u003e\n\u003cli\u003eVimberg, V., et al., \u003cem\u003eTranslation initiation region sequence preferences in Escherichia coli.\u003c/em\u003e BMC molecular biology, 2007. \u003cstrong\u003e8\u003c/strong\u003e(1): p. 100.\u003c/li\u003e\n\u003cli\u003eWang, C., et al., \u003cem\u003eStructural basis of transcription-translation coupling.\u003c/em\u003e Science, 2020. \u003cstrong\u003e369\u003c/strong\u003e(6509): p. 1359-1365.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-8581843/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8581843/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eHeterologous expression of proteins and enzymes in prokaryotic hosts such as \u003cem\u003eEscherichia coli\u003c/em\u003e and \u003cem\u003eRhodococcus ruber\u003c/em\u003e is central to biomanufacturing and biotechnology, yet a substantial fraction of proteins still fail to express for reasons that remain unknown. To address this long-standing problem, we focused on translation initiation, particularly the N-terminal coding-start region. By constructing large-scale 5\u0026prime;-synonymous codon libraries, we found that mRNA secondary structure alone cannot account for expression-fate of proteins across hosts. Instead, a previously unrecognized factor, designated as N-terminal-specified Incompatible Codons (NICs), were identified that decisively correlate protein-non-expression. Library mining revealed all of the NICs in \u003cem\u003eE. coli\u003c/em\u003e, spanning 13 of the 20 amino acids, such as GTC for Val, AAG for Lys, ACC for Thr and CAG for Gln. Further studies showed that these NICs are independent of codon rarity, tRNA abundance, and mRNA folding, but correlated to the wobble-base usages. Ribosome loading experiments demonstrated that NICs act as kinetic barriers that prevent ribosome loading thereby resulting in failed translation from initiation and early elongation to efficient elongation. We further performed deep DIA-based LC\u0026ndash;MS/MS analysis on intracellular soluble proteins and quantified protein abundance, obtaining a quantitative snapshot of the endogenous proteome. The top 500 most abundant proteins in \u003cem\u003eE. coli\u003c/em\u003e were selected and the N-terminal optimal codons (NOC) were summarized. By integrating both NIC avoidance and NOC preference, we designed a new two-section codon-usage strategy for heterologous protein overexpression in prokaryotes, including a one-to-one (one-amino-acid, one-codon) codon table for 48 bp N-terminus and a mixed table (optimal codon utilization\u0026thinsp;+\u0026thinsp;rare codon substitution) for the subsequent sequence. Using this novel approach, 8 previously non-expressed proteins such as lipase, laccase and cysteine hydrolase, were reversely overexpressed in \u003cem\u003eE. coli\u003c/em\u003e. An on-line codon-design tool \u003cb\u003eRiboLoad Codon Optimizer\u003c/b\u003e, has been available \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://47.86.169.8/\u003c/span\u003e\u003cspan address=\"http://47.86.169.8/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, serving for numerous proteins/enzymes overexpression with \u003cem\u003eE. coli\u003c/em\u003e and \u003cem\u003eR. ruber\u003c/em\u003e hosts. Together, these findings establish a mechanism-grounded codon-usage framework for overcoming translational bottlenecks and enabling efficient heterologous protein expression in prokaryotic hosts.\u003c/p\u003e","manuscriptTitle":"N terminal wobble base usage determines ribosome loading and thus protein expression fate","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-30 07:05:14","doi":"10.21203/rs.3.rs-8581843/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"nature-communications","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"NCOMMS","sideBox":"Learn more about [Nature Communications](http://www.nature.com/ncomms/)","snPcode":"","submissionUrl":"https://mts-ncomms.nature.com/","title":"Nature Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature Communications","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"63772723-b5a9-4d4d-a0a3-3ea7c19767e9","owner":[],"postedDate":"March 30th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":64348562,"name":"Biological sciences/Molecular biology/Translation"},{"id":64348563,"name":"Biological sciences/Biotechnology/Expression systems"}],"tags":[],"updatedAt":"2026-03-31T11:26:59+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-30 07:05:14","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8581843","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8581843","identity":"rs-8581843","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}