Receptor of activated protein C kinase 1 (RACK1) is a highly conserved eukaryotic protein that regulates several aspects of mRNA translation; yet, how it does so, remains poorly understood. Here we show that, although RACK1 consists largely of conserved β-propeller domains that mediate binding to several other proteins, a short interconnecting loop between two of these blades varies across species to control distinct RACK1 functions during translation. Mutants and chimeras revealed that the amino acid composition of the loop is optimized to regulate interactions with eIF6, a eukaryotic initiation factor that controls 60S biogenesis and 80S ribosome assembly. Separately, phylogenetics revealed that, despite broad sequence divergence of the loop, there is striking conservation of negatively charged residues amongst protists and dicot plants, which is reintroduced to mammalian RACK1 by poxviruses through phosphorylation. Although both charged and uncharged loop mutants affect eIF6 interactions, only a negatively charged plant – but not uncharged yeast or human loop – enhances translation of mRNAs with adenosine-rich 5′ untranslated regions (UTRs). Our findings reveal how sequence plasticity within the RACK1 loop confers multifunctionality in translational control across species.
Beyond transcriptional responses, regulated translation of individual mRNAs enables cells to rapidly adjust the levels of specific proteins during a wide range of processes. Although this regulation is exerted at all stages of translation, much of it occurs during initiation when the 40S ribosomal subunit is first loaded onto, and scans the 5′UTR of, an mRNA (Hinnebusch et al., 2016). Once the 40S subunit identifies a start codon it is joined by a 60S subunit to form a translationally competent 80S ribosome that can begin decoding the mRNA open reading frame (ORF). Although each step is regulated by several eukaryotic initiation factors (eIFs) to control translation efficiency, ribosomes themselves are emerging as central players in the regulation of translation rates of individual mRNAs (Dinman, 2016; Genuth and Barna, 2018). This concept of ‘ribosome specification’ posits that ribosomes are not homogenous and indiscriminate machines but, instead, vary in composition to selectively control translation through the activity of individual ribosomal protein (RP) subunits that operate on specific 5′UTR elements. However, we have a limited understanding of how and when ribosome diversification arises.
Receptor of activated protein C kinase 1 (RACK1) has recently emerged as a particularly intriguing small ribosomal protein (RPS) that functions in several aspects of translation. Containing seven Trp-Asp (WD) repeats and adopting a seven-bladed β-propeller structure, RACK1 is a highly conserved eukaryotic protein that is positioned on the head of the 40S subunit in the vicinity of the mRNA exit channel (Gallo and Manfrini, 2015; Sengupta et al., 2004) (Fig. 1A). Although RACK1 has both ribosomal and extra-ribosomal functions in some transformed cell lines, it is now clear that it in several cell types and many normal cells RACK1 predominantly functions on the ribosome (Gallo and Manfrini, 2015; Nielsen et al., 2017; and discussed later herein). In doing so, RACK1 serves as a hub for host signaling to the protein synthesis machinery as well as directly controling translation (Gallo and Manfrini, 2015; Nielsen et al., 2017). A number of its cytosol-facing β-propeller domains enable RACK1 to interact with eIF3c (Kouba et al., 2012), one of several subunits of the eIF3 complex that interacts with various RPSs to bridge the 40S subunit to the 5′-end of mRNAs (Hinnebusch et al., 2016). Similarly oriented propeller domains also mediate interactions with kinases, such as PKCβII, to transmit signals to ribosome-associated initiation factors (Dobrikov et al., 2018a,b; Grosso et al., 2008). However, RACK1 is generally not required for the efficient synthesis of most proteins but, instead, facilitates translation of certain viral transcripts that contain either internal ribosome entry sites (IRESs) or poly(A) tracts, as well as small subsets of mRNAs in mammals and yeast (Ceci et al., 2012; Gallo et al., 2018; Jha et al., 2017; Núñez et al., 2009, 2010; Rachfall et al., 2013; Romano et al., 2019; Ruan et al., 2012; Shor et al., 2003; Thompson et al., 2016). Indeed, RACK1 can directly regulate ribosome activity and controls for example, frameshifting and ribosome quality-control responses induced by specific mRNA ‘stall sequences’ (Garzia et al., 2017; Ikeuchi and Inada, 2016; Juszkiewicz and Hegde, 2017; Kuroha et al., 2010; Letzring et al., 2013; Sitron et al., 2017; Sundaramoorthy et al., 2017; Wang et al., 2018; Wolf and Grayhack, 2015). Yet, beyond its β-propeller domains that mediate several of its protein–protein interactions, how RACK1 regulates translation remains poorly understood.
We recently found that the poxvirus family member vaccinia virus (VacV) phosphorylates an STSS motif in a short variable loop that lies between the 6th and 7th β-propeller blade, and extends from RACK1 toward the ribosome (Jha et al., 2017) (Fig. 1A,B). Intriguingly, phosphorylation of this motif does not appear to occur outside the context of VacV infection, being driven by a unique viral kinase, and functions to promote translation of poxvirus mRNAs that contain unusual 5′poly(A)-leaders (Jha et al., 2017; Meade et al., 2018a). Moreover, the introduction of phosphate by VacV mimics the presence of negatively charged amino acids that are present in the RACK1 loop of the plants Nicotiana tabacum and Arabidopsis thaliana but are absent in human, mouse, worm and yeast loops (Jha et al., 2017). However, although these findings revealed a role for the RACK1 loop in VacV mRNA translation, the mechanistic basis by which the loop region functions and its broader importance beyond infection remain unknown. Here, we show that the RACK1 loop exhibits broad sequence plasticity across species and controls two distinct aspects of translation. First, independent of its charge status, RACK1 loop sequences are differently optimized in species to regulate interactions with the eukaryotic initiation factor eIF6, a factor that controls 60S biogenesis and 80S assembly pathways. Second, phylogenetics reveals that specific groups known to utilize mRNAs with 5′poly(A)-leaders also encode RACK1 loop regions that harbor negatively charged residues. Functional testing reveals that distinct from regulating eIF6 interactions, only a negatively charged plant RACK1 loop enhances translation of mRNAs with 5′poly(A)-leaders. Moreover, modeling and biochemical testing suggests that the RACK1 loop charge generates electrostatic forces that are carefully controlled through spatial organization, and probably remodel the mRNA exit channel to accommodate the unusual structures adopted by poly(A) leaders. Overall, our findings suggest that sequence plasticity in its loop region enables RACK1 to control distinct aspects of translation in different species.
Evolutionary divergence of RACK1 loop sequence and charge
Prompted by our recent studies of VacV that involved a limited comparison of negatively charged residues in RACK1 loops across seven organisms (Jha et al., 2017), we assembled a phylogenetic tree from a BLAST search of available and predicted eukaryotic RACK1 protein sequences in the UniProtKB Protein database to determine the broader extent to which loop sequences vary across species (Fig. 1C,D; Table S1). Once duplicates and non-RACK1 sequences were removed, we analyzed ∼1000 species variants of RACK1. This included 31 protists, 332 animals, 485 fungi and 131 plants. Although some subgroups had limited sequence availability, overall this approach provided broad coverage of kingdoms and most groups. In doing so, we noticed that 93.6% of all protist loops harbor negatively charged amino acids, 38.7% of which contain multiple consecutively arranged charged residues, whereas 54.8% contained singular or multiple, yet spatially separated, charge configurations (Fig. 1D). Examples of these different charge organizations in protists are shown in Fig. 1C, which notably includes the spaced configuration of D. discoideum, whose mRNAs were shown to contain long 5′poly(A) tracts (Steel and Jacobson, 1991). By contrast, negatively charged residues are almost completely absent in loop regions across all metazoans (vertebrate and invertebrate) as well as most plants, with two notable exceptions. 35.3% of chlorophytes contain negatively charged residues, which either occur singularly or in a spaced out configuration. Most strikingly, 100% of eudicotyledons (dicot plants) harbor negatively charged residues. While some harbor clustered charge configurations, 75.3% of RACK1 loop sequences in dicot plants contain multiple charged residues that are spatially separated. Finally, fungi contain a heterogenous mix of uncharged and charged loops, with most charged configurations again arranged singly or in a spatially separated fashion (Fig. 1D, Table S1). These overall phylogenetic patterns might represent several individual events (convergent evolution), wherein negative charge was introduced into the RACK1 loop in protists and dicot plants, and is actively explored across chlorophytes and fungi. Alternatively, these patterns might represent evolutionary pressures driving the loss and/or spatial reorganization of negative charge from ancestral forms of RACK1. In the latter scenario, clustered charge found more frequently in protists may suggest that ancestral RACK1 loops were highly charged and often contained consecutive negative residues. Mixed phylogenetics in fungi may indicate ongoing selective pressures against negative charge that are complete in animals and most plants, while a specific evolutionary advantage(s) may have led to the highly conserved retention of charge in dicot plants. Regardless of its evolutionary origins it was notable that, overall, 84.3% of all species with negatively charged loops utilized one single or several albeit spatially separated configurations, hinting at both structural and functional significances regarding this choice (Table S1).
Beyond negatively charged residues, most plants, fungi and animals exhibit broader variations in amino acid sequence. For example, many metazoan loops have polar, uncharged residues that produce an S[T/P][S/N]S motif, similar to the STSS motif in Homo sapiens (human) RACK1 that is targeted by VacV. Caenorhabditis elegans has a slightly modified SSGSS motif. Some fungi possess a distinct uncharged loop motif that is primarily composed of non-polar aliphatic or aromatic residues. For example, the human VISTSS motif is replaced by a FAGYS motif in the yeast Saccharomyces cerevisiae (Fig. 1C). Whereas the loop region of RACK1 is often considered flexible, glycine has a high conformational freedom due to its lack of a side chain, and the yeast loop appears to have a distinct flexibility and structure (Coyle et al., 2009; Krieger et al., 2005). Overall, these phylogenetic comparisons reveal that the RACK1 loop exhibits considerable sequence plasticity that could be of functional importance.
The loop region controls RACK1 interactions with eIF6
To explore the potential functional significance of some of these differences regarding charge or amino acid composition of RACK1 loops across different species, we created loop mutants and chimeras in the background of GFP-tagged human RACK1, and generated stably expressing pools of primary normal human dermal fibroblasts (NHDFs). We chose this experimental set-up for several important reasons, beyond human cells simply serving as our model species. First, primary NHDFs retain normal translational control pathways that are otherwise highly deregulated in most, if not all, commonly used transformed cell lines (Jha et al., 2017; McMahon et al., 2011; Meade et al., 2018b). Second, GFP-tagged RACK1 is functional (Cox et al., 2003; Jha et al., 2017) and allows us to both distinguish and directly compare the behaviors of exogenous and endogenous forms within the same cells. Indeed, expression of GFP-tagged wild-type (WT) RACK1 did not affect overall translation rates in NHDFs, distributed across polysomes similarly to the endogenous form and, as discussed later, only moderately affected polysome profiles within the normal range of variability in these cells (Figs 2 and 3A, Fig. S1A,B). Third, RACK1 has extra-ribosomal functions in several cell lines that might be linked to transformation or only operate in specific cell types or contexts (Gallo and Manfrini, 2015; Schmitt et al., 2017). By contrast, several cell types – including normal cells, such as NHDFs – degrade extra-ribosomal RACK1 and restrict its function to the ribosome (Ceci et al., 2003; Gallo et al., 2018; Gerbasi et al., 2004; Jha et al., 2017; Johnson et al., 2019; Romano et al., 2019; Sengupta et al., 2004). For this reason, exogenous expression of RACK1 downregulates endogenous RACK1 levels, such that RACK1 cannot be overexpressed in these cells (Fig. S1A) (Jha et al., 2017). This homeostatic balance means that supraphysiological levels of RACK1 do not arise in NHDFs, making these cells an ideal system to study RACK1 function.
We first examined the effects of various RACK1 loop mutations and chimeras on ribosome profiles in NHDFs, prompted by our recent observation that a loop mutant wherein the entire STSS motif was mutated to negatively charged, phosphomimetic glutamic acid (E) residues (hereafter referred to as STSS-EEEE mutant) caused a shift towards a predominance of 40S subunits over 80S and polyribosomes (Jha et al., 2017). However, whether this phenomenon is specific to this particular, far less prevalent, clustered charge organization, and the underlying mechanism remains unknown. Western blot analysis showed that endogenous and GFP-tagged forms of human RACK1 were distributed across 40S, 80S and polysome fractions, whereas the large ribosomal protein 11 (RPL11) was detected in 60S, 80S and polyribosomes (Fig. 2A). Similar distribution patterns and ready detection of 60S subunits along with 80S ribosomes occurred in NHDFs expressing GFP-tagged human RACK1, in which the VISTSS motif had been replaced with its yeast counterpart (S. cerevisiae; FAGYS) (Figs 1C and 2A, Fig. S1B). However, expression of GFP-tagged human RACK1, in which the VISTSS motif had been replaced with a dicot plant loop (A. thaliana; LKAEAEKADNSGPAAT; underlining indicates negatively charged residues) caused a notable shift in the sedimentation pattern of both endogenous and exogenous forms of RACK1 towards a predominance of 40S subunits This was accompanied by reduced levels of RPL11 in 60S, 80S and polysome fractions (Fig. 2A). Similar effects to those seen in the plant loop were also observed in NHDFs expressing human RACK1 with amino acid substitutions for negatively charged aspartic acid (D) or glutamic acid (E) residues within in the STSS motif; i.e. single (S278E; hereafter referred to as S-E), double (T277D and S278E; hereafter referred to as TS-DE) or quadruple [S276E, T277E, S278E and S279E (STSS-EEEE)] substitutions (Fig. 2B).
However, the effects of the STSS-EEEE mutant were also observed for another RACK1 loop quadruple mutant, in which the STSS motif had been changed to uncharged alanine (A) residues (STSS-AAAA) (Fig. 2B). Given the propensity of polyalanine motifs to form helices (Marqusee et al., 1989), we next determined the effects of a simple STSS deletion (ΔSTSS). Western blot analysis showed that expression of human ΔSTSS RACK1 also caused a shift towards 40S subunits, and was accompanied by reduced levels of 60S, 80S and polyribsomes (Fig. 2B). Uncropped blots demonstrate the relative abundance of RPL11 across sample sets throughout Fig. 2 and, although modest reductions in the overall levels of RPL11 were often evident, there was a disproportionately large reduction in detectable 60S subunits and 80S ribosomes in cells expressing negatively charged, STSS-AAAA or ΔSTSS RACK1 loop mutants. Representative traces further show the changes in ribosome profiles that accompany various mutations to a charged or uncharged loop (Fig. S1B). It is important to note that the NHDF traces shown in Fig. S1B reflect polysome levels and patterns associated with the low basal translation rates of normal cells, as reported by others (McKinney et al., 2014). This contrasts to traces more commonly associated with transformed cell lines (Fig. S1C), which are highly translationally hyperactivated compared to normal cells (McMahon et al., 2011). Overall, these findings reveal that changes in loop sequence or structure, rather than negative charge alone, caused changes in ribosome levels and profiles. However, the fact that the yeast FAGYS motif is able to operate in human cells suggests that mammalian loops can evolve this functionality in diverse ways and, some at least, are interchangeable despite their broad sequence diversity. By contrast, the lack of functionality of the plant loop in NHDFs suggests that it is distinctly optimized and does not function in other kingdoms.
A fortuitous and initially perplexing clue as to why these RACK1 loop mutations might exert such effects arose when NHDF pools expressing a subset of key RACK1 mutants were generated using reduced retroviral transduction efficiencies. Regardless of high or moderate transduction approaches, there were only subtle differences in the expression levels of GFP-tagged RACK1 (Fig. S2A). Despite this, under moderate transduction conditions both endogenous and exogenous RACK1 were readily detectable across 40S, 80S and polyribosome fractions, as was RPL11 in 60S, 80S and polyribosomes in NHDFs expressing either the S-E mutant or ΔSTSS, as well as either plant or yeast loop chimeras (Fig. 3A). In some instances, 80S levels appeared to be elevated in NHDFs expressing loop mutants that originally reduced 80S levels in high transduction pools. These seemingly contradictory phenotypes in high versus moderately transduced NHDF pools were reminiscent of phenotypes reported for the RACK1-binding eukaryotic translation initiation factor 6 (eIF6) (Ceci et al., 2003; Gallo and Manfrini, 2015). eIF6 binds 60S ribosomal subunits and performs two functions, (1) control of 60S biogenesis in the nucleus and, (2) control of 80S assembly in the cytoplasm by acting as a 60S anti-association factor that blocks binding to 40S subunits (Brina et al., 2011). Robust depletion of either eIF6 or RACK1 results in reduced levels of 60S subunits and impaired 80S assembly across mammals, yeast and plants (Guo et al., 2011; Sanvito et al., 1999; Thompson et al., 2016). By contrast, moderately smaller reductions in either eIF6 or RACK1 levels, as occurring e.g. in heterozygous knockout mice, do not impact 60S levels but, instead, cause an increase in 80S ribosome levels (Gandin et al., 2008; Volta et al., 2013). Albeit the underlying reasons for this seemingly paradoxical effect remain unknown, this dose-dependent functional parallel suggests that RACK1 loop mutations act in a dominant-negative manner in this pathway.
Although RACK1 is known to bind eIF6, how it does so and how this interaction is regulated remains poorly understood. To test whether the RACK1 loop is involved, we compared eIF6 binding across both high and moderately transduced NHDFs. Beginning with high transduction lines, the use of GFP-Trap Sepharose to recover GFP-tagged forms of RACK1 from cell extracts showed that binding of eIF3c – which is mediated by the β-propeller blades of RACK1 (Kouba et al., 2012) – was unaffected by loop mutations and simply reflected the level of RACK1-eGFP recovery across samples (Fig. 3B). Notably, the degree to which eIF3c was enriched in bound versus input samples was similar to that of RACK1 enrichment, suggesting that a large fraction of RACK1 binds a substantial fraction of the cellular eIF3c pool. This is in line with the notion that RACK1 functions predominantly as a 40S ribosomal subunit that interacts with initiation factors, and that these interactions are unaffected by loop mutations. By contrast, although WT RACK1 bound to eIF6, eIF6 was not enriched in bound versus input samples (Fig. 3B). This suggests that only a small fraction of RACK1 binds eIF6, again in line with evidence that RACK1 has only limited functions off the 40S ribosome in many contexts. However, negatively charged (STSS-EEEE), uncharged (STSS-AAAA) or deletion (ΔSTSS) mutants of RACK1 caused a large increase in eIF6 binding, whereas changes caused by the yeast FAGYS motif were minimal (Fig. 3B). Similar increases in eIF6 binding over WT RACK1 were observed for high transduction lines expressing either S-E or TS-DE mutants of human RACK1, or for human RACK1 harboring the dicot plant loop (Fig. 3B). Notably, although the TS-DE mutant did not reach statistical significance due to broader variability in the magnitude of binding, it nonetheless bound eIF6 in a similar range and exerted the same effects on 80S levels as other mutants. As such, across all RACK1 variants tested, a direct correlation was seen between high levels of eIF6 binding, and defects in 60S and 80S levels in high transduction NHDF pools. To test this further, we performed the same binding assay in NHDFs that had been moderately transduced and that did not exhibit reductions in 80S assembly (Fig. 3A). Remarkably, whereas eIF3c was, again, enriched in bound fractions and its recovery, again, reflected the relative recovery of RACK1-eGFP across each sample, eIF6 binding was not readily detectable in any of these lines (Fig. 3C). Inclusion of the ΔSTSS mutant from high transduction lines alongside the same ΔSTSS mutant from moderate transduction lines revealed the striking and specific difference in eIF6 compared with eIF3c binding, which was caused by differences in transduction efficiencies (Fig. 3C). As such, a correlation between eIF6 binding and the effects on 80S levels was found across multiple mutants and transduction levels. This suggested that only a small fraction of the total pool of RACK1 associates with eIF6. In line with this, immunofluorescence confirmed that, like in many other cell types (Brina et al., 2011; Gallo and Manfrini, 2015), highest levels of RACK1 are found in the cytoplasm of NHDFs, with small amounts in the nucleus. Highest eIF6 levels, however, are present in the nucleus with smaller fractions in the cytoplasm (Fig. S2B,C). Considering these findings alongside evidence that RACK1 has high affinity to and residence time on the 40S ribosome with only limited extra-ribosomal functions, we propose the following model from our observations (Fig. 4A-C): in non-transduced cells, most RACK1 is present on the 40S subunit, whereas only relatively small fractions interact with eIF6 (Fig. 4A). This small subpopulation of RACK1 influences the function of eIF6 in 60S biogenesis and its release from 60S ribosomes to allow 80S assembly. Both human and yeast loops are functional in this regard, while plant – as well as other charged or uncharged loop mutations – bind tightly to eIF6 and impair its function, resulting in reduced 80S levels (Fig. 4B). Tight binding and sequestration of factors is a common regulatory process even in the area of translational control, exemplified by how limiting amounts of the eIF2 guanine nucleotide exchange factor eIF2B are sequestered through higher affinity binding by small amounts of the phosphorylated form of its own substrate eIF2 (Wek, 2018). However, given that this is only a function for a small fraction of RACK1, even very small reductions in the level of exogenous RACK1 will quickly limit effects on eIF6 as most RACK1 is bound to the 40S ribosome (Fig. 4C) (Johnson et al., 2019; Nielsen et al., 2017; Sengupta et al., 2004). Although not the predominant function of RACK1, our findings reveal that a secondary function of the RACK1 loop is to regulate interactions with eIF6, and that loop sequences are differently optimized to accomplish this activity across species.
Negative charge in the RACK1 loop enhances translation of 5′poly(A) mRNAs
On the basis of our earlier observation that an STSS-EEEE loop mutant caused shifts in ribosome profiles, we originally hypothesized that this reflects changes in scanning behavior of the ribosome to offset ‘sliding’ (Jha et al., 2017), a phenomenon wherein poly(A) stretches cause initiating 40S or elongating 80S ribosomes to lose processivity, i.e. forward directionality (Arthur et al., 2015; Koutmou et al., 2015; Shirokikh and Spirin, 2008). Moreover, single or clustered charges, as well as spaced plant loop configurations, function analogously to enhance translation from poly(A) reporters (Jha et al., 2017). However, our data revealed that these altered ribosome profiles reflect changes in eIF6 interactions, are not unique to negatively charged RACK1 loops and can be largely bypassed through reduced transduction. We, therefore, used these reduced transduction NHDFs to minimize the influence of effects on eIF6 and to test the functionality of loops from species known to use poly(A) leaders, namely the dicot plant A. thaliana, versus those that do not use these leaders, namely H. sapiens or S. cerevisiae loops (Figs 1C and 5). As an additional control for potential contributions from effects on eIF6, rather than negative charge alone, we also included the ΔSTSS mutant that binds eIF6 similarly to charged mutants. 35S-metabolic pulsing and western blot analysis showed that steady-state protein synthesis and the levels of several cellular proteins were not affected by any of these loop modifications compared with a WT human loop (Fig. 5A). However, the plant loop chimera significantly enhanced luciferase production from a 5′ poly(A) reporter (Fig. 5B). By contrast, this enhancement was neither observed with the yeast chimera nor with the ΔSTSS loop mutant, suggesting that translational enhancement is specific to the negatively charged loop of A. thaliana and largely independent of any secondary effects on eIF6.
This finding established that the negative charge found in dicot plant loops is important for enhancing translation of mRNAs with poly(A) leaders. In order to explore how this enhancement is facilitated, we examined the potential effects of the negative charge on RACK1 contacts with the 40S ribosome. RACK1 sits near the mRNA exit channel with the loop facing away from the cytosol and towards the 18S ribosomal RNA (rRNA) (Figs 1A,B and 6A) (Coyle et al., 2009; Sengupta et al., 2004). The RACK1 loop is not required for ribosome binding in yeast (Coyle et al., 2009) and our data, after using various mutants, support this notion for human cells (Figs 2A and 3A). Therefore, to test whether a negative charge might generate electrostatic forces against the phosphate backbone of 18S rRNA, we homology-modeled the consequences of varying loop charges by using human RACK1. We then measured interatomic clashes without performing energy minimization to preserve electrostatic interactions of potential biological significance. Although there were no electrostatic clashes between the uncharged STSS motif of WT RACK1 and the 18S rRNA, the S-E substitution mutant, which mimics single-site phosphorylation through VacV or singly charged residues present in other species, resulted in 15 clashes (Fig. 6A). This increased to 24 clashes for an STSS-EEEE mimetic of the far less prevalent, clustered charge organization present in some protists. However, despite containing three negatively charged residues, the A. thaliana loop only generated 15 clashes, like the S-E substitution mutant. Overall, this modeling suggests that the number and spatial organization of charged residues controls the degree of electrostatic force between RACK1 and the 18S rRNA.
To test these predictions biochemically, we examined how several GFP-tagged RACK1 loop mutants associated with other small subunit proteins in binding assays. To do this, RACK1-eGFP was isolated from cell extracts by using GFP-Trap Sepharose followed by western blot analysis of the recovery of RPS3a and RPS10. Compared with an unmodified WT human loop, and in line with modeling predictions, the S-E mutant or a chimeric human RACK1 with an A. thaliana loop exhibited similar, modest reductions in association (Fig. 6B). Effects on RPS3a were not statistically significant while effects on RPS10 showed some degree of significance. By contrast, both TS-DE and STSS-EEEE mutants resulted in a larger, statistically significant reduction in recovery of both RPSs. This suggests that a clustered charge, even as few as two consecutive residues, greatly increases electrostatic forces on the ribosome and may explain, in part, the single or spaced charge configurations favored by most organisms. It is important to note that these assays employ buffer conditions that differ from those used for ribosome sedimentation assays, and results do not mean that RACK1 is released from the ribosome. Rather, they simply reveal differences in the strength of RACK1 contacts with the ribosome that arise from electrostatic forces. We confirmed this in two ways. First, negatively charged mutants do not become detectable in the free fractions in ribosome sedimentation assays (Fig. S3). Second, the most-severe defects in RPS recovery associated with the extremely clustered charge mutant STSS-EEEE could be readily reversed simply by changing buffer conditions to those used for ribosome sedimentation (Fig. 6C). In addition, recent biochemical studies of interactions between RACK1 and the ribosome also examined the S-E mutant and, independently, found that it remains bound to the 40S ribosome (Johnson et al., 2019). Further in line with the notion that a negative charge generates electrostatic forces on the 40S subunit – rather than the loop structure itself being required for binding, uncharged STSS-AAAA or ΔSTSS mutants, as well as the uncharged yeast chimera exhibited no significant defects in RPS association when using these assays (Fig. 6D). Overall, these findings suggest that on the ribosome, electrostatic forces are increased by a negatively charged RACK1 loop that is likely to remodel the loop or the mRNA exit channel in order to accommodate the unusual structures adopted by poly(A) leaders.
RACK1 is emerging as a key regulator of ribosome activity and specificity towards certain transcripts (Gallo and Manfrini, 2015). Although it consists of several widely studied β-propeller domains that mediate protein–protein interactions, a recent study by us provided initial clues to the potential importance of the understudied loop region of RACK1, based on its role in VacV translation (Jha et al., 2017). Here, we reveal the broader multifunctionality of the RACK1 loop in regulating two distinct aspects of translation, together with evidence that its sequence plasticity underlies functional diversification of ribosomes in different organisms.
Although RACK1 is known to bind eIF6 across diverse species (Ceci et al., 2003; Guo et al., 2011), how these two proteins functionally couple remains poorly understood. We found that a small fraction of RACK1 binds eIF6 and this interaction is regulated by the RACK1 loop domain. Charged, uncharged and deletion mutants suggest that the loop is unlikely to be the direct binding site but, rather, that it regulates the strength or dynamics of interactions with eIF6. A diverse range of experimental mutations, independently of their charge status and down to single amino acid changes, affected eIF6 interactions and 80S levels. This suggests that, despite its sequence plasticity, the RACK1 loop is intolerant of random changes in amino acid composition. Paradoxically, the functionality of a yeast loop in human cells demonstrates that some mammalian loops are, indeed, interchangeable despite considerable sequence diversity, suggesting that they evolved precise – yet quite different – loop sequences optimized to this activity. However, loops from organisms of more-distant kingdoms, such as plants, appear to be too evolutionarily divergent in this regard. Notably, eIF6 in higher organisms is phosphorylated by PKCβ, but is not in plants and most fungi. Moreover, plants encode two forms of eIF6 that have diverged from their mammalian counterparts (Brina et al., 2011). It is tempting to suggest that sequences in the RACK1 loop coevolve with changes in eIF6 or vice versa to retain functionality of this pathway that is conserved across plants and mammals. As such, changes in eIF6 might be one evolutionary driver of the sequence diversity in the RACK1 loop across species. Functionally, RACK1 loop mutants exhibited dose-dependent effects that closely mirrored those reported for varying degrees of eIF6 or RACK1 depletion, namely reductions of 60S subunits and 80S ribosomes with more-robust depletion or increases of 80S subunits with more-moderate depletion (Gandin et al., 2008; Guo et al., 2011; Sanvito et al., 1999; Thompson et al., 2016; Volta et al., 2013). This suggests that RACK1 loop mutants act in a dominant-negative fashion, most likely by binding too tightly to eIF6, thereby affecting its functions in ribosome biogenesis and 80S assembly pathways. Interestingly, despite this fact, we did not observe any notable effects on steady-state translation in NHDFs with high or moderate transduction levels. Similar observations were made with eIF6 or RACK1 perturbations in other systems. This suggests that changes in ribosome levels do not dramatically impact steady-state translation in some cell types, at least in those with a low protein synthesis demand, such as NHDFs. Alternatively, these seemingly dramatic changes to ribosome profiles might also, in part, reflect changes in 80S assembly kinetics that do not significantly impact steady-state protein synthesis. However, loss of RACK1 or eIF6 functionality does impair translational upregulation in response to certain stimuli as well as translation of specific subsets of mRNAs (Brina et al., 2011; Ceci et al., 2012; Gandin et al., 2008; Núñez et al., 2009, 2010; Rachfall et al., 2013; Ruan et al., 2012; Shor et al., 2003; Thompson et al., 2016; Volta et al., 2013). Thus, RACK1's control of eIF6 through the loop region may contribute to this form of transcript discrimination in some biological contexts. However, the dominant-negative effects of charged loop mutants would be expected to impair, not enhance, translation of transcripts controlled by this pathway. Indeed, under the moderate transduction conditions and the ΔSTSS mutant used by us, we suggest that contributions from eIF6 are limited in regulating translation of mRNAs with poly(A) leaders.
Most higher organisms select against poly(A) in either 5′UTR or coding regions (Arthur et al., 2015; DiGiuseppe et al., 2018; Koutmou et al., 2015; Xia et al., 2011). This is due, at least in part, to the fact that poly(A) stretches ≥11 nucleotides induce sliding of both initiating 40S and elongating 80S ribosomes (Arthur et al., 2015; Koutmou et al., 2015; Shirokikh and Spirin, 2008), a phenomenon that contributes to the specific function of the 3′poly(A) tail in mRNA quality control (Joazeiro, 2017). It is interesting that, amongst reported functional studies of poly(A) leaders, these elements are generally absent and lack enhancer activity in human or yeast cells (Dhungel et al., 2017; Jha et al., 2017; Xia et al., 2011), both of which have no negative charge in their RACK1 loop and whose loops – our assays functionally validated – lack poly(A)-enhancer activity. By contrast, many 5′UTRs of the dicot plant A. thaliana harbor long poly(A) stretches (Guo et al., 2016). Moreover, A-rich sequence motifs were recently identified in the 5′ leaders of a motif containing almost exclusively purines (R-motif) in translationally upregulated mRNAs during A. thaliana responses to microbial peptides (Xu et al., 2017). In addition, plant viruses encode adenosine-rich enhancers, such as the tobacco mosaic virus (TMV) Ω-leader that functions analogously to poxvirus poly(A) leaders. However, the Ω-leader primarily works in natural dicot hosts of the TMV, including N. tabacum, and functions poorly in other species (Gallie et al., 1988). Transcripts of the protist D. discoideum, which also harbors a multiple, spaced charge RACK1 loop configuration, also contain a consensus motif of 5′ poly(A)≥15 nucleotides whose functional significance is unknown (Steel and Jacobson, 1991). Among the few studies of its function, mutations within the RACK1 loop impair D. discoideum growth and development (Omosigho et al., 2014); but, precisely how the loop contributes to these processes remains unclear. Our findings suggest that a negatively charged RACK1 loop enables these species to utilize poly(A) leaders more efficiently. It would be interesting to determine whether other species that harbor a negatively charged RACK1 loop also utilize adenosine-rich leaders and whether species that lack loop charge do not. However, this would require extensive functional studies across a wide range of species and might, ultimately, be complex to interpret. For example, the use of poly(A) leaders may be restricted to specific conditions, such as immune responses in dicot plants (Xu et al., 2017), while some species may utilize alternative strategies to RACK1 loop charge. Other species may simply not exploit the regulatory potential of their charged RACK1 loop, or may not have evolved the leader sequences to do so yet. Regardless, our data show that a negative charge in the RACK1 loop represents one strategy to enhance translation of 5′ poly(A) mRNAs and closely correlates with reports of their use, or lack thereof, across a number of species. The importance of this strategy is further reinforced by the fact that poxviruses evolutionarily ‘reintroduced’ a RACK1 loop charge to mammalian cells through phosphorylation to enhance translation of their 5′ poly(A) mRNAs (Jha et al., 2017).
Finally, our data regarding the effects of loop charge on RACK1 interactions with the ribosome offer explanations on how it might function and why species overwhelmingly choose single or spaced charge configurations. Modeling electrostatic interactions and biochemical analyses suggest that negative charge generates electrostatic forces that are likely to either restructure the loop itself or, more generally, remodel the exit channel. This would potentially better accommodate the unusual helical structures known to be adopted by adenosine-rich RNA elements (Kovtun et al., 2007), which cause ribosome sliding (Arthur et al., 2015; Koutmou et al., 2015; Shirokikh and Spirin, 2008). Although these structural and mechanistic details remain to be determined, our analyses also offer an explanation as to why most organisms avoid clustered charges. Beyond retaining functionality with eIF6 pathways in their host, a second evolutionary pressure on charge organization within the RACK1 loop might exist to minimize the extent of electrostatic forces on the 40S ribosome that alter RACK1 affinity – providing an optimal balance of ribosome binding, while retaining negative charge that enables enhanced translation of 5′ poly(A) mRNA. While these evolutionary concepts require further experimental testing, our findings here reveal how sequence plasticity across the poorly studied loop of RACK1 plays species-specific roles in two distinct translational control pathways.
MATERIALS AND METHODS
Human dermal fibroblasts from neonatal foreskin (NHDF-Neo) were purchased from Lonza Walkersville, Inc. (CC-2509) and are fully characterized by the supplier. 293T cells used to generate viruses were obtained from Dr Mojgan Naghavi, Northwestern University. NHDF and 293T cells were cultured in Dulbecco's modified Eagle's medium (DMEM; MT15013CV, Fisher Scientific) supplemented with 2 mM L-glutamine, 1× penicillin-streptomycin and 5% fetal bovine serum (FBS) and maintained at 37°C, 5% CO2. Lentivirus vectors for RACK1-eGFP expression were produced by co-transfection of 293T cells with pLVX-hygro-based plasmids, described below, together with p8.91 (gag-pol) and p-VSV-G (envelope). Supernatants containing lentivirus were then filtered and used to transduce NHDFs. NHDFs stably expressing RACK1-eGFP forms were generated by selection using 100 µg/ml hygromycin. For high transduction Polybrene was used. For moderate transduction, Polybrene was not used. All cell lines were routinely checked for mycoplasma by DNA staining and imaging.
Plasmids and cloning
Sub-cloning of human receptor of activated protein C kinase 1 (RACK1) comprising a C-terminal eGFP tag into the pLVX-IRES-hygromycin plasmid, as well as generation of S-E and STSS-EEEE mutants, and the A. thaliana plant loop chimera (VISTSS>LKAEAEKADNSGPAAT) forms of RACK1-eGFP have been described previously (Jha et al., 2017). The ΔSTSS and STSS-AAAA mutants, and S. cerevisiae yeast loop chimera (VISTSS>FAGYS) forms of RACK1-eGFP were created by Gibson cloning using Gibson assembly master mix (NEB Biolabs) and the following gBLOCK DNA fragments (Integrated DNA Technologies). Note, uppercase letters within the gBLOCK represent the mutations made to the STSS motif. The uppercase letters at the 3′ end of the gBLOCK represent introduction of a BamHI restriction site (GGATCC) and the inclusion of an 18 nt overhang for cloning. Lowercase letters indicate the original human RACK1 sequence. ΔSTSS: 5′-gtgactgtctctccagatggatccctctgtgcttctggaggcaaggatggccaggccatgttatgggatctcaacgaaggcaaacacctttacacgctagatggtggggacatcatcaacgccctgtgcttcagccctaaccgctactggctgtgtgctgctacaggccccagcatcaagatctgggatttagagggaaagatcattgtagatgaactgaagcaagaagttatcaaggcagaaccaccccagtgcacctccctggcctggtctgctgatggccagactctgtttgctggctacacggacaacctggtgcgagtgtggcaggtgaccattggcacacgcGGGGTACCGCGGGCCCGGGATCCACCGGTCGCCACCatggt-3′. STSS-AAAA:5′-gtgactgtctctccagatggatccctctgtgcttctggaggcaaggatggccaggccatgttatgggatctcaacgaaggcaaacacctttacacgctagatggtggggacatcatcaacgccctgtgcttcagccctaaccgctactggctgtgtgctgctacaggccccagcatcaagatctgggatttagagggaaagatcattgtagatgaactgaagcaagaagttatcGCTGCCGCTGCCaaggcagaaccaccccagtgcacctccctggcctggtctgctgatggccagactctgtttgctggctacacggacaacctggtgcgagtgtggcaggtgaccattggcacacgcGGGGTACCGCGGGCCCGGGATCCACCGGTCGCCACCatggt-3′.
S. cerevisiae loop chimera (VISTSS>FAGYS): 5′-gtgactgtctctccagatggatccctctgtgcttctggaggcaaggatggccaggccatgttatgggatctcaacgaaggcaaacacctttacacgctagatggtggggacatcatcaacgccctgtgcttcagccctaaccgctactggctgtgtgctgcTacaggccccagcatcaagatctgggatttagagggaaagatcattgtagatgaactgaagcaagaaTTCGCCGGCTACAGCaaggcagaaccaccccagtgcacctccctggcctggtctgctgatggccagactctgtttgctggctacacggacaacctggtgcgagtgtggcaggtgaccattggcacacgcGGGGTACCGCGGGCCCGGGATCCACCGGTCGCCACCatggt-3′.
gBLOCK DNA fragments were digested with BamHI and ligated into BamHI-digested peGFP-N1-RACK1 (Addgene, plasmid #41088). Insertions were confirmed by sequencing and used as templates for PCR amplification and subcloning into pLVX-IRES-Hygromycin, using the following primers and SpeI and NotI digestion: RACK1 Fwd SpeI 5′-AAAAAACTAGTCTCAAGCTTATGACTGAGCAGATG-3′, RACK1 Rev NotI 5′-AAAAAGCGGCCGCTTACTTGTACAG-3′.
The TS-DE mutant was generated by site-directed mutagenesis using the following primers: RACK1 TS277-278DE FWD: 5′-GAAGTTATCAGTGATGAAAGCAAGGCAG-3′, RACK1 TS277-278DE REV: 5′-CTGCCTTGCTTTCATCACTGATAACTTC-3′.
The luciferase reporter plasmid was generated by modifying the vector, pQCXIN (Clontech) to replace the linker region between the transcription start site and open reading frame with specific UTRs. First, pQCXIN was digested with XbaI and EcoRI to remove the linker between the CMV promoter/TSS and major cloning site, and replaced with the following sequence in the form of a gBlock (the TSS and AgeI site used to insert reporters are underlined): 5′-agatctgggggatcgatcctctagagtccgttacataacttacggtaaatggcccgcctggctgaccgcccaacgacccccgcccattgacgtcaataatgacgtatgttcccatagtaacgccaatagggactttccattgacgtcaatgggtggagtatttacggtaaactgcccacttggcagtacatcaagtgtatcatatgccaagtacgccccctattgacgtcaatgacggtaaatggcccgcctggcattatgcccagtacatgaccttatgggactttcctacttggcagtacatctacgtattagtcatcgctattaccatggtgatgcggttttggcagtacatcaatgggcgtgaatagcggtttgactcacggggatttccaagtctccaccccattgacgtcaatgggagtttgttttggcaccaaaatcaacgggactttccaaaatgtcgtaacaactccgccccattgacgcaaatgggcggtaggcgtgtacggtgggaggtctatataagcagagctcgtttagtgaaccgtcagatgcggccgcaccggtaggcctcgtacgcttaattaacggatccggaattc-3′.
Poly(A)-containing 5′UTR elements ahead of the Firefly luciferase (FLuc) ORF were inserted by transferring these elements from our original pBABE-puro-based FLuc reporters (Jha et al., 2017) using AgeI and BamHI digestion followed by ligation into the modified pQCXIN vector. This resulted in the following leader ahead of the FLuc ORF, with the TSS and ATG start sites underlined: 5′-tcagatgcggccgcaccggtAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAcatatg-3′.
All plasmid inserts were verified by sequencing at NUSeq Core Facility, Northwestern University or ACGT, Inc.
Sucrose gradient centrifugation and isolation of eGFP-tagged RACK1
Sucrose gradient centrifugation and polysome analysis was performed as described recently (Jha et al., 2017), loading lysates based on equal amounts of RNA (RNA levels were also found to be equal by cell number). Please also note that Lysate samples were not TCA precipitated and for this reason, differences in buffers causes RACK1-eGFP to migrate differently to RACK1-eGFP in precipitated fractions. To isolate RACK1-eGFP complexes, 6-cm or 10-cm dishes of confluent NHDFs were washed with ice cold PBS and collected in lysis buffer [50 mM HEPES pH 7.4, 1.5 mM MgCl2, 2 mM EDTA, 1.6 mM Na3VO4, 25 mM glycerophosphate and complete mini EDTA-free protease inhibitor cocktail (Roche)]. To rescue binding of the STSS-EEEE mutant as shown in Fig. 6C, the same assays were performed using lysis buffers from polysome approaches (Jha et al., 2017). To detect eIF6 in RACK1-eGFP complexes, 50 mM NaCl was used in lysis buffers. After 40-80 min rocking at 4°C, lysates were clarified by centrifugation and incubated for 4 h with GFP-Trap_A beads (ChromoTek). Beads were then washed with lysis buffer before elution of bound protein by boiling in Laemmli buffer.
Luciferase assays and qRT-PCR analysis
For luciferase assays or RNA analysis, 10 cm dishes of NHDFs stably expressing the RACK1-eGFP were electroporated with 2 µg pQCXIN reporter plasmid. For Luciferase assays, immediately following electroporation the cell suspension was seeded onto 12-well plates. 24 h post transfection, cultures were washed with PBS and lysed with 200 µl Luciferase Cell Culture Lysis Reagent (Promega). Lysates were clarified by centrifugation at 10,000 g for 2 min. 20 µl supernatant was added to 96-well plates and luminescence was measured using either a Spectramax (Molecular Devices) or a CLARIOstar microplate reader (BMG Labtech). For qRT-PCR analysis, immediately following electroporation cells were seeded onto 6-well plates. 24 h post transfection, cells were harvested and total RNA was isolated using Trizol (Thermo Fisher Scientific). 10 µl was reverse transcribed using Transcriptor First cDNA synthesis kits (Roche Life Science) or RevertAid First Strand cDNA Synthesis kits (Thermo Fisher Scientific). Primers used were described and confirmed for specificity previously (Jha et al., 2017). Quantitative Real-Time PCR (qRT-PCR) was performed using PowerUp SYBR Green Master Mix (Applied Biosystems) on a 96-well plate using 7500 Fast Real-Time PCR System (Applied Biosystems). Absolute quantification of firefly luciferase reporter RNA was performed using the standard curve method. The pQCXIN poly(A) reporter plasmid linearized using XhoI (NEBiolabs) was used as the template for the standard curve. 115 ng of sample cDNA was used as the template for the reactions.
Western blotting was performed as described previously (Jha et al., 2017; Walsh and Mohr, 2006). Briefly, whole cell lysates were resolved by SDS-PAGE and transferred to nitrocellulose membranes that were blocked for by incubation for 1 h at room temperature in a solution of 5% non-fat milk in Tris-buffered saline (TBS) and 0.1% Tween. After rinsing in TBS-T, membranes were incubated overnight at 4°C in primary antibody diluted in 3% bovine serum albumin in TBS and 0.1% Tween. Antibodies used were from Cell Signaling Technologies; RACK1 (Cat# 5432S), RPL11 (Cat# 14382S), RPL13a (Cat# 2765S), anti-eIF3A (#3411), anti-eIF3C (#2068), β-actin (Cat# 3700S), anti-eIF6 (#D16E9); Abcam; RPS10 (Cat# ab151550), RPS3a (Cat# ab174894), Millipore Sigma; α-tubulin (Cat# ab18251), Santa Cruz Biotechnology; GAPDH (#sc-25778), BD Transduction Laboratories; eIF4E (#610270). Rabbit anti-eIF4G and anti-PABP was described previously (Walsh and Mohr, 2006). Primary antibodies were used at 1:1000 dilution except for β-actin (1:10,000), RPS3a (1:500), RPS10 (1:500), PABP (1:5000), α-tubulin (1:4000) and eIF4G (1:5000). Next day, membranes were washed and incubated with the appropriate HRP-conjugated secondary antibody. After washing, detection was performed using Pierce ECL western blot substrate (Thermo Fisher) followed by exposure to X-ray film. Western blots were quantified using densitometry as described previously (Procter et al., 2018).
Immunofluorescence was performed as described previously (Jha et al., 2017; Meade et al., 2018a). Briefly, NHDFs were seeded on glass coverslips and fixed 24 h later in 4% formaldehyde in PBS (Affymetrix) for 20 min at room temperature. Samples were then washed in PBS and blocked in a mixture of 10% FBS and 0.25% saponin in PBS. Samples were incubated with anti-eIF6 antibody (Cell Signaling, #D16E9) in antibody dilution buffer (10% FBS and 0.025% saponin in PBS) overnight at 4°C. Primary antibody was then removed and samples were washed in PBS containing 0.025% saponin and detected using alexa-647-conjugated donkey anti-rabbit secondary for 1 h. After washing, samples were stained with Hoechst and imaged using a Leica DMI6000 wide-field microscope configured as described previously (Meade et al., 2018a). RACK1-eGFP was imaged directly.
Metabolic labeling was performed by incubating cultures in methionine/cysteine-free DMEM (17-204-CL, Corning) supplemented with 17.5 µCi 35S-methionine/cysteine (NEG072, Amersham) for 30 min before lysis in Laemmli buffer.
Phylogenetic analysis and structure modeling
The UniProt Basic Local Alignment Search Tool (BLAST) was used to search the UniProtKB database and identify known or expected RACK1 homologues. H. sapiens, D. melanogaster, S.cerevisiae, A. thaliana, Phytophthora parasitica and D. discoideum RACK1 sequences were used as templates to retrieve sequences for vertebrate and invertebrate animals, yeast, plants, stramenopiles and amoeboids, respectively. For the search parameters, the E-threshold was set at 10. The search results were manually filtered for non-RACK1 sequences and duplicates, such that there would only be one sequence per species. In the cases that there were multiple sequences returned for a single species, only the longest and/or most recently updated sequence was retained. After filtering, the remaining 979 sequences were aligned using Clustal Omega and further analyzed using ESPript 3.0 (http://espript.ibcp.fr/ESPript/ESPript/index.php). All conserved residues are highlighted in black and similar residues in bold. Loop regions were manually analyzed for the presence of negative charge and specific charge organizations.
The NCBI common tree generator was used to produce the cladogram, which was further processed and annotated using Evolview (https://www.evolgenius.info//evolview/#login).
All structures analyzed were obtained from the Protein Data Bank (PDB) and visualized using UCSF Chimera (https://www.cgl.ucsf.edu/chimera/). For the structures generated in Fig. 1, all unstructured regions of human RACK1 (PDB: 4AOW) were built with the standard loop modeling protocol of UCSF Chimera interface to MODELLER (https://salilab.org/modeller/), using the default settings. One model was generated for display. Propeller blades were color coded by WD repeats and numbered as done by McCahill et al. (2002).
Clash modeling of the RACK1 charged mutants was performed with the comparative modeling tool of the MODELLER interface in UCSF Chimera, using the default settings. RACK1 present in the cryo-EM structure of the 80S human ribosome (PDB: 5T2C) was used as the template. The target sequences modeled included the single site (S-E) and quadruple-site (STSS-EEEE) phosphomimetics, and the A. thaliana plant loop (VISTSS>LKAEAEKADNSGPAAT) chimera. Of the five models generated, the structure that contacted the 18S rRNA phosphate backbone was selected for display. Interatomic clashes between loop motifs and the 18S rRNA were calculated with the FindClashes/Contacts tool in UCSF Chimera, using the default settings. Values presented in the table in the main figure represent the charged residue of the loop motif that makes contacts with the 18S rRNA phosphate backbone.
GraphPad Prism 7 software was used for each analysis. Results are displayed as the mean±standard error of the mean (±s.e.m.). For comparisons, one- or two-way analysis of variance (ANOVA) followed by either a Dunnett's multiple comparison or Sidak's multiple comparison post-hoc test was performed to determine the statistical significance (ns, P>0.05; *P≤0.05; **P≤0.01; ***P≤0.001 and ****P≤0.0001). Additional statistical details can be found in the figure legends.
Conceptualization: M.G.R., S.J., E.T.B., D.W.; Methodology: M.G.R., S.J., E.T.B.; Formal analysis: M.G.R., D.W.; Investigation: M.G.R., S.J.; Resources: S.J.; Writing - original draft: M.G.R., D.W.; Writing - review & editing: S.J., D.W.; Visualization: E.T.B.; Supervision: D.W.; Project administration: D.W.; Funding acquisition: E.T.B., D.W.
This work was supported by grants from the National Institutes of Health (NIH) (grant no. R01AI127456 to D.W. and grant no. R50CA221848 to E.T.B.). M.G.R. was supported by training grant T32 GM008061. Deposited in PMC for release after 12 months.
The authors declare no competing or financial interests.