Abstract Two novel approaches were recently suggested for genome-wide identification of protein aspects synthesized at a given time. Ribo-Seq is based on sequencing all the ribosome protected mRNA fragments in a cell, while PUNCH-P is based on mass-spectrometric analysis of only newly synthesized proteins. Here we describe the first Ribo-Seq/PUNCH-P comparison via the analysis of mammalian cells during the cell-cycle for detecting relevant differentially expressed genes between G1 and M phase. Our analyses suggest that the two approaches significantly overlap with each other. However, we demonstrate that there are biologically meaningful proteins/genes that can be detected to be post-transcriptionally regulated during the mammalian cell cycle only by each of the approaches, or their consolidation. Such gene sets are enriched with proteins known to be related to intra-cellular signalling pathways such as central cell cycle processes, central gene expression regulation processes, processes related to chromosome segregation, DNA damage, and replication, that are post-transcriptionally regulated during the mammalian cell cycle. Moreover, we show that combining the approaches better predicts steady state changes in protein abundance. The results reported here support the conjecture that for gaining a full post-transcriptional regulation picture one should integrate the two approaches. __________________________________________________________________ Gene expression is a multi-step process, with the first stage of this process (transcription) and its product (mRNA levels) comprehensively studied and measured. However, it was shown that the correlation between mRNA and protein levels is relatively limited[28]^1,[29]^2,[30]^3. Consequently, recently various technologies for studying post transcriptional regulation, and specifically translation, have emerged to close this gap[31]^1,[32]^3,[33]^4,[34]^5,[35]^6,[36]^7,[37]^8,[38]^9,[39]^10,[40] ^11,[41]^12,[42]^13,[43]^14,[44]^15,[45]^16,[46]^17,[47]^18,[48]^19,[49 ]^20,[50]^21,[51]^22,[52]^23,[53]^24,[54]^25,[55]^26. Currently the most common technology for studying translation is ribosomal profiling (Ribo-Seq). Although ribosomal profiling was introduced only several years ago it has already been successfully employed for answering fundamental biological questions related to post transcriptional regulation of gene expression[56]^5,[57]^27,[58]^28,[59]^29,[60]^30. [61]Figure 1A includes the major steps of the ribosomal profiling approach: Cells are treated with cycloheximide (or a different drug) to arrest translating ribosomes; extracts from these cells are then treated with RNase to degrade regions of mRNAs not protected by ribosomes; the resulting 80S monosomes, many of which contain a ~30-nucleotide ribosomal protected footprint (RPF), are purified (e.g. using sucrose cushion) and then treated to release the RPFs, which are processed for Illumina high-throughput sequencing. The next steps are computational: the RPFs are mapped to the transcriptome, and based on them it is possible to infer various biophysical properties related to the translation elongation process. For example, each ribosomal footprint read is related to a certain codon along the mRNA, and was generated when the codon in one of the mRNA molecules is covered by a ribosome. Thus, from a biophysical perspective, relatively slower codons along the mRNA can be detected based on the fact that they are covered by ribosomes for longer periods of time, creating a higher number of reads. Figure 1. [62]Figure 1 [63]Open in a new tab A schematic illustration of the study: Ribo-Seq (A) and PUNCH-P (B) data are used simultaneously in-order to augment the information which can be extracted based on each alone (description of the methods appears in the main text). (C) Predictive power of steady-state protein levels based on the two approaches are assessed. (D) Protein-protein interaction (PPI) analyses are performed based on differentially expressed (DE) genes in the cell-cycle phases M and G1 based on each approach. (E) Pathway enrichment analyses are performed based on the DE genes detected based on each approach. (F) Clustering of the PPI sub-networks induced by DE genes detected based on each approach is performed. Recently a new approach called PUNCH-P[64]^31,[65]^32 was proposed. This approach is based on the combination of biotinylated puromycin with MS analysis to globally label newly synthesized proteins, enabling identifying the proteins translated in a certain condition. The method involves isolation of ribosomes by ultracentrifugation followed by cell-free labeling of nascent polypeptide chains with 5′ biotin-dC-puromycin 3′ (Biot-PU), capture on immobilized streptavidin, and analysis by liquid chromatography-tandem MS (LC-MS/MS). This work flow leads to the identification of thousands of newly synthesized proteins in a certain condition, generating a snapshot of the cellular translatome, see [66]Fig. 1B. It is easy to see that both approaches measure very similar but non-identical aspects related to protein synthesis ([67]Fig. 1A,B). Roughly, Ribo-Seq is based on the total number of ribosomes on the mRNA molecules related to a certain gene; PUNCH-P, on the other hand, is based on the total amount of nascent peptide emerging from the ribosomes on the mRNA molecules related to a certain gene which are translating at the time of the experiment. Since not all ribosomes on the mRNA (i.e. can be detected by Ribo-Seq) are actually translating[68]^33,[69]^34,[70]^35 at a certain moment (i.e. can be detected by PUNCH-P), the signal detected by these approaches is not identical. Furthermore, the different approaches are expected to have different experimental biases/noise as they are based on different experimental/analysis techniques: sequencing vs. proteomics. It is important to mention that a priori it is not clear which approach (or if there is an approach that) performs better. This is true not only due to the different biases related to the different methods, but also due to the fact that each of them is expected to capture biological meaningful signals not detected by the other: changes in the number of translating ribosomes, but also the total number of ribosomes, are expected to be relevant to protein levels regulation. Thus, the aim of this study is to compare these two methods, which were both performed at the G1 and M phases of the cell cycle, and discern if their integration can yield improved predictions of relevant genes/proteins, and uncover otherwise elusive biological phenomena. To this end we: 1. Tested the predictive power of steady-state protein levels of each approach and their combination ([71]Fig. 1C). 2. Uncovered significant M/G1 differentially expressed genes with each of the approaches. 3. Exploiting these genes we discovered relevant intra-cellular pathways with each of the approaches ([72]Fig. 1E). 4. Discerned biological relevant properties related to the differentially expressed genes detected by each approach and the protein-protein interaction network ([73]Fig. 1D,F). In points 2.-3. we specifically studied the genes/proteins detected by only one of the methods. Data for PUNCH-P was taken from[74]^31 (see Methods), while we generated the Ribo-Seq data via two experiments, one with 3 replicates, and the other with one replicate, totalling 4 technical replicates per each cell-cycle phase, G1 and M (see [75]Methods and Supplementary Methods). Results Correlations based on Ribo-Seq and PUNCH-P with steady state protein levels Steady state protein levels are expected to be affected by all gene expression steps (e.g. transcription, translation, mRNA degradation, protein degradation). Thus, steady state protein levels (PSS) are expected to correlate with mRNA levels, PUNCH-P (PP), and Ribo-Seq (RP). In addition, it is easy to see that RP and PP (which encapsulate both the mRNA levels and the translation step, or the number of ribosomes on the mRNA molecules) are expected to have higher correlation than mRNA levels with steady state protein levels. Moreover we expect to see relatively high correlation between PP and RP as they measure similar variables. Finally, we also expect that the combination of the different measures can improve the prediction of steady state proteins, as each of them encapsulates non-identical aspects of gene expression and exhibits different experimental biases. All these points are verified in this sub-section. At the first step, we aimed at providing estimations for the effect of transcription and translation on steady state protein levels via the correlation of the products of these stages. Our analyses demonstrate that the correlation (all correlations reported in the paper are Spearman, see Methods) between the G1 and M phases of steady state protein levels and Ribo-Seq (r(PSS,RP)) are: 0.70 (p < 10^−454) (see [76]Fig. 2A,E) and 0.70 (p < 10^−454) respectively (see [77]Fig. 2B,F); the correlation is significant and high also when controlling for mRNA levels (r (PSS,RP|mRNA): 0.45 (p = 2.4·10^−252) (see [78]Fig. 2C,E) and 0.47 (p = 4·10^−280) (see [79]Fig. 2D,F). The correlation between M and G1 phases of steady state protein levels and PUNCH-P (r(PSS,PP)) are: 0.68 (p < 10^−454) (see [80]Fig. 3A,E) and 0.68 (p < 10^−454) (see [81]Fig. 3B,F); the correlation is significant and high also when controlling for mRNA levels (r (PSS,PP|mRNA): 0.48 (p = 3.3·10^−213) (see [82]Fig. 3C,E) and 0.48 (p = 2.7·10^−208) (see [83]Fig. 3D,F). The correlation of PSS with mRNA levels is indeed lower than with both RP and PP, (r(PSS,mRNA)): 0.61 (p < 10^−454) and 0.60 (p < 10^−454), for the G1 and M phases respectively (see [84]Supplementary Figure S4). Figure 2. [85]Figure 2 [86]Open in a new tab (A) Scatter plot of steady state protein levels (PSS) (y-axis , data is log2-scaled) and Ribo-Seq (RP) (x-axis, read count log2-scaled RPKM (see Methods)) G1 phase. (B) Scatter plot of PSS (y-axis log2(intensity)) and RP levels (x-axis, read count log2-scaled RPKM (see Methods)) M phase. (C) Correlation between PSS and RP G1 phase for different bins of genes sorted by mRNA levels (the y-axis is the correlation, the x-axis is the mRNA levels RPKM ranges, scatter plots are log2-scaled). (D) Correlation between PSS and RP M phase for different bins of genes sorted by mRNA levels (the y-axis is the correlation, the x-axis is the mRNA levels RPKM ranges, scatter plots are log2-scaled). (E) A summary of the 2 correlations performed with PSS: RP, and RP controlled for mRNA levels (partial correlation) for G1 phase. F. A summary of the 2 correlations performed with PSS: RP, and RP controlled for mRNA levels (partial correlation) for M phase. Figure 3. [87]Figure 3 [88]Open in a new tab (A) Scatter plot of steady state protein levels (PSS) (y-axis log2(intensity)) and PUNCH-P (PP) (x-axis log2(intensity)) G1 phase. (B) Scatter plot of PSS (y-axis log2(intensity)) and PP levels (y-axis log2(intensity)) M phase. (C) Correlation between PSS and PP G1 phase for different bins of genes sorted by mRNA levels (the y-axis is the correlation, the x-axis is the mRNA levels RPKM ranges, scatter plots are log2-scaled). (D) Correlation between PSS and PP M phase for different bins of genes sorted by mRNA levels (the y-axis is the correlation, the x-axis is the mRNA levels RPKM ranges, scatter plots are log2-scaled). (E) A summary of the 2 correlations performed with PSS: PP, and PP controlled for mRNA levels (partial correlation) for G1 phase. (E) A summary of the 2 correlations performed with PSS: PP, and PP controlled for mRNA levels (partial correlation) for M phase. PP data is log scaled. Next we estimated the correlation between the two methods PP and RP to evaluate the similarity between the prediction obtained by the two methods. Our analyses demonstrate that the correlations between the M and G1 phases of PUNCH-P and Ribo-Seq are 0.63 (p < 10^−454) (see [89]Fig. 4A,E) and 0.63 (p < 10^−454) (see [90]Fig. 4B,F) respectively; the correlations are high also when controlling for mRNA levels (r (PP,RP|mRNA)): 0.31 (p = 2.5·10^−112) (see [91]Fig. 4C,E) and 0.32 (p = 2.2·10^−117) (see [92]Fig. 4D,F) for M and G1 respectively. Figure 4. [93]Figure 4 [94]Open in a new tab (A) Scatter plot of PUNCH-P (PP) (y-axis log2(intensity)) and Ribo-Seq (RP) G1 phase (x-axis, read count log2-scaled RPKM (see Methods)). (B) Scatter plot of PP (y-axis log2(intensity)) and RP levels M phase (x-axis, read count log2-scaled RPKM (see Methods)). (C) Correlation between PP and RP G1 phase for different bins of genes sorted by mRNA levels (the y-axis is the correlation, the x-axis is the mRNA levels RPKM ranges, scatter plots are log2-scaled). (D) Correlation between PP and RP M phase for different bins of genes sorted by mRNA levels (the y-axis is the correlation, the x-axis is the mRNA levels RPKM ranges, scatter plots are log2-scaled). (E) A summary of the 2 correlations performed with PP: RP, and RP controlled for mRNA levels (partial correlation) for G1 phase. (E) A summary of the 2 correlations performed with PP: RP, and RP controlled for mRNA levels (partial correlation) for M phase. PP data is log scaled. Finally, as can be seen in [95]Fig. 5, regressors based on PP and RP for the M and G1 phases (see Methods), as a function of RP coverage from >0 to ≥60%, achieve improved correlation with steady state protein levels in comparison to a regressor based only on either PP or RP, while including mRNA levels further improves the correlation (but not substantially), see [96]Supplementary file Supplementary_Table_S1_RegressorCorrs.xlsx. Figure 5. Correlations with M and G1 steady state protein levels (PSS) for three regressors based on PP, PP and RP, and PP, RP and mRNA respectively, as a function of the RP coverage (>0 – 60%), the y-axis is the correlation. [97]Figure 5 [98]Open in a new tab We performed a 2-fold cross validation 100 times per Spearman linear regressor, with the standard deviation of all the regressors being between 0.0062–0.0141, with the variation being lower for the most part as coverage increases and with more measurements combined. One can see that combining all 3 measurements improves correlations with steady state protein levels. See [99]Supplementary file Supplementary_Table_S1_RegressorCorrs.xlsx. The results demonstrate that the correlation between PP and RP is high (as expected) but is far from being perfect. In addition, these results support the hypothesis that the variance in protein levels can be explained by PP, RP, and mRNA levels; thus both changes in mRNA levels (regulated among others via transcription) and changes in ribosomal densities (as part of the translation step) effect the changes in protein abundance (translation, and not only transcription, as traditionally thought, has important contribution to changes in protein levels). The results also show that PP and RP have significant predictive power of protein levels. Finally, we demonstrate how a regression based both on PP and RP improves the prediction of steady state protein levels. Since steady state protein levels may also be affected by proteins not translated at the moment of the experiment, a predictor based on both PP and RP improves the prediction of steady state protein levels upon a predictor based on PP or RP alone. There are relevant genes detected to be differentially expressed exclusively by each method At the next step our objective was to show that both PP and RP can be used for detecting relevant differentially transcriptional and post transcriptional regulated genes, and that each of these methods exclusively detects relevant genes. To demonstrate this point we first inferred the set of differentially expressed (DE) genes between the G1 and M phases of the cell cycle detected for PUNCH-P (PP) and Ribo-Seq (RP) separately. M/G1 differentially expressed (DE) genes were determined according to DESeq[100]^36 for Ribo-Seq (RP), where the top 10% most significant FDR p-values were selected (See [101]Methods and Supplementary Methods), and for PUNCH-P (PP) according the top 10% ANOVA significant fold change (see Methods,[102]^31). At the next step we defined three DE gene groups: 1. RP-PP (genes that are significantly DE in RP but not in PP; 1,090 genes). 2. PP-RP (genes that are significantly DE in PP but not in RP; 200 genes). 3. RP∩PP (genes that are significantly DE both in PP and in RP; 125 genes). These two DE sets, and the three DE groups derived from them will be employed throughout the paper. We performed pathway and biological process enrichment for each of the groups (Methods). To achieve our objective, we aimed to show that relevant pathways and biological processes are significantly enriched with DE genes in all three cases. As can be seen in [103]Fig. 6 (for a full pathway list please see [104]Supplementary Information Table 1 ([105]section 3.2), and for a full biological process list see Supplementary files [106]Supplementary_Table_S3_RPDavidReports.xlsx, [107]Supplementary_Table_S4_PPDavidReports.xlsx, [108]Supplementary_Table_S5_RPiPPDavidReports.xlsx, see further details in Methods), each technique enables detecting meaningful genes/proteins that are not detectable by the other. The detected differentially expressed post-translational regulatory pathways related to the three sets described via enrichment analysis include: central cell cycle, central gene expression regulation, DNA damage and replication, and chromosome arrangement. Figure 6. Figure 6 [109]Open in a new tab Selected pathways and biological processes which are significantly enriched by the 3 groups of DE genes (for a full pathway list please see [110]Supplementary Information Table 1 ([111]section 3.1), and for a full biological process list see Supplementary files Supplementary_Table_S3_RPDavidReports.xlsx, Supplementary_Table_S4_PPDavidReports.xlsx, Supplementary_Table_S5_RPiPPDavidReports.xlsx). For example, all three sets are enriched with genes related to the cell cycle and M phase; RP-PP and RP∩PP are enriched with genes related to apoptosis regulation, while PP-RP is enriched with genes related to cell proliferation; RP-PP is enriched with genes related to Spindle Organization and DNA Damage response, while PP-RP and RP∩PP are enriched with genes related to Spindle Microtubule/Microtubule organization center and DNA replication. We would like to emphasize the fact that aside from detecting distinct biologically relevant pathway enrichments, there are cases that the sets RP-PP and PP-RP are enriched with genes related to the same (or very similar) pathways, suggesting that the different techniques tend to find different parts of the same relevant pathways. This evidence again demonstrates the advantage of combining/considering the two methods. Now, in order to further demonstrate that each of the techniques, RP and PP, uncovers biologically relevant protein-protein interactions that cannot be detected by the other technique, three PPI network colouring schemes were defined, where “black” nodes represent differentially expressed genes (DE; see Methods) as above between the G1 and M phases of the cell cycle. In the first case, the black nodes were defined as genes that are DE according to RP but not PP (RP-PP); in the second case the black nodes were defined as genes that are DE according to PP but not RP (PP-RP); in the third case the black nodes were defined as genes that are DE according to both RP and PP; similarly to the previous analysis. We computed the mean distance (md) between all black nodes in each of the aforementioned three cases. For each case, we computed a PPI empirical p-value by randomizing each PPI network 100 times respectively generating random networks with a similar degree distribution as the original one, and calculating the black node distance, showing that the mean distances are shorter in the real graph in comparison to the random ones (see details in the Methods section). Shorter distances between DE PPI nodes means more meaningful biological signals, as if indeed we uncover real regulatory changes in signalling pathways, we expect them to be clustered/close in the PPI network (we expect to see physical interactions between DE genes). All p-values were <10^−2 (when 100 permutations are performed a p-value <10^−2 means that the observed distance was always shorter than the distances obtained during all 100 random permutations), with the mean distance being shorter (2.01) in the case of the RP∩PP than in the case of the RP-PP and the PP-RP groups (2.12 and 2.13, respectively) (see [112]Fig. 7). Figure 7. Three PPI network colouring schemes were defined, where black nodes represent DE genes (based on PP and/or RP): 1. RP-PP. 2. PP-RP. 3. RP∩PP DE. Figure 7 [113]Open in a new tab We compute the mean distance (md) in each between all black nodes. For each case we compute a PPI empirical p-value by randomizing each PPI network 100 times respectively and calculating the black node distance. Shorter distances between DE PPI nodes means more meaningful biological signals (we expect to see physical interactions between DE genes). Our analyses demonstrate that genes detected by each of the methods (even if not detected by the other) tend to be closer to each other than expected by the null model in the PPI network. Thus, this result supports the hypothesis that not all biological meaningful genes detected by one of the methods are detected by the other. Modules of differentially post-transcriptionally expressed genes and physical interactions To better understand the differentially expressed genes detected by PP and RP we performed a clustering analysis (Newman algorithm[114]^37, see Methods), on the PPI network using the previously described DE genes according to RP and PP respectively, divided into the following three aforementioned groups: 1. RP-PP. 2. PP-RP. 3. RP∩PP (See [115]Supplementary Figure S5 for RP∪PP ([116]Supplementary section 3.3)). We projected each of the 3 groups on to the PPI network respectively, and only selected genes from each group that have a neighbour in that group in the PPI. In each case, the Newman algorithm partitions the PPI networks to sub-networks, where each sub-network is modular and includes nodes related only to the corresponding group. To understand the pathways related to each module we performed pathway enrichment based on the genes in each module (for all significantly enriched pathways see [117]Supplementary file Supplementary_Table_S6_ClusterPathwayEnrichment.xlsx). As can be seen in [118]Fig. 8, the number of modules detected for each of the groups RP-PP/PP-RP/RP∩PP were 4/13/15 respectively. The modules in all cases were enriched with relevant pathways related to the cell-cycle, DNA Damage and replication, and gene expression regulation and signalling. This analysis demonstrates again that meaningful sub networks of physical interactions are detected by each of the methods separately and together. Figure 8. [119]Figure 8 [120]Open in a new tab (A) RP-PP clusters: 879 genes participate, resulting in 4 clusters. (B) PP-RP clusters: 96 genes participate, resulting in 13 clusters. (C) RP∩PP clusters: 90 genes participate, resulting in 15 clusters. The functional enrichment related to each cluster appears in the figure. There are 4 node sizes depicted in the figure, according to their centrality (the 4^th size being equal for most nodes is a coarse-grained portrayal for simplicity). For the full cluster pathway enrichment see Supplementary_Table_S6_ClusterPathwayEnrichment.xlsx. Genes detected to be of opposite regulatory direction based on the different methods Finally, we aimed to examine if there are genes that are detected to be significantly expressed based on both RP and PP but in opposite directions. To this end we looked at the following groups: (a) Genes that have RP M/G1 fold-change >0 and PP M/G1 fold-change <0 (b) Gens that have RP M/G1 fold-change <0 and PP M/G1 fold-change >0 In total 78 genes appear in the first group and 68 genes in the second (the list of genes appears in [121]Supplementary file Supplementary_Table_S7_RPopPPdiffGenes.xlsx). Both lists of genes were enriched with relevant pathways related to gene regulation and cell cycle (see [122]Supplementary table 2 in Supplementary section 3.4). For example, the first group is enriched with genes related to DNA Replication and cell cycle control, while the second group is enriched with genes related to various central signalling pathways. This result suggests that increasing/decreasing ribosomal density as detected by Ribo-Seq is not always related to increasing/decreasing the ribosomal density involved in protein synthesis at a certain time point as detected by Punch-P. There can be various explanations for this discrepancy which may be related (among others) to the fact that translation elongation (and not only translation initiation) is controlled during the mammalian cell cycle. For example, regulatory changes that cause ribosomal stalling during elongation[123]^33,[124]^34,[125]^35 may cause traffic jams, for example, near the beginning of the ORF where the ribosomes are not translating, or there is no nascent peptide emerging from the ribosome; since such ribosomes can theoretically be detected by RP and not PP they may increase RP but decrease PP. It is also possible that in some cases, due to traffic jams, the RNase does not accurately digest the mRNA between ribosome protected regions. This may result in underestimation of ribosome density and may lead to a decrease in measured ribosome density when the actual density increases (see, for example,[126]38). It is also important to emphasize that aspects related to changes in mRNA levels can’t trivially explain the observed discrepancies since both RP and PP are expected to be proportional to mRNA levels (if there are no traffic jams and biases). Details regarding some of the post-transcriptionally regulated genes detected The major aim of this study was to show in an objective, large scale, quantitative manner that combining RP and PP measurements (in comparison to each measure independently) is expected to improve the ability to detect meaningful post transcriptional regulation signals. Thus, we focused on objective quantitative measures. Nevertheless, in this section we provide some biological examples related to meaningful/relevant biological cell cycle signals detected by PP and RP. To this end, we will focus on the module inference/clustering analysis performed based on protein-protein interactions among genes detected to be differentially expressed based both on PP and RP (90 genes, see [127]Fig. 8C and [128]supplementary table Supplementary_Table_S9_RPiPP_ClusterPEDetails.xlsx). As mentioned, we detected 15 modules (see [129]Fig. 8C); here we will discuss in further detail the four largest modules. The first module of size 11 genes/proteins includes many genes that encode ribosomal proteins (e.g. RPL3, RPL34, RPS10, RPL35, RPL32, RPL29) which are down regulated (in terms of both RP and PP) in M in comparison to G1. This result supports the hypothesis that translation (specifically the canonical regulatory mechanisms) is globally down regulated during M phase[130]^39,[131]^40,[132]^41 in mammalian cells, and that the down regulation occurs and can be detected also post transcriptionally. The second module of size 27 genes/proteins includes various M phase specific genes/proteins mainly related to spindle morphogenesis and chromosome movement that are found to be up-regulated based on PP and RP in M phase: for example, one hub in this module is the gene/protein ESPL1 which stabilizes cohesion between sister chromatids before anaphase, and their timely separation during anaphase is critical for chromosome inheritance. Another hub is the gene/protein BUB3 that is involved in spindle checkpoint function, which is up-regulated in M phase together with BUB1. Interestingly the module also includes several kinesins KIF22, KIF20A, KIF18A, KIF23, KIF2C, KIFC1; it was suggested that kinesins and proteins interacting with them are known to have important spindle morphogenesis and chromosome movement in cell division[133]^42,[134]^43,[135]^44,[136]^45, and our analysis emphasizes their post-transcriptional regulation. Naturally this module also includes (among others) cell cycle regulatory proteins such as CDC20 and CDC8 that are involved in nuclear movement prior to anaphase, chromosome separation, and spindle formation. It also includes various Kinases (e.g. PLK1, CDK1, and TTK) that are involved in regulating the processes mentioned above. The third module includes 13 genes/proteins related mainly to DNA replication. One hub in this module is the gene FZR1; it is up-regulated in M-phase and is a key regulator of ligase activity of the anaphase promoting complex/cyclosome. The module includes genes/proteins related to DNA replication regulation, and activation and maintenance of the checkpoint mechanisms in the cell cycle that coordinate S phase and mitosis: MCM6, CDC6, MCM3, PCNA, RFC4; all these genes are down regulated (based on RP and PP) at the M-phase as there is no DNA replication during M-phase[137]^46. The module also includes various genes related to gene expression regulation and proliferation such as the gene DMAP1 which represses transcription and is up-regulated in M-phase. Finally, it includes genes related to cell cycle progression such as the genes CCNA2 and CDK4 which are up-regulated in M-phase. The fourth module (module number 14) includes 12 genes/proteins which are related among others to dynamic microtubules polymerization, which is an important step of the M-phase[138]^46. For example, the module’s main hub, TUBB4B (Tubulin, beta 4B class IVb), and 3 additional tubulins (TUBB6, TUBA4A, TUBB4A) are up regulated (according to RP and PP) in M-phase; this fact emphasizes the post transcriptional regulation of microtubules polymerization during M-phase. To summarize the details depicted above, the genes/proteins detected by RP and PP are highly relevant to cell-cycle biology and teach us about the central role of post transcriptional regulation during the cell cycle. Discussion This study includes the first comparison of RP and PP. We report various analyses that demonstrate that RP and PP can exclusively detect relevant differentially expressed genes. Specifically, based on enrichment and PPI network analyses, we show that genes that are detected by each of these methods, but not by the other, tend to include biologically relevant signals. We evince that the prediction of steady state protein levels can be improved by combining PP and RP measurements. Furthermore, we show that the relevant DE genes detected by each of the methods may have opposite fold-change, demonstrating that the two techniques can detect different aspects of translational regulation, and are thus in part synergistic. There are three major explanations to the fact that the correlation between 1) a model based on RP, PP, and mRNA and 2) steady state protein levels is not prefect: First, steady state protein levels are a result of many gene expression steps such as the regulation of protein degradation, post-translational regulation, and secretion of proteins. Second, there are different biases in the cases of the various experiments/measurements. For example, the sequencing based experiments have biases related to RNase, while the proteomic based approaches have biases related to protein digestion; in addition, the distribution of protein/peptide length is different in PUNCH-P (where truncated proteins are generated at the first stage) and in steady state protein levels measurements. Third, some of the differences are due to natural variability among technical repeats and may also be related to the stochasticity (specifically for lowly expressed genes) of the gene expression steps (see, for example,[139]47). We would like to summarize some of the different biases in the RP and PP experiments. The RP major biases can be related to preferences/non-uniform efficiency of the RNase, sequencing biases,