The Modular Adaptive Ribosome

The ribosome is an ancient machine, performing the same function across organisms. Although functionally unitary, recent experiments suggest specialized roles for some ribosomal proteins. Our central thesis is that ribosomal proteins function in a modular fashion to decode genetic information in a context dependent manner. We show through large data analyses that although many ribosomal proteins are essential with consistent effect on growth in different conditions in yeast and similar expression across cell and tissue types in mice and humans, some ribosomal proteins are used in an environment specific manner. The latter set of variable ribosomal proteins further function in a coordinated manner forming modules, which are adapted to different environmental cues in different organisms. We show that these environment specific modules of ribosomal proteins in yeast have differential genetic interactions with other pathways and their 5’UTRs show differential signatures of selection in yeast strains, presumably to facilitate adaptation. Similarly, we show that in higher metazoans such as mice and humans, different modules of ribosomal proteins are expressed in different cell types and tissues. A clear example is nervous tissue that uses a ribosomal protein module distinct from the rest of the tissues in both mice and humans. Our results suggest a novel stratification of ribosomal proteins that could have played a role in adaptation, presumably to optimize translation for adaptation to diverse ecological niches and tissue microenvironments.


INTRODUCTION 45
A single celled organism displays a range of phenotypes to survive in diverse environments. 46 In complex multicellular organisms, in addition to the external environment, tissue specific 47 6 Dharmacon Inc.). Additional double-gene deletions were generated using a previously 149 described protocol [30]. SK1 wild type and deletion strains were obtained from Wilkening et 150 al. [31]. Strains were phenotyped in YPD, YPD+Menadione (50µM) and YPD+CdCl 2 151 (10µM). Spot dilutions, ranging from 10 -3 to 10 -8 dilutions were incubated at 30ºC and 152 phenotyped at 24, 36 and 48h. The strain and primer list is given in S4 Table. 153

Sequence analysis of coding and 5'UTR regions in SGRP collection 154
S. cerevisiae and S. paradoxus ribosomal protein and control gene sequences from the SGRP 155 strains [32] were downloaded from http://www.moseslab.csb.utoronto.ca/sgrp/blast_original. 156 Sequence alignments, estimation of the maximum likelihood tree (1,000 permutations) and 157 nucleotide diversity were performed using MEGA 6.06 [33] with default parameters (S5 158   Table). 159 For the three gene clusters based on the interaction modules, the nucleotide diversity in the 160 coding and promoter regions of the genes in each cluster was evaluated using Shannon 161 Entropy function as a metric. The sequences of the 5'UTR and coding regions of each gene 162 corresponding to the different strains were aligned separately using the software MUSCLE 163 [34] with default parameters. In the aligned set of sequences for each gene, the mutated sites 164 were identified, and corresponding to each kind of base at the given site, were assigned a 165 value, 166 p i = Number of sequences which base of kind i present at given site Total number of sequencesin the aligned set 167 Here i = 1, 2, 3, 4 corresponds to A, T, G, C respectively. 168 The Shannon Entropy at each mutation site was computed using the standard definition: 169 Variations within a given sequence set can occur in two principal ways: (i) variations in the 171 proportion of bases at any given mutation site, and (ii) variations in the number of sites of 172 mutations. The simplest quantity that accounts for both these is the sum of the H values for 173 all the mutation sites. A higher value for the sum would indicate greater nucleotide diversity 174 in the corresponding 5'UTR or coding region respectively for each gene across the different 7 strains. To eliminate any length bias in the comparisons, the sum of the H values was 176 normalized by the length of the aligned sequences for each 5'UTR or coding region (S5 177 Table). 178

Prediction of transcription factor binding sites 179
The database YEASTRACT [35] was used to download known transcription factor (TF) 180 binding sites of the various ribosomal proteins and to predict potential binding sites on UTR 181 of ribosomal proteins of different SGRP strains. Of the 700 TFs reported in YEASTRACT,182 216 have been shown to experimentally bind to the promoter region of at least one ribosomal 183 protein present in the three Clusters identified in our study (S6 Table). These 216 TFs were 184 enriched for various signaling pathways and chromatin remodeling complexes (S6 Table), 185 substantiating the enrichment of chromatin remodelers among the positive genetic interactors 186 of ribosomal proteins. Of these 216, the TFs binding exclusively to ribosomal proteins in 187 Cluster A, B and C were identified. 188

Analysis of human and mouse ENCODE and GTEx data 189
Tissue specific count (transcripts per million or TPM) data for human and mouse were 190 downloaded from ENCODE (https://www.encodeproject.org/), and mapped to Entrez genes 191 using annotation packages org.Hs.eg.db and org.Mm.eg.db in R. The genes that were not 192 expressed in any replicate were discarded. Replicates in which an unusually high (as 193 determined from sfigx_h and sfigx_m) fraction of Entrez genes were not expressed were 194 discarded as well. In the remaining replicates, to reduce relative systematic error among 195 replicates, the median for each gene was normalized to unity in each replicate by dividing the 196 count for the gene by the median count in each replicate array. These median adjusted TPM 197 values were log transformed to obtain the final expression X of each gene in each replicate as 198 follows: 199 This normalization ensured that genes that were not expressed at all were mapped to X = 0, 201 and the median of all genes in a replicate was mapped to X = 10. The mean and standard 202 deviation (sd) of expression levels over replicates was computed for each gene-tissue pair to 203 generate a distribution of x = log 10 sd mean ( ) for human and mouse data respectively (Fig   204      8   S1). Based on these distributions cutoffs x h and x m were established and the gene-tissue pair 205 was excluded from further analysis if x was greater x h ~ -0.4 and x m ~ -0.6 for human and 206 mouse data respectively. For the gene-tissue pairs that passed this check, the expression of a 207 gene in a tissue was defined as the mean over replicates. If a gene had to be excluded in many 208 tissues, then that gene was excluded altogether. Sixty-six ribosomal proteins in 110 tissues in 209 humans (S7 Table), and 42 ribosomal proteins in 18 tissues in mice (S8 Table), passed this 210 filter. The R package pvclust was used to perform bootstrapping [36] of hierarchical 211 clustering of ribosomal proteins and tissues in mouse and humans. Similar filtering was 212 performed for GTEx data. GTEx data consists of RNAseq data of 54 tissues from 544 donors 213 amounting to a total of 8,555 samples. Gene-tissue pair with x g > 0.1 were excluded from the 214 data. Seventy-nine ribosomal proteins in 54 tissues passed this filter (S9 Table). 215

Phenotypic variability of ribosomal proteins in yeast 218
In all organisms, the ribosome is a ribonucleoprotein complex composed of two subunits 219 each with an RNA core and large number of ribosomal proteins. In eukaryotes, the 60S large 220 subunit consists of 46 proteins, and the 40S small subunit consists of 33 proteins [37]. In 221 yeast that has undergone whole genome duplication [38], most of the ribosomal proteins 222 (paralogs) are duplicated, as a result of which it contains 137 ribosomal proteins, of which 223 107 are non-essential [22]. 224 Deletion collection in yeast allows testing of phenotypic effect of deletions of non-essential 225 genes in yeast in diverse environments. In order to identify genes which show maximum 226 phenotypic variability across environments, we reanalyzed deletion phenotypes for 4,769 227 single gene deletions grown in 293 diverse environments using a previously published dataset 228 [22] (S1 Table) for all genes, including ribosomal proteins. The surprising observation was 229 that across all yeast genes, deletions of ribosomal proteins had the highest differential effect 230 on growth in different environments i.e., no effect in some environments and strong effect in 231 others. Among the 191 genes with variance σ 2 > 0.8 across the 293 environments, 232 components of the ribosome were significantly enriched (21/191, P < 0.01, S10 Table, Fig  233   S2). Fourteen out of these 21 genes belonged to the large ribosomal subunit. These 21 genes 234 9 contain only one paralog of the ribosomal proteins, either A or B, indicating a possible but 235 small differential role of paralogs in responding to environmental heterogeneity. Only 236 ribosomal protein RPL34 was an exception to this where both paralogs -RPL34A and 237 RPL34B showed high phenotypic variance. 238 To test whether this phenotypic variability was a non-specific cellular effect or was specific 239 to the ribosome, we compared phenotype variability in growth for deletions of genes in 90 240 different pathways and protein complexes across 293 stress conditions versus growth in rich 241 media (YPD). These 90 pathways and complexes were defined using both a biased 242  Table). Differences in variance of a 246 pathway or a complex between stress and YPD indicate variable roles of its constituents in 247 different conditions. High correlation of constituents of a pathway between stress and YPD 248 would indicate that independent of the essentiality of the pathway; different constituents have 249 similar functions in both conditions. Moreover, a higher variance in YPD compared to stress 250 would indicate that the constituents of the pathway show a more diverse response in YPD but 251 show similar phenotype in stress (Fig 1A). Such a co-ordination of stress specific genes has 252 previously been observed [39] in multiple stresses where the whole pathway is essential to 253 respond to the stress. On the other hand, higher variance of the pathway in stress compared to 254 YPD would indicate that different components of the pathway have differential roles in stress 255 and therefore function in a different manner than in YPD ( Fig 1B). Deletions of constituent 256 proteins in 13 pathways showed a significant difference in variance in 3 or more stress 257 environments compared to YPD (P < 0.01 by Brown Forsythe test, S2 Table, Fig S2), with 258 the higher variance in YPD in most cases, showing that the pathway was essential in stress. 259 Additionally, constituent of these pathways showed high correlation of phenotype across 260 YPD and stress indicating that the functional hierarchy of the genes was conserved 261 (functional homogeneity), but the phenotypic contribution of the module increased during 262 stress, reducing phenotypic variance. In contrast, for the cytoplasmic ribosomal proteins, 263 there was a significantly higher variance in 28 stress environments compared to YPD (Fig S2,  264 S2 Table), suggesting that ribosomal proteins are differentially used in the stress condition. 265 Unlike other pathways, poor correlation was observed between phenotype of ribosomal 266 proteins across YPD and stress, indicating overall functional heterogeneity in stress and YPD 267 10 ( Fig 1B). These results independently show that among different pathways, deletions of 268 genes in the ribosomal pathway has the greatest effect on growth in stress versus rich media, 269 suggesting a unique property of the ribosomal genes that they are the most variable proteins 270 in the cell when comparing diverse environments. A heatmap of growth for single deletions 271 of the 68 ribosomal proteins with consistent replicate data in 25 stress conditions and YPD 272 ( Fig 1C) reinforces the above results and shows that a number of these ribosomal proteins 273 have high phenotypic variability, i.e. that they are required for growth in some environments 274 but expendable in others. 275 We next asked whether these ribosomal proteins have different phenotypic profiles across 276 environments i.e., whether they work independently, or whether they form modules, whose 277 constituents show coordinated regulation across different environments. A clustering analysis 278 of the Pearson correlation of these ribosomal proteins across environments showed functional 279 modularity (Fig 2A) in the form of three distinct clusters (S11 Table). Ribosomal proteins in 280 Cluster 1 were both highly correlated, enriched in large subunit proteins and had high 281 phenotypic diversity across environments. On the other hand, proteins in Cluster 2, although 282 highly correlated and important for growth across most environments, were enriched in small 283 subunit and pre-ribosomal components (important for ribosomal assembly), which explains 284 the constitutive growth defect when these proteins were deleted (S10 Table). Proteins in 285 Cluster 3, however, showed low correlation amongst themselves and were important in 286 different environments. The conclusion that emerges from this analysis is that subsets of 287 ribosomal proteins in Cluster 1 act together in diverse environments, whereas proteins in 288 Cluster 2 act together in most environments. Proteins in Cluster 3 on the other hand, seem to 289 play specialized roles in specific environments. 290 These results strongly suggest that the yeast ribosomal proteins do not function in a uniform 291 manner when the environment is varied. While this environmental variability of deletions of 292 ribosomal proteins has previously been observed, we have identified a novel underlying 293 modularity among these ribosomal proteins, potentially to optimize yeast growth in different 294 environments. This is the main finding of the present work that distinguishes it from previous 295 studies. In summary, whereas a core set of ribosomal proteins are important in all 296 environments, different combinations of a subset of variable ribosomal proteins are 297 functional in different environments to optimize growth. 298 11 To test whether the deletion phenotype of ribosomal proteins is conserved among yeast 299 strains, we compared growth of ribosomal protein deletions in a soil isolate, SK1, in an 300 oxidative stress (Fig S3). While different ribosomal protein deletions show diverse 301 phenotypic defects, indicating differential use of ribosomal proteins in SK1, the identity of 302 the variable ribosomal proteins was different among the two strains, SK1 and S288c. This  (Fig 2B) identified three clusters, which had 335 a highly significant overlap with the previous clustering based on phenotypic profiling (90% 336 overlap between Clusters 1 and A and 89% between clusters 2 and B, Fisher Exact test, P = 337 0.006, S4 Table). Ribosomal proteins in Cluster A interacted mainly with genes involved in 338 mRNA processing, whereas those in Cluster B interacted with other ribosomal proteins. 339 Ribosomal proteins in Cluster C interacted with genes involved in diverse pathways (S10 340   Table). This strong overlap of corresponding clusters identified independently through 341 phenotype association (Clusters 1, 2 These are RPL6A and RPL6B, and RPL9A and RPL9B in Cluster A and RPS0A and RPS0B, 349 and RPS29A and RPS29B in Cluster B (S11 Table). All the remaining paralogs fall into 350 separate clusters or are in Cluster C, indicating that they have differential genetic interactions. 351 Thus, in spite of sequence similarity amongst paralogs, the regulation of these modules in 352 yeast seems to have evolved since the duplication event to create novel functions for these 353

paralogs. 354
Positive non-ribosomal interactors of the ribosomal proteins were enriched in chromatin 355 regulators and remodelers (S10 Table). Since ribosomal proteins showed diverse phenotypes in oxidative stresses (Fig 3), we tested 364 the premise of these genetic interactions by phenotyping double deletions of GCN5 and 365 ribosomal proteins in YPD, and oxidative stresses menadione and CdCl 2 . While gcn5∆ 366 rpl38∆ behaved the same as gcn5∆ in menadione, indicating that GCN5 controls cellular 367 proliferation using RPL38, double deletion gcn5∆ rps6b∆ resulted in a 20-fold reduced 368 growth compared to either of the single deletions, indicating parallel or independent roles of 369 both the genes in growth. Furthermore, deletion of the remaining ribosomal proteins (rpl11b∆ 370 and rpl26b∆) rescued the growth phenotype of gcn5∆ (Fig 3). This observed antagonistic 371 effect suggests that a more likely scenario is that these ribosomal proteins have a direct  Table). While the coding sequence nucleotide diversity was similar for both 384 sets, the diversity in the 5'UTR regions within ribosomal proteins was twice that of the 385 control genes (P < 0.005, Fig S4A, S5 Table). Furthermore, in the YEASTRACT database 386 [35], this variability altered the predicted transcription factor binding motifs on ribosomal 387 proteins in diverse strains (S7 Table). Thus, while ribosomal proteins among the SGRP 388 strains have similar coding sequences, their promoter regions have been significantly altered, 389 presumably to adapt to different ecological niches. 390 Coding and 5'UTR regions of ribosomal proteins in the three clusters in Fig 1C were  391 compared using normalized Shannon entropy (S5 Table). All three clusters showed a 392 significant difference in entropy between 5'UTR and coding regions (Fig S4B). Cluster A 393 14 showed significantly high entropy (diversity) for both 5'UTR and coding regions of 394 ribosomal proteins compared to Clusters B and C (Fig S4B). This differential variability 395 shows that the ribosomal proteins in the clusters are evolving at different rates across the 396 SGRP population. 397 Using the YEASTRACT database, we find that whereas most transcription factors (TFs) bind 398 to ribosomal proteins in all three clusters, some are cluster specific. Transcription Factors 399 regulating ribosomal proteins in Cluster A are enriched in the histone deacetylase complex 400 while those that bind to ribosomal proteins in Cluster C are enriched in the HIR (Histone 401 Regulatory) complex (Fig S5, S6 Table). This could be a possible explanation for the 402 different rates of evolution of the 5'UTR regions of the proteins in these three clusters. 403

Modular ribosomal proteins in higher eukaryotes 404
Our above results establish modularity in both phenotype and genetic interactions of 405 ribosomal proteins in yeast. It can be argued that this is merely a unique feature of the yeast 406 ribosome, presumably because of the whole genome duplication event, which might have 407 allowed differential adaptation of duplicated ribosomal proteins. To understand whether the 408 ribosomal modularity observed in yeast extends to higher eukaryotes, we investigated 409 expression levels of ribosomal proteins in mice and humans, which have a single copy of 410 most ribosomal proteins. 411 In complex eukaryotes, the analog of adaptation of unicellular organisms like yeast to 412 different environments is adaptation to different cellular and tissue microenvironments. We 413 therefore expect that if our thesis of the modularity of ribosomal proteins is valid beyond 414 single celled eukaryotes, ribosomal proteins should be differentially used across cell types 415 and tissues in mice and humans. In multicellular systems like humans and mice, ribosomal 416 proteins are present in a single copy, whose deletion results in both cellular and organismal 417 lethality. Our hypothesis for complex eukaryotes would then be that ribosomal proteins are 418 expressed at significantly different levels in different tissue microenvironments in mice and 419 humans. We tested this hypothesis by comparing the expression levels of ribosomal proteins 420 in diverse cell types and tissues in human and mouse samples using RNASeq data for mRNA 421 transcript levels from the ENCODE and the GTEx projects. 422 A total of 66 ribosomal proteins with consistent transcript expression levels across replicates 423 in the ENCODE data were identified in 110 cell types and tissues in humans (see Methods). 424 15 As previously observed [41], we found that while the majority of ribosomal protein 425 transcripts are highly expressed across diverse tissues and cell types, a few showed low 426 expression levels throughout. It has been shown that ribosomal proteins can show differential 427 expression levels based on the proliferation or turnover rate of the cell type. To normalize 428 such global differences, each ribosomal protein within each tissue was assigned a rank based 429 on its expression level (rank 1 for lowest expression and rank 66 for highest expression, S7 430 Table). Despite being a part of the ribosome, it is known that not all ribosomal proteins are 431 equally expressed in a given tissue. In complex eukaryotes, just as in yeast, some of the 432 ribosomal proteins are involved in ribosomal assembly. However, since this is true for all 433 tissues and cell types, their rank normalized expression levels should be consistent across 434

tissues. 435
Hierarchical clustering of all ribosomal proteins expression ranks across all tissues resulted in 436 a single highly correlated cluster (Fig S6). However, our results from yeast show that while 437 some ribosomal proteins are essential and behave similarly across environments, others show 438 high variability. To identify these highly variable ribosomal proteins, the 66 ribosomal 439 proteins were filtered based on their ranks. In each tissue, the expression level ranks of the 440 proteins were stratified into 4 classes: class I (rank 1-17), II (17-34), III (35-51) and IV (51-441 66) (see Methods). This showed that across tissues, 46 out of 66 ribosomal proteins were 442 classified into the same or adjacent classes, while the remaining 20 were classified into 3 443 classes for 11 or more tissues types per class (S7 Table). These 20 were termed as variable 444 ribosomal proteins and analyzed further. Note that had the rank assignments merely amplified 445 small differences in expression levels (noise) for a given protein across tissues, such 446 stratification would not have been observed. Instead, we would have seen a random 447 assignment of ranks across tissues, which is not what was observed. 448 The 20 variable ribosomal proteins spanned mostly classes II, III and IV i.e., their transcripts 449 were both highly expressed and highly variable across tissues, thereby eliminating technical 450 noise as the cause of the observed variability in ranks (S7 Table). Hierarchical clustering of 451 the ranks of these 20 proteins across cell types and tissues showed that these proteins assort 452 into distinct groups (Fig 4A). An identical clustering can also be observed in a heat map of 453 Pearson rank correlations (Fig 4B, correlation P < 0.05). In the hierarchical clustering, 454 distinct sets of ribosomal proteins were associated with two discrete clusters of epithelial 455 cells, a cluster of the nervous tissue (tissue from different sections of the brain and spinal 456 16 cord) and a cluster of human cell lines (Fig 4A). We note that cell lines cluster separately, 457 indicating that similar to modification of their signaling pathways [42], expression patterns of 458 ribosomal proteins are also rewired in these cell lines compared to other human cells and 459 tissues. To investigate this tissue specific ribosomal modularity further, we separately 460 analyzed the 20 ribosomal proteins in the clusters associated with epithelial and nervous 461 tissues. This again showed that the nervous tissue cluster is quite distinct in its use of variable 462 ribosomal proteins compared to epithelial cells (Fig 5A, 5B) Table). We extracted expression values of ribosomal proteins from these samples 476 and performed the same analysis as for the ENCODE data (see Methods). Seventy nine 477 ribosomal proteins passed our filtering criteria and were classified into ranks ranging from 1 478 for the least expression and 79 for the highest expression in each tissue. These were further 479 stratified into 4 classes: class I (rank 1-20), II (21-40), III (41-60) and IV (61-79). We found 480 that ribosomal proteins from the GTEx data showed less variability in classes across tissues 481 compared to the ENCODE data (S9 Table). Consequently, ribosomal proteins that fell into 482 two or more classes with at least 10 tissues per class were identified as variable ribosomal 483 proteins. A total of 18 ribosomal proteins were identified to be variable, of which 7 were the 484 same as those identified in the ENCODE data, showing a significant overlap (Fisher's Exact 485 test, P < 0.1) between variable ribosomal proteins identified using ENCODE and GTEx data 486 ( Fig 6A). The variable ribosomal proteins in the GTEx data separated into two modules 487 (Pearson correlation, P < 0.01, Fig 6A). Similar to the ENCODE data, the nervous tissues 488 (brain and spinal cord) formed a separate cluster, validating our previous observation that a 489 17 different module of ribosomal proteins is used in the nervous system compared to other 490 tissues ( Fig 6B). However, bulk tissues did not cluster separately on the basis of variable 491 ribosomal proteins (Fig 6B). The lack of modularity of ribosomal proteins at the level of 492 tissues indicates that cell specific differences in ribosomal protein expression levels are lost 493 when dealing with data from bulk tissue, because cell specific identity is lost in the GTEx 494 data. 495 A similar analysis was carried out for expression levels of ribosomal proteins in various 496 tissues in mice from the ENCODE data. We note that the signal to noise ratio in the mouse 497 data in ENCODE was significantly higher than in the human data. Consequently, only 42 498 ribosomal proteins in 18 different tissues passed our filtering criteria and were stratified into 499 four classes (S8 Table for details of classes). As in the human data, while the majority of 500 mice ribosomal proteins showed high expression and invariant classification, 14 of these 42 501 ribosomal proteins were as variable across tissues (spanning 3 classes in more than 2 tissues). 502 As in ENCODE and GTEx data from humans, mouse brain tissues also formed a unique 503 cluster, indicating that nervous tissue in general uses a distinct module of ribosomal proteins 504 compared to other tissues (Fig S7A, 7B). 505 We identified RPL38 as a variable protein in ENCODE and GTEx data in humans as well as 506 ENCODE data in mice (Fig 5, 6, S7B). RPL38 is the most extensively studied ribosomal 507 protein associated with ribosomal heterogeneity. This heterogeneity is due to its specialized 508 translation of only the hox mRNA, without affecting translation of other mRNA [19]. 509 Identification of RPL38 as a variable ribosomal protein in our study serves as an independent 510 validation of our analyses. While different sets of ribosomal proteins were found to be 511 variable in mice and humans, 6 out of 18 ribosomal proteins variable in mice were also 512 variable in humans (4 in ENCODE data and 5 in GTEx data, Fig S7B). Furthermore, 4 of 513 these 6 conserved variable ribosomal proteins fell in one cluster in mouse indicating a partial 514 conservation of variability of ribosomal proteins across the two species (Fig S7B). 515 516 DISCUSSION 517 Our study provides several arguments and multiple evidences for the existence of modularity 518 of ribosomal proteins across eukaryota, presumably to facilitate optimized translation 519 efficiency in different environments. We show that, at least in yeast, we see evidence that the 520 18 5'UTRs of ribosomal proteins that form the modules seem to be under selection pressure, 521 which suggests that they play a role in evolutionary adaptation. We interpret our results as 522 evidence for a hitherto unrecognized ribosomal code, wherein specific ribosomal proteins are 523 used in an environment specific manner in yeast and in cell and tissue specific ways in mice 524 and humans. The existence of such a dynamic modularity of ribosomal proteins is the main 525 finding of this paper. The mechanisms that regulate these modules remain to be elucidated 526 and are outside the scope of this paper. 527 Our study also uncovered some general, conserved properties of ribosomal proteins. We find 528 that a subset of variable ribosomal proteins contribute to the plasticity of the ribosome by 529 functioning independently or in concert across different environments by forming modules, 530 defined as sets of proteins functioning in a coordinated manner. This modularity indicates 531 that they have been optimized over the course of evolution based on the need for functional 532 adaptation. We note that only a subset of ribosomal proteins vary among cell types and 533 tissues, with the core ribosome remaining unaffected. Hence cell type specific structural 534 changes resulting from such variation may be difficult to detect. 535 Our findings would argue that, in spite of high sequence conservation [43], the inability of 536 human ribosomal genes to substitute for yeast ribosomal genes [44] is probably because of 537 species specific functioning of ribosomal modules. This, along with differential expression 538 variability of ribosomal proteins in mice and humans, indicates that each species optimizes 539 the composition of its ribosome to adapt to species specific selection pressures, not by 540 substantially altering the sequence of the ribosomal proteins but by regulating their 541 expression in an environment dependent manner using mechanisms yet to be discovered. 542 Our results from two independent expression datasets (ENCODE and GTEx) show that the 543 nervous tissues use a unique ribosomal code compared to the rest of the tissues in both mice 544 and humans. While an overall reduced expression of ribosomal proteins in the brain has been 545 observed previously, it has been attributed to the reduced proliferation of nervous cell types. 546 Here, we show that along with a reduced expression, a unique composition of ribosomal 547 proteins is utilized by nervous tissue. These may play a role in the fundamental physiological 548 differences observed between the brain and the rest of the body. 549 In mice and humans, recently evolved paralogs RPL27L, RPL22L1, RPL7L1, and RPL39L, 550 showed poor but highly tissue specific expression compared to core ribosomal proteins. This 551 19 suggests that there is an ongoing process of adaptation driving modular ribosomes, with 552 recent paralog proteins still evolving in response to selection pressures on them and on other, 553 more ancient ribosomal proteins and pathways. 554 Our results show that ribosome modularity is a dynamic, evolving process which seems to be 555 involved in the evolution of species specific ribosomal proteins, the diversification of their 556 sequences and functions and the creation of novel, species specific ribosomal proteins [ RPP1B  RPL43B  RPS24A  RPS22A  RPS1A  RPL15B  RPL38  RPL34B  RPS25A  RPL40A  RPL41B  RPL22B  RPS9A  RPL7B  RPL2B  RPS8A  RPL18B  RPS27A  RPL26A  RPL29  RPL16A  RPS14B  RPS12  RPS14A  RPS17B  RPS26B  RPS10B  RPS0A  RPS7B  RPS30A  RPS23A  RPS11B  RPL21B  RPL19B  RPL24B  RPL9A  RPL35B  RPP2A  RPL37B  RPL6B  RPL11B  RPL23A  RPL36A  RPL8B  RPL8A  RPL33B  RPL17B  RPL9B  RPL4A