Translation attenuation via 3′ terminal codon usage in bovine csn1s2 is responsible for the difference in α_s2- and β-casein profile in milk

Abstract

The rate of secretion of α_s2-casein into bovine milk is approximately 25% of that of β-casein, yet mammary expression of their respective mRNA transcripts (csn1s2 and csn2) is not different. Our objective was to identify molecular mechanisms that explain the difference in translation efficiency between csn1s2 and csn2. Cell-free translational efficiency of csn2 was 5 times that of csn1s2. Transcripts of csn1s2 distributed into heavier polysomes than csn2 transcripts, indicating an attenuation of elongation and/or termination. Stimulatory and inhibitory effects of the 5′ and 3′ UTRs on translational efficiency were different with luciferase and casein sequences in the coding regions. Substituting the 5′ and 3′ UTRs from csn2 into csn1s2 did not improve csn1s2 translation, implicating the coding region itself in the translation difference. Deletion of a 28-codon fragment from the 3′ terminus of the csn1s2 coding region, which displays codons with low correlations to cell fitness, increased translation to a par with csn2. We conclude that the usage of the last 28 codons of csn1s2 is the main regulatory element that attenuates its expression and is responsible for the differential translational expression of csn1s2 and csn2.

Abbreviations

40S

small ribosomal subunit

60S

large ribosomal subunit

AA

amino acid

aaRS

aminoacyl-tRNA synthetase

aa-tRNA

aminoacyl-tRNA

Apaf-1

apoptosis protease activating factor 1

ARE

AU-rich element

cDNA

complementary DNA

DLG1

disc large 1 ncosuppressor

eEF

eukaryotic elongation factor

eIF

eukaryotic initiation factor

eRF

eukaryotic termination factor

FMR1

fragile X mental retardation 1

HRP

horseradish eroxidase

IE

inhibitory element

IRE

iron-responsive element

IRES

nternal ribosome entry site

IRP

iron-regulatory protein

MACT

bovine mammary epithelial cell

m⁷G

7-methylated uanidine

mRNA

messenger RNA

PABP

poly(A) binding protein

PAGE

polyacrylamide gel electrophoresis

PCR

polymerase chain reaction

PVDF

polyvinylidene fluoride

qPCR

real-time polymerase chain reaction

RACE

rapid amplification of cDNA ends

RBP

RNA-binding protein

RRL

rabbit reticulocyte lysate

RT

reverse transcription

sAUG

start codon

SDS

sodium dodecyl sulfate

SE

standard error

STR

single-stranded nucleic acid binding protein

TBS-T

Tris-buffered saline containing 0.5% Tween 20

TfR

transferrin receptor

tRNA

transfer RNA

uAUG

upstream start codon

uORF

open reading frame

UTR

untranslated region

ΔG

free energy

Introduction

The proteins secreted into bovine milk by exocytosis from mammary epithelial cells supply an ideal amino acid profile for infant growth.¹ In addition, bioactive peptides derived from fermentation or digestion of milk proteins possess functional value in treatment and prevention of metabolic syndromes such as obesity, atherogenesis, arterial hypertension and type 2 diabetes.^1-4 The bovine milk proteome is 82% casein and 18% whey proteins.⁵ The caseins are amphipathic and aggregate into surfactant particles called casein micelles in colloidal suspension with Ca₂PO₄ in milk. On consumption, the micelles facilitate protein, Ca and PO₄ digestion in the developing neonatal digestive tract and they have drawn attention in recent years due to their unique properties as novel gastrointestinal nutrient and drug delivery nanoparticles.^6,7 Therefore, understanding the factors that regulate casein synthesis is important for the endeavor to make high quality protein available to the global population for both nutritional and functional uses.

The concentrations in milk of the 4 casein variants, αs₁-, αs₂-, β-, and κ-casein, do not match the mammary expression of their respective mRNA transcripts, csn1s1, csn1s2, csn2, and csn3,^8,9 suggesting that each variant is not translated with the same efficiency. According to mRNA:protein ratios, csn2 and csn1s2 exhibit the highest and lowest translational efficiencies, respectively, with a 4-fold difference.⁸ Both β- and α_s2-casein are calcium-sensitive caseins, of the secretory calcium-binding phosphoprotein gene family, that arose from a common ancestral sequence by multiple duplications and amino acid replacements while retaining similar 5′ untranslated regions (UTRs) of the mRNA, signal sequences and phosphorylation sites.¹⁰ β-casein is present in all mammalian milks, while the proportion of α_s2-casein in milk varies considerably between species and is absent from human and marsupial milk.

Translational differences between species of mRNA in the same cell are due to cis- and/or trans-elements such as GC content, secondary structure, 5′ and 3′ UTRs, codon usage efficiencies, and protein-binding sites that influence ribosome recruitment and transit. Such features of mRNA are known to produce differential expression of regulatory proteins such as p53,¹¹ interleukin-2¹² and gastrin.¹³ As secretory, but not regulatory proteins, the caseins represent a novel example of regulation where the variation in ease of mRNA translation has been exploited evolutionarily to modify the profile of proteins secreted from a cell.

The specific aims of this study were to characterize sequences of csn2 and csn1s2 and to define the molecular features that account for differences in translational efficiency and protein accumulation. Reticulocyte lysate (RRL) was used for in vitro translation because of its low background and efficient translation of exogenous mRNA at low concentrations. Using this system, in which csn2 was translated 5-times faster than csn1s2, we undertook a comprehensive study of effects of primary and secondary structures of csn2 and csn1s2 coding regions and UTRs on translational efficiencies. We found that the last 28 codons upstream of the stop codon were responsible for the relatively low translational efficiency of csn1s2, specifically those from 195 to 207 of the 222-codon transcript. We propose that codon usage bias contributes to maintenance of the profile of caseins in milk.

Results

In vitro transcription rates are similar but β-casein accumulates to a greater extent than α_s2-casein

Length, GC content and ΔG of the 5′ UTRs, 3′ UTRs and complete sequences of csn2 (Submitted to NCBI: KC993858.1) and csn1s2 (NCBI: NM_174528.2) that were obtained from RACE PCR (Supplementary Fig. S1) and the NCBI database, respectively, were compared (Supplementary Table S2). No distinct differences were found according to the in silico characteristics of sequences. Both mRNAs lack upstream start codons and open reading frames in the 5′ UTR. The preferred nucleotide context for detection of the AUG start codon by the scanning ribosome in bovine transcripts is GCCAACAUGGCG.¹⁴ The csn1s2 start codon context is GCAAACAUGAAG, for which 5 upstream nucleotides out of 6 match the preferred sequence; the A at −4 is the only exception. In contrast, the csn2 context is AGAGCCAUGAAG, which contains the less common A, G, G and C in positions −6, −5, −3 and −2, respectively. This finding does not implicate a suboptimal context for the start codon in csn1s2 as an impediment to initiation of translation. In addition, both csn2 and csn1s2 share the same AAG sequence following the start codon and a G following the stop codon. The base following the stop codon is known to influence translational termination efficiency.^15,16 Both transcripts lack early termination codons. Termination codons within the coding region are known to trigger a nonsense mediated decay (NMD) process of the transcript.¹⁷ Thus, these basic sequence characteristics cannot explain the difference in translational efficiency between csn2 and csn1s2.

Two sets of primers (Supplementary Table S1) were designed to clone CSN2 and CSN1S2 cDNA into a pTZ19R vector which possesses a T7 promoter. Each EcoRI-linearized construct was transcribed and translated to produce csn2 and csn1s2 (Fig. 1A) and protein (Fig. 1B) with expected molecular sizes of 1095 nt (24 kDa) and 1023 nt (25.2 kDa), respectively. The transcriptional efficiency was not different between CSN2 and CSN1S2 but protein accumulation was higher for β-casein than α_s2-casein. When the number of positions available for biotinylated lysine incorporation into the proteins was taken into account - β-casein contains 12 lysine residues (5.33 molar%) and α_s2-casein has 25 (11.21 molar%) - product formation was approximately 6-fold higher from csn2. Both transcripts produced maximum protein at 20 min of translation time (Fig. 1B).

Figure 1.

In vitro transcription and translation of csn2 and csn1s2. (A) Electrophoresis of transcribed full length csn2 and csn1s2 on 1.2% RNA agarose gel containing 0.6 M formaldehyde. Expected product sizes are 1095 nt and 1023 nt, csn2 and csn1s2, respectively. (B) Electroblotting of translated csn2 (24kDa) and csn1s2 (25.2kDa). Equal amounts of (A) were incubated for 15 min in rabbit reticulocyte lysate system with biotinylated lysine tRNA. The amount of biotin labeled proteins synthesized were blotted against Streptavidin-HRP. Quantification of the total protein stained with fast green on the blot were used for normalization. Representative bands from electroblots are also shown in (B). Values are means ± SE. Means without a common letter differ at P ≤ 0.05. M: a size marker, nt: nucleotides

Efficient translation in RRL does not require a 7-methyl GTP (m⁷G) cap on the 5′ end of the mRNA¹⁸ but initiation of translation is typically accelerated by interaction of eIF4E with the m⁷G cap.¹⁹ To determine whether the difference in protein accumulation between csn2 and csn1s2 is cap-dependent, both mRNAs, m⁷G capped and uncapped, were translated for 15, 30, 45 and 60 min. The addition of a 5′ cap increased protein accumulation by 36% for both transcripts but did not alter the 6-fold difference between csn1s2 and csn2 (Supplementary Fig. S2). The use of uncapped mRNAs hereon retains the focus on factors involved in translation differences between the 2 transcripts.

csn2 is translated 5 times more efficiently than csn1s2

When RRL is used to assay translational efficiency, mRNA and protein degradation influence protein accumulation independent of mRNA translation. To account for these effects and estimate translational efficiencies, the compartmental model of Hargrove and Schmidt,²⁰ with modifications, was fit to mRNA and protein accumulation time-courses over 60 and 90 min of incubation in RRL. The model (Fig. 2A) is composed of an RNA pool (R) and a protein pool (P). R and P are degraded by first-order kinetics with rate constants k_d1 and k_d2, respectively; P is synthesized from R according to translational efficiency k_s; k_s decreases by first-order kinetics with rate constant k_d3. Timecourses of R and P were fitted by the model with r²-values between 0.88 and 0.99 (Fig. 2C and D). Fitted values of k_d1 (Fig. 2C) show that csn1s2 degraded 2.1-fold faster than csn2 with a half-life (ln2/k_d1) of 38 min compared to 82 min for csn2. There was no difference in the rate of protein degradation (k_d2) between β- and αs2-casein, where half-lives were 4 min. According to k_d3 estimates, translational activity declined 37% faster in the presence of csn1s2 compared to csn2. In addition to a more stable mRNA, efficiency of csn2 translation, k_s, was 5 times higher than that of csn1s2 translation (Fig. 2B).

Figure 2.

Translational efficiency with mRNA and protein degradation rates taken into account. (A) Two-compartment mathematical model. Boxes indicate mRNA (R) and protein (P) concentrations; arrows represent fluxes with rate constants for mRNA degradation (k_d1), protein degradation (k_d2), translation efficiency (k_s) and translation capacity (k_d3). (B) Rate constant values best fit to observations of mRNA and protein accumulation in translation reaction system. Efficiency of csn2 translation (k_s) is 5 times higher than that of csn1s2. (C) mRNA and (D) protein accumulation over time; points indicate observations and lines indicate fitted model predictions.

Elongation is suppressed more than initiation in csn1s2

To determine whether the difference in translational efficiency is due solely to initiation, translation incubations were ultracentrifuged through a sucrose gradient to identify polysomal distribution of csn2 and csn1s2. Gradients were collected into 15 fractions from top (free ribosomes, fraction1) to bottom (polyribosomes, fractions 9–15). There was higher expression of csn1s2 on polyribosome fractions whereas more csn2 was found on lighter ribosomes (Supplementary Fig. S3A). Sucrose gradient centrifugaton of bovine mammary tissue homogenates yielded similar results with no significant presence of csn1s2 in the monosome fraction (Supplementary Fig. S3B). An increase in the rate of initiation relative to elongation or termination results in a shift of the mRNA from smaller to larger polysomes, whereas the opposite occurs if the rate of elongation or termination is increased relative to initiation.²¹ The shift of csn1s2 into larger polysomes indicates that elongation and/or termination, relative to initiation, is slower from csn1s2. Since initiation of csn1s2 translation is also slower, as evidenced by the lower translational efficiency, the polysome profile indicates that elongation and/or termination are suppressed more than initiation in csn1s2 compared to csn2.

mRNA secondary structure impedes csn1s2 translation

Stable secondary structures formed in RNA constitute an impediment to ribosome binding and scanning. We wondered if the slower initiation and elongation of α_s2-casein is due to more stable secondary structure, i.e. lower ΔG, of csn1s2 than csn2. Nucleotide sequences of csn2 and csn1s2 were input into mfold and Rfold software^22,23 that predicts optimal secondary structures. Interestingly, the 5′ UTR (Fig. 3D) and full length of csn1s2 (Fig. 3B) were less stable than those of csn2 (Fig. 3A and C), by 0.80 and 17.70 kcal/mol (Supplementary Table S2), respectively, which would be expected to favor ribosome transit and therefore do not account for slower initiation and elongation of csn1s2. In contrast, the structure of the 3′ UTR of csn1s2 (Fig. 3F) was 6.5 kcal/mol more stable than that of csn2 (Fig. 3E; Supplementary Table S2).

Figure 3.

Predicted optimal secondary structures in csn2 and csn1s2 using Mfold and Rfold softwares (14,15) for comparison. Secondary structure of (A and B) full-length, (C and D) 5’ UTRs, and (E and F) 3′ UTRs of csn2 and csn1s2 are presented. Arrow and triangle indicate start and stop codon, respectively. Outlined arrow and bullet shape indicate 5’ and 3′ ends, respectively. Two hairpin structures on 5’ UTR of csn1s2 are circled. (G) Electroblotting of in vitro translated full-length csn2 and csn1s2. Denatured or re-folded transcripts were in vitro translated for 15 or 30 min. (H) Protein accumulation from denatured and re-folded full-length csn2 and csn1s2. Values are means ± SE (n = 3 independent experiments). Means without a common letter differ at P ≤ 0.05 for csn1s2. * Indicates significance at P ≤ 0.05 compare to denatured csn2.

The 5′ UTR of csn1s2 contains 2 hairpin structures close to the start codon (Fig. 3D), which are absent in csn2. In addition the 3′ UTR of csn2 folds away from the coding region whereas the 3′ UTR of csn1s2 interacts entirely with its coding region (Fig. 3F). To test the degree to which secondary structure affects translational efficiency, protein accumulation from denatured and re-folded mRNAs were compared at 15 and 30 min of incubation (Fig. 3G and H). Disrupting secondary structure by denaturing indeed increased translational efficiency and the difference was greater with csn1s2. At 15 min, protein accumulation from denatured csn1s2 was 100% greater than from re-folded csn1s2, compared to only 40% stimulation for csn2. At 30 min, denaturing stimulated protein accumulation by 100% from csn1s2 and had no significant effect on csn2. The results indicate that secondary structure inhibits translation of both csn2 and csn1s2 but it has a greater effect on csn1s2 translation.

The 5′ UTR of csn1s2 and 3′ UTR of csn2 are stimulatory to luciferase reporter activity

There are several cis-acting features of the UTRs of mRNAs expressed from various genes and species.²⁴ Primary and secondary structures of the 5′ UTR upstream of the start codon influence initiation of translation while the 3′ UTR can affect mRNA stability and, to a lesser extent, elongation and/or termination rates. The 5′ UTRs of csn2 and csn1s2 are relatively short, at about 60 nt, with no upstream start codons. The 3′ UTRs are much longer at 300 nt and represent promising targets to mediate post-transcriptional control. An AU-rich element (ARE) present at nucleotide 897 of murine csn2 was found to protect the transcript from degradation via HuR binding,²⁵ whereas the ARE present in caprine csn1s1 was suggested to have a potential role in RNA degradation.²⁶ This ARE is present in the 3′ UTR of bovine csn2 at position 877. In the 3′ UTR of bovine csn1s2, 2 AREs are present at nt 870 and 947 (Supplementary Fig. S1C and D). Modeling the timecourse of protein accumulation indicated that csn1s2 degraded 2.1-fold faster than csn2, which may be related to differences in their respective AREs. However, since translational efficiency is independent of protein and mRNA stability, the cis-elements that affect mRNA stability cannot explain the difference in translational efficiency between csn2 and csn1s2. Furthermore, RNA affinity chromatography of RRL using 5′ and 3′ UTRs of csn2 and csn1s2 revealed no differences in the profile or abundance of proteins bound to the UTRs (Supplementary Fig. S4). We then evaluated the influence of 5′ and 3′ UTRs of casein transcripts on translation by assaying firefly luciferase (FLuc) reporter activity from mRNA constructs transfected into the MAC-T line of bovine mammary epithelial cells (Fig. 4). To normalize for transfection efficiency and cell viability after transfection, we employed renilla luciferase (RLuc) as a reference. We have already shown that, compared to csn2, the 5′ UTR of csn1s2 possesses a more common nucleotide context for the start codon, lower thermal stability and less GC content, while the secondary structure of the 3′ UTR is more stable. Thus, we expected higher reporter activity from constructs containing the 5′ UTR of csn1s2 and the 3′ UTR of csn2.

Figure 4.

The relative Fluc/Rluc ratio following transfection of various constructs into MAC-T cells. (A) Schematic diagram of modified pGL3-control vector with firefly luciferase (FLuc) coding region that was transiently transfected into MAC-T cells. Renilla luciferase expression construct (RLuc) was co-transfected as control. Solid and double lines represent csn2 and csn1s2 sequences, respectively. A box between the various UTR combinations indicates Fluc coding sequence. (B) csn2 or csn1s2 5’UTR effects on Firefly luciferase activity. (C) csn2 or csn1s2 3′UTR effects on Firefly luciferase activity. (D) Different combination of csn2 or csn1s2 UTR effects on Firefly luciferase activity. The relative Fluc/Rluc ratio was calculated by normalizing the Fluc activity by Rluc. Data shown are means ± SEM (n = 3 independent experiments). Means without a common letter differ at P ≤ 0.05.

As expected, we observed 2.1-fold higher FLuc/RLuc activity from the A5L construct carrying the csn1s2 5′ UTR in comparison to the B5L construct with the csn2 5′ UTR (Fig. 4B). The LB3 construct carrying a csn2 3′ UTR showed 2.3-fold higher FLuc/RLuc activity than the LA3 construct with a csn1s2 3′ UTR (Fig. 4C). The results demonstrate beneficial effects of the csn1s2 5′ UTR and the csn2 3′ UTR on translational efficiency.

The ALB construct with both fast UTRs (5′ of csn1s2 and 3′ of csn2) yielded 83% higher FLuc/RLuc activity than the BLA construct carrying the 2 slow UTRs (csn2 5′ UTR and csn1s2 3′ UTR; Fig. 4D). Interestingly, when a 3′ UTR was present, replacement of the slow 5′ UTR with a fast 5′ UTR (ALA vs. BLA, and ALB vs. BLB) had no effect on FLuc/RLuc ratio. On the other hand, the 3′ UTR of csn2 was stimulatory to FLuc/RLuc activity when either the csn2 or csn1s2 5′ UTR was present (ALB vs. ALA, and BLB vs. BLA). These findings suggest that the csn1s2 3′ UTR overrides the stimulatory effect of the csn1s2 5′ UTR and the csn2 3′ UTR overrides the inhibitory effect of the csn2 5′ UTR. With luciferase in the coding region, the 3′ UTRs of the casein mRNAs exert stronger effects on translational efficiency than the 5′ UTRs.

The 5′ UTR of csn1s2 stimulates translation of csn2 and csn1s2 with or without 3’ UTRs

Untranslated regions can influence translational efficiency by forming secondary structures with other parts of the mRNA molecule. To determine if stimulatory effects of the csn1s2 5′ UTR and the csn2 3′ UTR occur in their native context, various constructs were designed around coding regions of csn2 and csn1s2 and translated (Fig. 5A and C).

Figure 5.

Schematic diagram of various constructs with csn2 or csn1s2 as a coding region and corresponding expression level in rabbit reticulocyte lysate (RRL) translation system. (A and C) Solid and double lines represent the csn2 and csn1s2 sequences, respectively. Vertical solid and dotted lines indicate start (AUG) and stop (TAA) codons, respectively. Boxed T7 indicates T7 promoter on the plasmid. (B and D) Relative band intensity of translated constructs represents translational efficiency. Equal amounts of transcribed RNA of various constructs were translated as described in Material and Method in in vitro translation section. In (B), constructs with csn1s2 5’ UTR alone and with csn2 3′ UTR have significant higher expression levels whereas the construct with both csn1s2 5’ and 3′ UTR significantly reduced expression of β-casein. In (D), on the other hand, constructs with csn1s2 5′ UTR have significantly higher expression levels and constructs without any UTRs or with csn1s2 3′ UTR show the least expression of α_s2-casein. Data shown are means ± SE (n = 3 independent experiments). Means without a common letter differ at P ≤ 0.05

Protein accumulation results are presented in Fig.s 5B and D. Consistent with the result we obtained from transfection of luciferase constructs, the csn1s2 5′ UTR was stimulatory to translation of the csn2 coding region, compared to the csn2 5′ UTR, when no 3′ UTR was present (AB0 vs. BB0). It continued to be stimulatory when the 3′ UTR of csn2 was present (ABB vs. BBB) and it also stimulated csn1s2 translation with either 3′ UTR (AAA vs. BAA, and AAB vs. BAB). It was not stimulatory to csn1s2 translation without a 3′ UTR (AA0 vs. BA0) and it inhibited csn2 translation in the presence of the csn1s2 3′ UTR (ABA vs. BBA). The stimulatory effect of the csn1s2 5′ UTR is consistent with our expectation from its lower thermal stability and more common start codon context but it runs counter to the lower efficiency of csn1s2 translation.

The 3′ UTRs of csn2 and csn1s2 have no effect on translation of csn2 and csn1s2 with or without 5′ UTRs

The effects of the 5′ UTR on casein synthesis appeared to dominate the 3′ UTR effects, in contrast to the luciferase findings where 3′ UTRs dominated. Addition of the csn2 3′ UTR onto 0B0 or BB0 neither increased nor decreased csn2 translation (Fig. 5B). Similarly, the csn1s2 3′ UTR had no effect on csn1s2 translation when added to 0A0 or AA0 (Fig. 5B). Interchanging 3′ UTRs between csn2 and csn1s2 did not change translational efficiency (BBA vs. BBB, and AAA vs. AAB). The csn2 3′ UTR, that was stimulatory to luciferase translation, had little effect on casein translation except for a stimulation of translation of csn2 with a csn1s2 5′ UTR (ABB vs. ABA).

The coding region of csn1s2 is responsible for its low translational efficiency

The relatively low efficiency of translation of csn1s2 compared to csn2 was not improved by exchanging UTRs with csn2. In fact, none of the csn1s2 constructs carrying various combinations of UTRs were translated more efficiently than the full-length csn1s2 (AAA). Thus, the difference in translational efficiency between csn2 and csn1s2 is not due to their UTRs and we decided to subject the coding regions of csn2 and csn1s2 to further analysis. Without any UTRs, the coding region of csn2 was translated 4.7 times more efficiently than that of csn1s2 (BhBh vs AhAh; Fig. 6), verifying an impediment to translation in the coding region itself. We divided each coding region into 2 halves and cross-exchanged them between constructs (Fig. 6). Intriguingly, the construct containing the 5′ half of csn1s2 and the 3′ half of csn2 (AhBh) was translated 1.6 times more efficiently than the complete csn2 coding region (BhBh). When compared to the csn1s2 coding region by itself (AhAh) the translational efficiency of AhBh was 7.4-fold higher. This implicates the second half of csn1s2 in the low translational efficiency. Indeed, replacement of the second half of csn2 with the second half of csn1s2 (BhAh) suppressed translational efficiency compared to BhBh by 50% (Fig. 6).

Figure 6.

In vitro translational efficiency of full csn1s2 (AhAh) and csn2 (BhBh) coding regions and constructs with cross-exchanged halves of coding regions (AhBh and BhAh). Replacing the 3′ half of csn2 coding region with that from csn1s2 significantly suppressed translation whereas replacing the first half of csn2 coding region with that from csn1s2 exerted a stimulatory effect. Data shown are means ± SE (n = 4 independent experiments). Means without a common letter differ at P ≤ 0.05.

The last 28 codons of csn1s2 are the main cause of its low translational efficiency

In the coding region of a gene, there are multiple codons that code for the same amino acid through synonymous tRNA isoacceptors that are present in different abundances in the cell. If several of these low-efficient codons are clustered together, a bottleneck in ribosome transit may occur and slow down translation. The codon usage and order has been shown to influence the accuracy and the speed of translation.^27–30 We analyzed the distribution of codon efficiencies along the coding region of both csn2 and csn1s2 using the correlations between codon usage and cell fitness reported by Navon and Pilpel ³¹ (Fig. 7A). Only statistically significant correlation values were used in the analysis. Values are arranged so a negative correlation indicates a burden on cell fitness. When these correlation values were applied to the codon sequence of csn2 and csn1s2 coding regions, 2 distinct patterns emerged. The csn2 coding region exhibits fairly balanced codon usage except in the first 4 to 21 codons which have either non-significant or positive correlation values (Fig. 7A). The csn1s2 coding region, on the other hand, contains clusters of low values, especially in the 3′ half and, most noticeably, in the last 28 codons (Fig. 7B). To investigate whether these last 28 codons of csn1s2 with low correlation values contribute to its low translational efficiency, mRNA constructs of the csn1s2 coding regions after deleting the last 112, 70 or 28 codons were subjected to translation (Fig. 7C). Deletion of the last 28 codons improved efficiency of translation of the csn1s2 coding region by 2.9 fold, while deletion of the final 70 and 112 codons did not increase translation but reduced it (Fig. 7C). These results suggest that the last 28 codons of csn1s2 have a strong inhibitory effect on its translational efficiency. The AhA3 construct lacking the last 28 codons of the csn1s2 coding region was translated with equivalent efficiency to the BhBh construct coding for the entire amino acid sequence of csn2.

Figure 7.

Deletion of coding fragments from csn1s2 and effects on translational efficiency. (A) Schematic representation of the constructs is on the top panel. Solid and dashed lines represent csn1s2 and csn2 codons, respectively. Arrows indicate deletion locations. Numbers indicate codon distance from the start codon. 3′ terminal 112, 70 or 28 codons of csn1s2 coding region were deleted as shown in (A) and the last codon of each construct was mutated to a stop codon (TAA). (B) Deletion of the last 28 codons improved efficiency of csn1s2 coding region (AhAh) translation 2.9-fold, while deletion of the last 70 or 112 codons reduced translation. Translational efficiency of csn1s2 shortened by 28 codons was 28% higher than BhBh which encodes β-casein. (C) Deletion of the 3′-terminal 20, 24 or 28 codons increased translational efficiency compared to the full csn1s2 coding region. Data shown are means ± SE (n = 4 independent experiments). Means without a common letter differ at P ≤ 0.05.

To narrow down within the last 28 codons, 4 codons at a time were deleted from the 3′ terminus of the csn1s2 coding region (Fig. 7D). Deletion up to 16 codons did not alter the translational efficiency but protein accumulation was improved with each 4-codon deletion thereafter up to codon 28 (Fig. 7D). This analysis pinpoints the sequence between codons 16 and 28 upstream of the stop codon as responsible for the 5-fold difference in translational efficiency between csn2 and csn1s2.

Discussion

Analysis of mammary tissue collected from goats, sheep and cows has revealed that csn2 and csn1s2 mRNA are expressed at similar levels but β-casein accumulation in milk is 4 times that of α_s2-casein.⁸ In mammary tissue from water buffalo, csn2 and csn1s2 represent 23 and 56% of casein transcripts, respectively, while their corresponding protein concentrations in milk are 54 and 5%, respectively, of total caseins, indicating approximately 10-fold more efficient translation of csn2.⁹ To define molecular features of the mRNA that account for this difference in translational efficiency, we first needed to determine if the difference could be recapitulated in a cell-free RRL. Indeed, we found that translational efficiency of uncapped bovine csn2 in RRL was 5 times more than that of csn1s2 (Fig. 2D), which is close to the 4-fold difference reported in vivo.⁸ Furthermore, this difference was retained when just the coding regions of csn2 and csn1s2, without UTRs, were translated in RRL.

It has been suggested that the relatively low translational efficiency of csn1s2 might be due to a weak sequence context around the start codon or a putative hairpin structure in the 5′ UTR of csn1s2.^8,9,32 The nucleotide context in which the start codon lies influences detection by the scanning ribosome. Point mutation analysis showed that the sequence around the start codon that was most frequently observed in a survey of 699 vertebrate genes³³ enhances translation initiation.³⁴ Bevilacqua et al.⁸ compared all 4 casein variants with the Kozak sequence, GCCGCCRCCAUGG, and considered both csn2 and csn1s2 to possess a weak start codon context. However, subsequent studies report a diversity of preferred sequences across vertebrate species. According to the preferred bovine sequence of GCCAACAUGGCG compiled from 1337 genes of Bos taurus,¹⁴ csn1s2 contains the most frequently found nucleotides in 5 of the 6 positions immediately upstream of the codon, whereas csn2 contains only 1 nucleotide of the preferred context. In addition to the nucleotide sequence, location of hairpins close to the start codon significantly reduces translation of the adjacent coding sequence.³⁵ In silico analysis predicted that the 5′ UTR of csn1s2 contains 2 hairpin structures before the start codon and more complicated loops at the beginning of the coding region (Fig. 3D). The 5′ UTR of csn2 (Fig. 3C), in contrast, forms a fairly linear structure before and after the start codon. However, contrary to the suggestion that the 5′ UTR of csn1s2 is responsible for its relatively inefficient translation, we found that, compared to the 5′ UTR of csn2, the csn1s2 5′ UTR enhances translation of downstream csn2, csn1s2 and firefly luciferase coding regions in the MAC-T cell line and in cell-free RRL (Fig. 4). This suggests that the 5′ UTR of csn1s2 is not the reason for its low translational efficiency. Rather, it enhances translation, possibly in part due to a preferred nucleotide sequence around the start codon.

Generally, lengths of the UTRs correlate with translational efficiency across the range of mammalian 5′ UTR-lengths from 20 to 936 nucleotides, and 3′ UTR-lengths from 37 to > 3000 nucleotides.³⁶ The lengths of csn2 and csn1s2 UTRs are not very different from each other; both mRNAs have fairly short 5′ and 3′ UTRs, approximately 60 and 300 nucleotides, respectively. However, the 3′ UTR of csn1s2 interacts with its coding region to form secondary structures whereas the 3′ UTR of csn2 folds away from the coding region, leaving it exposed (Fig. 3E and F). Our expectation that the csn1s2 3′ UTR would be inhibitory to translation was confirmed in the luciferase assay in cultured cells and when added to the csn2 coding region in an in vitro translation system. Thus, the 5′ and 3′ UTRs of csn1s2 exert opposing regulatory effects.

When one fast and one slow UTR were inserted at either end of the luciferase coding region, the 3′ UTR masked the 5′ UTR effect. No matter which 5′ UTR was present, the csn1s2 3′ UTR always inhibited luciferase translation compared to the csn2 3′ UTR. At first glance, this suggests that regulatory elements possess different strengths, where the effect of one UTR can be masked by stronger elements within the same mRNA sequence. However, findings with the casein constructs did not yield the same ranking of UTR regulatory strengths and the concept must be discarded as too simplistic. Instead of masking 5′ UTR effects, the 3′ UTR of csn1s2 was inhibitory to only one of the 6 casein constructs (Fig. 5B and D). The 5′ UTR of csn1s2 was stimulatory to translation of csn2 carrying a 3′ UTR from csn2 and then inhibitory with a csn1s2 3′ UTR. But when that same 3′ UTR from csn1s2 was linked to the csn1s2 coding region, the 5′ UTR of csn1s2 was stimulatory again. These inconsistencies indicate that UTR effects are dependent on the entire sequence of the transcript with which it has the opportunity to form secondary structure. Our results show that the common practice of studying UTR effects on translation efficiency using a reporter construct cannot be guaranteed to reproduce their effects in a native context.

The conclusion that the coding region of csn1s2, and not its UTRs, is responsible for the low translational efficiency compared to csn2 is based on 1) an inability to stimulate csn1s2 translation with csn2 UTRs (Fig. 4D), 2) the 4.7-fold difference in protein accumulation in RRL from csn2 and csn1s2 constructs without UTRs (Fig. 6), and 3) the absence of differences in profile of RNA-bound proteins (Supplementary Fig. S4). Involvement of the coding region in the translation attenuation is also consistent with the finding from sucrose density gradient centrifugation (Supplementary Fig. S3) that elongation of csn1s2, which occurs as the ribosome scans the coding sequence, was suppressed more than initiation, in which UTRs play a role. In simple terms, the rate of protein synthesis is equal to the rate of translation initiation, where elongation plus termination represents the time delay between the two. Thus, elongation itself does not influence efficiency of protein synthesis unless initiation is also affected. Initiation can be affected when pauses in elongation cause a traffic jam of ribosomes on the mRNA so that new ribosome attachment is sterically impaired.³⁷ Distribution of csn1s2 into heavier polysomes, while protein synthesis is slowed, identifies the occurrence of ribosome congestion due to an impaired elongation rate.

Possible explanations for the elongation attenuation include stable secondary structures in the coding region, because the ribosome must pause to melt helices and hairpins in order to scan the primary nucleotide sequence,³⁸ and low availability of acylated tRNA acceptors for the codons on display. Removing secondary structure by denaturing full-length csn2 and csn1s2 transcripts significantly improved translational efficiency compared to re-folded transcripts, and a greater improvement was observed in csn1s2 (Fig. 3G and H). However, the translation of csn2 was still more efficient than that of csn1s2 in the completely denatured state (Data not shown). Thus it appears that secondary structure can be only partly responsible for the translational difference between the casein mRNAs.

The multiple tRNA acceptors for each amino acid in a cell are not all present at the same abundance, which creates a codon usage bias. Codons read by abundant tRNA isoacceptors are typically translated faster than those that are decoded by rare isoacceptors.^39,40 Replacement of rare codons in an mRNA transcript with common synonymous codons strongly improves the efficiency with which the sequence is translated.^41–43 According to the correlation coefficients between codon frequency and cell fitness reported by Navon and Pilpel,³⁰ we identified a 12-nucleotide segment of low-efficiency codons between codons 195 and 207 at the 3′ terminus of the csn1s2 coding region that, when deleted, increased translational efficiency to a par with csn2. It is becoming apparent that it is the clustering of codons of like efficiencies together that affects translational efficiency, rather than average codon usage bias over the full mRNA sequence.^25,28,44 For example, Tuller et al.⁴⁰ discovered that the first 30 to 50 codons of eukaryotic coding regions tend to code for rare tRNAs, which was proposed to govern the speed of translation initially, preventing congestion of ribosomes further down the line. The last 50 codons, in contrast, are recognized by abundant tRNA acceptors.⁴⁰ In contrast to a highly expressed gene, csn1s2 contains a cluster of low-efficiency codons at the 3′ terminus, suggesting that elongation would be slowed as the ribosome approaches the stop codon, consistent with the distribution of csn1s2 transcripts into heavier polysomes than csn2 transcripts. Interestingly, csn2 does not display an optimal codon sequence either. The first 21 codons contain a high number of those that correlate positively with cell fitness. Acceleration of ribosome transit at the beginning of the coding region can create ribosomal traffic jams at more distal parts of the mRNA.⁴⁰ Indeed, replacement of the first half of the coding region of csn2 with that from csn1s2 enhanced translation (Fig. 6).

The section of the csn1s2 sequence from codons 194 to 206 that we identified as impeding translation codes for predominantly hydrophobic amino acid residues and a high proportion of residues with positively charged side chains.⁴⁵ Codons 139–222 and 57–137 of bovine csn1s2 display a homology that suggests they arose by gene duplication, although the C-terminal half of the second segment codes for more hydrophobic and positively charged amino acid residues than that of the first segment. In fact, the 37 residues at the C-terminus of α_s2-casein constitute the most positively charged section of all the caseins.⁴⁵ Across species, there is a large variation in amino acid sequence in this section of the α_s2-like caseins³¹ but the hydrophobic character is preserved. This hydrophobicity is considered important for formation of the micelle with α_s1-, β- and κ-caseins in milk that stabilizes high concentrations of Ca₂PO₄ for delivery through the immature neonatal digestive tract in support of skeletal development. The positive charges are possibly required for binding to inorganic phosphate or phosphoserine residues of the other caseins.

Approximately 82% of the mass of protein in cow's milk is made up of α_s1-, α_s2-, β-, and κ-caseins in a 4:1:4:1 ratio, respectively. A wide range of casein variants and proportions is found across mammalian species, and even within some species due to polymorphisms in casein genes⁴⁶ and differential splicing of transcripts.⁴⁷ Human milk protein is less than 40% casein⁴⁸ and contains no α_s2-casein.⁴⁹ Rat milk is 70% casein⁵⁰ and contains 2 α_s2-like caseins in addition to α_s1-, β-, and κ-caseins.⁵¹ The proportion of the different caseins in milk influences the functionality of the milk. For example, increasing α_s2-casein content of goat's milk at the expense of α_s1-casein improves the uptake of milk Ca into bone when consumed by rats ⁵² and changes the renneting properties of milk during cheesemaking.^53,54 Nutritional and hormonal interventions in individual dams that affect daily milk protein yields, such as those that occur at the onset of lactation or under varying nutrient supplies, have little influence on the ratio of caseins, one to another, in their milk.^55–57 The profile, then, is set by genetic mechanisms. Strategies that have been exploited under natural and, for dairy animals, artificial selection pressures to produce the wide range of casein proportions across individuals include modification of casein gene transcriptional control, mRNA splicing, stop codon location, and transcript stability.⁴⁶ Our findings show that codon usage bias also contributes to the observed proportion of caseins present in milk. Occurrence of synonymous codon usage bias in vertebrates has largely been attributed to the non-selective processes of mutation bias and genetic drift, but recent arguments and evidence have been put forth to support the existence of selective advantages of alternative synonymous codon usage.^58,59 Maintaining the ratio of proteins one to another in milk may offer a novel example of selective advantage of codon usage bias. The presence of low-efficiency codons at the 3′ terminus of bovine csn1s2 keeps α_s2-casein expression low and in correct proportion to the other micelle-forming caseins. Identification of the nucleotide sequence responsible for low efficiency of casein translation carries significance for endeavors to use synthetic biology to study casein micelle formation and to accelerate production of high-quality milk protein for a burgeoning human population.⁶⁰

Materials and Methods

5′- and 3′-RACE PCR of csn2

Mammary tissue was obtained from a lactating Holstein cow at slaughter, frozen in liquid nitrogen and stored at −80°C. Total RNA was isolated from ∼100 mg mammary tissue using 1 ml of TRizol Reagent (Invitrogen, #15596-018), according to manufacturer's instructions. The integrity (RIN 8.3) and quantity of the RNA were verified using an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, USA) and NanoDrop 8000 spectrophotometer (Thermo Scientific, Wilmington, USA). cDNA was then synthesized from 1 µg of total RNA using SuperScriptase III Reverse Transcriptase (Invitrogen, #18080–044) with Oligo-dT primer, following manufacturer's instructions. The cDNA was purified using the High Pure PCR Product Purification Kit (Roche Applied Sciences, #11732668001), according to manufacturer's instructions. The purified cDNA was incubated at 37°C for 15 min with 400 U Terminal Transferase (TdT, Roche Applied Sciences, #03333566001), 50 pmol dATP, 1ċ TdT buffer, and 5 nmol CoCl₂. Following incubation, 1 µl of the reaction was used as a template for PCR to amplify the 5′ end with Oligo-dT and BC_R1 primer and the 3′ end with BC_F1 and Oligo-dA primer (Supplementary Table S1). PCR was performed using Platinum Pfx DNA Polymerase (Invitrogen, #11708013), according to manufacturer's instructions. The amplicons were gel-purified using QIAquick Gel Extraction Kit (Qiagen, #28704) and cloned into pGEM-T easy vector (Promega, #A1360), following each manufacturer's instructions after addition of adenosine triphosphate to the 3′ ends of each amplicon with Taq DNA Polymerase (Thermo Scientific, #EP0402) and 1x Taq Buffer and 2 mM dATP at 72°C for 10 min. The cloned 5′-and 3′-ends of csn2 were sequenced using an Applied Biosystems 377 automated sequencer (Perkin-Elmer Biosystems, Boston, USA).

Cloning and sequencing of the full-length transcripts of csn2 and csn1s2

The cDNA synthesized using the same procedure as for 5′- and 3′-RACE PCR of csn2 was used to amplify the full-length transcripts of csn2 and csn1s2. Each set of gene-specific primers with restriction enzyme sites, EcoRI and HindIII, was designed based on the full-length nucleotide sequences of csn2 and csn1s2 that were obtained from 5′- and 3′-RACE PCR and the NCBI database (NCBI Reference Sequence: NM_174528.2), respectively (Supplementary Table S1). PCR was performed using 10 ng cDNA, gene-specific primers for each cDNA, and Platinum Pfx DNA Polymerase (Invitrogen, #11708013). The amplicons were purified using the High Pure PCR Product Purification Kit (Roche Applied Sciences, #11732668001). The purified PCR products were digested with restriction enzymes, EcoRI and HindIII (Thermo Scientific, #ER0271 and ER0501) and ligated into EcoRI- and HindIII-digested pTZ19R vectors, under the control of a T7 promoter (Thermo Scientific, #SD0141) using T4 DNA Ligase (Thermo Scientific, #EL0014), followed by transfection into Escherichia coli DH5α competent cells. Positive colonies containing each construct were confirmed by Colony PCR using Taq DNA Polymerase (Thermo Scientific, #EP0402). Isolation of plasmid DNA was carried out using PureLink Quick Plasmid Miniprep Kit (Invitrogen, #K2100–10), according to the manufacturer's instructions. The full-length sequences of csn2 and csn1s2 were confirmed using an Applied Biosystem Model 377 automated sequencer (Perkin-Elmer Biosystems, Boston, USA). ClustalW-Multiple Sequence Alignment tool accessed at http://www.ebi.ac.uk/Tools/msa/clustalw2/ was used for multiple sequence alignments and sequencing analysis.

Transient transfection assay

The MAC-T bovine mammary epithelial cell line was used for transient transfection assays. Cells were grown on Dulbecco's Modified Eagle medium (Invitrogen, #11995–065) supplemented with 10% fetal bovine serum, 1 mM sodium pyruvate, 40 U/mL penicillin/streptomycin, 25 mM HEPES, 5 µg/mL bovine insulin (Sigma Aldrich, #I6634) at 37°C in a humidified atmosphere with 5% CO₂. A day before transfection, cells were plated at a density of 1x10⁴ cells/32 mm in 12-well cell culture plate (BD Biosciences, #351143). At ∼80% confluence, the MAC-T cells were co-transfected with 1.5 µg of various firefly luciferase (FLuc) reporter constructs (Fig. 4A) combined with 0.5 µg of pRL-TK renilla luciferase (RLuc) control vector in 6 µL FuGENE HD Transfection Reagent (Promega, #E2311) according to the manufacturer's recommendations. All reporter constructs were sequenced prior to transfection to confirm lack of any mutations. 24 h post-transfection, cells were rinsed in cold PBS then lysed with 20 µL of passive lysis buffer provided in Dual Luciferase Assay System (Promega, #E1910). After a brief centrifugation, 10 µL of each supernatant was assayed for luciferase activity in a luminometer where light emission was measured for 15 s with a Victor³ luminometer (Perkin Elmer). The FLuc activity for every reporter construct was normalized to the corresponding RLuc activity to account for varying transfection efficiency.

Rabbit reticulocyte lysate (RRL) system for in vitro transcription and translation

Upon sequencing analysis, the plasmid DNAs containing the full-length cDNA sequences of csn2 and csn1s2 were designated as pTZ19R-BBB and pTZ19R-AAA, respectively. The pTZ19R-BBB and pTZ19R-AAA plasmids were linearized with EcoRI (Thermo Scientific). To amplify fragments, 5′ or 3′ UTR, of interest for various recombinant constructs using Platinum Pfx DNA Polymerase (Invitrogen, Carlsbad, CA, USA), the pTZ19-BBB or pTZ19-AAA plasmid DNA was used as a template with the gene-specific primers (Supplementary Table S1). An extra round of blunt-end ligation and PCR amplification was carried out to generate recombinant plasmids in which fragments of csn2 and csn1s2 sequences were cross-exchanged (Fig. 4A). Sequence of all constructs was confirmed prior to further experiments. The recombinant constructs were purified using the High Pure PCR Purification Kit (Roche Applied Sciences) and 1 µg of each construct was transcribed with T7 RNA polymerase (Promega, Madison, WI, USA). Transcribed RNAs were separated from DNA templates using phenol/chloroform (pH 4) and ethanol/ammonium acetate precipitation. The integrity and quantity of RNAs were determined by 1.2% formaldehyde-agarose gel electrophoresis and NanoDrop 8000 spectrophotometer (Thermo Scientific, Walthman, MA, USA). Equivalent concentrations (500 ng) of each mRNA were translated in nuclease-treated RRL (Promega, #L4960) system, according to manufacturer's instructions, with addition of 0.5 µl Transcend tRNA (Promega, #L5061) in 25 ul reactions for biotinylated lysine labeling of nascent proteins during translation. Number of moles in 500 ng mRNA were calculated for each construct and used for normalization of translated proteins.

Each in vitro translation reaction product with laemmli sample buffer (Sigma Aldrich, #S3401) was separated using 12% SDS-PAGE. The proteins were then transferred onto PVDF membrane and blotted using Streptavidin-Horseradish Peroxidase (HRP), followed by chemiluminescent detection exposed on X-ray film using an SRX101A Tabletop X-ray Film Processor (Konica Minolta Medical Imaging USA Inc.). Exposed PVDF membrane was stained with 0.1% (w/v) Fast Green (Fisher Scientific) to confirm consistent loading and running on the gel. Band intensities on X-ray film and PVDF were scanned and analyzed by densitometry using NIH Image/J 1.6 Software available at http://rsb.info.nih.gov/ij/.

In vitro RNA and protein decay assay

Concentrations of the mRNA constructs and protein products in the RRL system were measured by RT-qPCR and Transcend Non-Radioactive Translation Detection System (Promega, #L5080), respectively. Upon EcoRI enzyme (Thermo Scientific) digestion of the pTZ19R-BBB and pTZ19R-AAA plasmids, 1 µg of each linearized plamid was purified and transcribed using the High Pure PCR Product Purification Kit (Roche Applied Sciences) and the T7 RNA polymerase (Promega, #P2075), respectively. During in vitro transcription, the reactions were collected at successive time intervals and RNAs were then extracted using TRIzol Reagent (Invitrogen, #15596–018). The integrity and quantity of RNAs were determined by 1.2% formaldehyde-agarose gel electrophoresis and NanoDrop 8000 spectrophotometer (Thermo Scientific). cDNA was synthesized from 500 ng of extracted total RNA with random hexamers using High Capacity cDNA Reverse Transcription Kit (Applied BioSystem), following manufacturer's instructions. qPCR was carried out using PerfeCta SYBR Green FastMix (Quanta BioScience) with an Applied Biosystems 7300 Real Time PCR instrument. The data obtained were analyzed by the 2^−ΔΔCt method.⁶¹ Primer sequences used are listed in Supplementary Table S1. During in vitro translation, aliquots of 10 µl were removed at successive time intervals. The removed reactions were analyzed by the Transcend Non-Radioactive Translation Detection System (Promega, Madison, WI, USA). Band intensities on X-ray film were scanned and analyzed by densitometry using NIH Image/J 1.6 Software available at http://rsb.info.nih.gov/ij/.

In silico RNA secondary structure predictions

The secondary structures and the minimum free energy, ΔG, of the full-length and individual regions of csn2 and csn1s2 were predicted using the Mfold and Rfold programs available at http://www.bioinfo.rpi.edu/applications/mfold²³ and http://www.ncrna.org/software/rfold/.²²

The effect of RNA secondary structure

To determine the effect of secondary structure on translational efficiency, pTZ19R-BBB and pTZ19R-AAA in vitro transcribed by T7 RNA polymerase (Promega) were denatured or refolded before translation. To denature secondary structure, each mRNA was heated at 65°C for 15 min and immediately placed on ice. For refolding, pre-denatured mRNAs were cooled at 0.1°C/s to a final temperature of 15°C in 5 mM MgCl₂.⁶² Within a minute of being put on ice, denatured and refolded mRNAs were translated in RRL system for 15 and 30 min.

Mathematical model

To account for mRNA and protein degradation influencing protein accumulation, the compartmental model of Hargrove and Schmidt,²⁰ with modifications, was fit to mRNA and protein accumulation time-courses over 60 and 90 min of incubation in RRL.

Differential equations for the model are:

$${\displaystyle \frac{\text{dR}}{\text{dt}}=-{\text{k}}_{\text{d1}}\text{R}}$$ (1)

and

$${\displaystyle \frac{\text{dP}}{\text{dt}}={\text{k}}_{\text{s}}{\text{e}}^{{\text{−k}}_{\text{d3}}\text{t}}\text{R}-{\text{k}}_{\text{d2}}\text{P}}$$ (2)

Analytical solutions for the timecourse of R and P concentrations during the RRL incubation are:

$${\displaystyle {\text{R}}_{\text{t}}={\text{R}}_{\text{0}}{\text{e}}^{-{\text{k}}_{\text{d1}}\text{t}}}$$ (3)

and

$${\displaystyle {\text{P}}_{\text{t}}=\frac{{\text{k}}_{\text{s}}{\text{R}}_{\text{0}}}{{\text{k}}_{\text{d1}}-{\text{k}}_{\text{d2}}+{\text{k}}_{\text{d3}}}\left({\text{e}}^{-{\text{k}}_{\text{d2}}\text{(t}-\text{TT)}}-{\text{e}}^{-\left({\text{k}}_{\text{d1}}+{\text{k}}_{\text{d3}}\right)\text{(t}-\text{TT)}}\right),}$$ (4)

where R₀ is the initial concentration of R at time 0, and TT is the transit time between attachment of ribosomes and release of a complete protein. Parameters R₀ and k_d1 were estimated by least-squares fits of (3) to mRNA timecourse curves using Solver in Microsoft Excel, and k_s, k_d2, k_d3, and TT were estimated by fits of (4) to protein timecourses. Arbitrary units representing density of protein immunoblots were normalized to their respective peak densities at 20 min of incubation to generate curves for fitting. Normalized densities for β-casein were multiplied by 3 to account for the difference in protein accumulation compared to α_s2-casein (Fig. 1B), and by 25/12 to account for the difference in number of biotinylated lysine residues.

Statistical analysis

Data are presented as means ± SE. Results are from 3 independent experiments unless stated otherwise. Differences between means were determined by one-way analysis of variance using Tukey's test for multiple comparisons. Differences were considered significant at P ≤ 0.05.

Disclosure of Potential Conflicts of Interest

No potential conflicts of interest were disclosed.

Acknowledgement

We acknowledge the support received from the Natural Sciences and Engineering Research Council of Canada. The authors also appreciate the technical help provided by Jing Zhang (Genomic Facility, University of Guelph).

Funding

This work was supported by the Natural Sciences and Engineering Research Council of Canada.

Supplemental Material

Supplemental data for this article can be accessed on the publisher's website.