Gradient Encoding Results Are Concordant with Traditional Polysome Profiling
We used a genome-wide polysome profiling technique, called Gradient Encoding, to identify genes translationally regulated by v-Abl transformation, and to define the contribution of mTOR to this regulation. Lysates harvested before and after 12 hours of treatment either with 2.5 µM imatinib, a v-Abl kinase inhibitor, or 10 ng/mL (10.9 nM) rapamycin, an mTOR inhibitor, were fractionated by sedimentation through linear sucrose gradients. Gradient fractions were “encoded” such that the mRNA from successive fractions was labeled with increasing ratios of Cy5 to Cy3. mRNAs derived from fractions in the lighter portion of the gradient therefore have a lower Cy5 to Cy3 ratio, whereas those deeper in the gradient have a higher Cy5 to Cy3 ratio. The ratio of Cy5 to Cy3 for each mRNA therefore reflects its average position within the gradient (). We thus “encoded” the sedimentation rate of each mRNA across the entire gradient. The resulting ratios were quantitatively measured for each mRNA species by hybridization to DNA microarrays, and related to the 260 nm absorbance peaks representing different numbers of ribosomes bound per mRNA (, see Materials and Methods
). We assume that changes in the number of ribosomes bound to a particular transcript in response to treatment represent a change in the rate of translation of that transcript, provided that 1) the sedimentation rate is determined by the number of ribosomes bound to each transcript, 2) initiation is rate-limiting and the elongation rate is uniform, and 3) the ORF length remains constant.
Schematics for the genomic (gradient encoding) and proteomic (pSILAC) techniques employed.
We confirmed the Gradient Encoding results for nine selected genes by high resolution polysome profiling and RT-PCR before and after imatinib treatment (, Fig. S1
). The analysis for the two most extreme of these nine genes, Tcf12 (NM_011544) and Hsp90ab1 (NM_008302), is shown here (). The relative change in mRNA abundance was measured simultaneously by quantitative RT-PCR (qRT-PCR) and is displayed in the insets of the gradient profiles (, inset).
Changes in translation inferred by gradient encoding agree with those obtained by qRT-PCR of sucrose gradient fractions and Western analysis.
We quantitatively analyzed changes in transcript abundance and translation following imatinib or rapamycin treatment in three biological replicate experiments; all three replicate experiments were used in subsequent statistical analyses. The change in average ribosome number between conditions is expressed as the Log2
of the ratio of average ribsome number in treated versus untreated cells for each mRNA. The transcripts with the largest negative values are therefore those with the greatest apparent reduction in translation upon imatinib or rapamycin treatment. To quantitatively evaluate the performance of the encoding method, we calculated the expected “encoded” values, based on the results from qRT-PCR analysis of individual fractions. The results from the individual gene qRT-PCR (qPCR) were in agreement with the results from Gradient Encoding (GE) (Pearson r
0.90, ). Western analysis of Hsp90ab1 and Tcf12, the two most extreme examples of translational regulation confirmed by qRT-PCR, highlighted in green and red, respectively (), confirmed the dramatic induction of Tcf12 after 12 hours of imatinib treatment and showed modest reduction of Hsp90ab1 after 12 and 18 hours of imatinib treatment (). However, Hsp90ab1 protein levels were not measurably reduced after 12 or 18 hours of rapamcyin treatment, despite significant reduction in translation after 12 hours (the average Log2 ratio was −1.34 for imatinib and −0.86 for rapamycin). This may reflect a difference in protein stability, which is not measured by our genomic methods, a point which will be addressed below.
Translational Regulation Contributes Significantly to Global Reprogramming of Protein Composition in v-Abl Transformation
We expected a change in translation to result in a corresponding change in protein production. Proteins with low turnover rates, however, might show only small changes in overall abundance even if their synthesis rates changed abruptly after 12 hours of drug treatment, as observed for Hsp90ab1 after rapamycin treatment (). In order to more accurately measure gene-specific changes in protein synthesis over the course of drug treatment we used a pSILAC (p
abeling by A
mino acids in C
ell culture) method whereby proteins are metabolically labeled only for the duration of drug treatment () 
. v-Abl-transformed cells were cultured in standard “light” media and then switched to either “medium” or “heavy” media containing L-arginine with stable isotope-labeled carbon (13
C-L-arginine) or both carbon and nitrogen (13
N-L-arginine), respectively, in place of 12
N-L-arginine upon addition of drug. Protein samples isolated from heavy-labeled, untreated cells and medium-labeled, imatinib-treated cells were mixed and ~60 µg of this 1
1 mixture of total lysate was subjected to PAGE (Polyacrylamide Gel Electrophoresis). The “reverse” experiment was performed comparing heavy-labeled, imatinib-treated cells and medium-labeled, untreated cells and served as a biological replicate. Each of the two lanes was cut into 8 slices that were separately analyzed by mass spectrometry. For each tryptic peptide that we could confidently identify, we determined the ratio of the medium to heavy isotope-labeled species, which should reflect the change in production of the corresponding protein after 12 hours of imatinib treatment. The data from both the “forward” and “reverse” experiments were used to determine the average ratios of imatinib-treated/untreated protein levels. We restricted this analysis to proteins for which at least two unique tryptic peptides were identified. To evaluate the relative contributions of translational and transcriptional regulation to the observed changes in protein production, we compared the correlation between changes in protein production +/− imatinib and the corresponding changes in estimated translation (), transcript abundance (), or the product of these two estimates () for 258 genes (see Table S1
Changes in translation (inferred from gradient encoding) predict changes in protein abundance.
The observed changes in translation (Spearman r
0.55, Pearson r
0.55) more accurately predicted changes in protein production than did changes in mRNA abundance (Spearman r
0.34, Pearson r
0.27), suggesting that translation independently accounts for a large component of the reprogramming of the cell’s proteome (). We include both Pearson (based on values) and Spearman (based on rank) correlations here, but only Spearman correlations are displayed in the figures. The genes for which we have pSILAC data qualitatively represent the total population, and are not skewed or clustered into regions of extreme change in either translation or abundance (Fig. S2
The genes that made the greatest contribution to the better correlation of translation than transcription with protein synthesis rates were, as expected, components of the translation apparatus. Nevertheless, even when the ribosomal proteins, elongation factors, and initiation factors were excluded from the pSILAC dataset (resulting in the removal of data for 25 genes), the correlations between the changes in synthesis of specific proteins and either translation (Spearman r
0.47, Pearson r
0.48), abundance of the corresponding transcripts (Spearman r
0.49, Pearson r
0.44), or the product of change in mRNA abundance and translation (Spearman r
0.60, Pearson r
0.53) confirm the large contribution of translational regulation to reprogramming the protein composition of the cell.
Although gene-specific changes in protein turnover rates, which cannot be inferred from transcript abundance or translation rates, could play an important role, genome-wide quantitation of both changes in translation and changes in mRNA abundance greatly improves our ability to infer the global reprogramming of the cells proteome by enabling quantitative analysis of more than ten times as many genes as can be evaluated by current proteomic methods.
A Major Role for mTOR in Reprogramming of Gene Expression by v-Abl
Concerted, large-scale activation of translation is a hallmark of the mTOR pathway 
. BCR-ABL1 induces phosphorylation of 4E-BP1 and S6K via mTOR through activation of both the PI3K and MAPK pathways 
. Because the genes encoding components of the ribosome and translation complex subunits comprise a substantial fraction of the ribosome-associated mRNAs in dividing cells, the decreased translation of these genes may explain much of the global redistribution of mRNAs from polysomes to monosomes, reflected in the 260 nm trace of the polysome profile following imatinib treatment (, and 
The translational response of v-Abl-transformed pre-B cells to imatinib (2.5 µM for 12 hours) or rapamycin treatment (10.9 nM for 12 hours) or no treatment.
Qualitatively, rapamycin treatment produced a less prominent shift in ribosomes towards monosomes than did imatinib (). Rapamycin sterically hinders formation of mTORC1, but not mTORC2 
and mTORC1 inactivation can activate mTORC2, due to loss of a negative feedback from mTORC1 on the mTORC2 complex 
. However, chronic exposure to rapamycin has been shown to also inhibit mTORC2 
. We tested the degree of mTORC1 and mTORC2 inhibition by the loss of phosphorylation of Ribosomal protein S6 (RpS6) (an mTORC1 target) or AKT-Ser473 (an mTORC2 target) (). No augmentation of mTORC2 activation by rapamycin was observed; phosphorylation of RpS6 and AKT-Ser473 was reduced to a comparable extent by either imatinib or rapamycin, at the concentrations and duration of drug used here.
mTOR canonically induces translation of genes harboring a 5′TOP sequence 
. The requirement of this signal for mTOR-dependent translational regulation, however, is not stringent, and cell-type-dependent responses to mTOR inhibition are common 
. Therefore, we compared the translational reprogramming after imatinib to that after rapamycin treatment in order to determine the contribution of mTOR to the translational re-programming of v-Abl. shows genes statistically significantly activated (in red) by imatinib (by inference repressed by v-Abl) and repressed (in green) by imatinib (by inference activated by v-Abl) in scatterplots relating changes in translation to mRNA abundance (). Of the 10507 unique genes for which we had data in three replicate experiments, 175 unique genes were deemed significantly activated by imatinib (10% FDR (false discovery rate) cut off, using significance analysis of microarrays (SAM)), and 591 genes were repressed (10% FDR cut off, SAM). Despite equal dephosphorylation of RpS6 and AKT-Ser473, the range of change in translation affected by rapamycin was smaller than that following imatinib treatment. None of the 10688 unique genes for which we acquired rapamycin data in triplicate were translationally induced and 192 unique genes were repressed at a 10% FDR threshold ().
Changes induced by imatinib correlate with changes induced by rapamycin.
Despite the difference in magnitude, the pattern of changes in translation after imatinib treatment correlated well with changes following rapamycin treatment (Pearson r
0.65, ), suggesting that mTOR activation could in principle account for much of v-Abl induced translational induction. In addition, rapamycin-induced changes in mRNA abundance correlated quite well with those induced by imatinib (Pearson r
0.47, ) suggesting that mTOR activation could also (perhaps indirectly) account for a substantial component of the change in mRNA abundance in these cells. However, there are isolated cases where imatinib and rapamycin caused discordant effects on the translation of specific genes, which will be discussed below.
Extensive Gene-Specific Reprogramming of Translation Following Abl Inhibition
Imatinib treatment induced gene-specific changes in mRNA abundance that generally paralleled the corresponding changes in translation (Spearman r
0.31, ). Despite this overall correlation, 109 of the 175 genes with significant increases in translation and 390 of the 591 genes with significantly reduced translation showed changes in mRNA abundance that were within one standard deviation (SD) of the mean for all detectable mRNAs (, purple lines represent 1 SD). Translational regulation thus appeared to be the principal mode of regulation for more than half of the ~770 genes either activated or repressed by v-Abl (see Tables S2
Genes with functional roles in translation were highly over-represented (p<10−32
, GO term enrichment) among the genes whose ribosome density decreased upon addition of imatinib, very likely reflecting the activation of mTOR by v-Abl. Genes encoding mitochondrial proteins (p<10−11
, GO term enrichment) were also over represented in this group, suggesting that v-Abl activity can promote translation of genes involved in aerobic respiration. In general, the translational response to rapamycin was similar but weaker than that to imatinib for both mitochondrial and translation apparatus genes (Tables S3
). Comparing average changes in translation across experiments confirms the general concordance between rapamycin and imatinib translational regulation for these subsets with one notable exception; C1qbp (Complement 1q-binding protein) is translationally induced by imatinib treatment while rapamycin has the opposite effect (, Fig. S3A
). C1qbp is a mitochondrial/cell surface protein implicated in promoting tumorigenesis 
Heatmaps depicting changes in translation upon imatinib and rapamycin treatment for genes either significantly translationally activated (A,C) or repressed (B,D) by v-Abl.
Genes involved in antigen processing and presentation were also significantly over-represented among the transcripts whose translation was inhibited by imatinib (p<0.025, KEGG pathways, Tables S2
). A similar pattern in response to rapamycin treatment (Fig. S3A
) suggests that the inhibition of MHCI antigen presentation canonically seen upon rapamycin treatment is due not only to reduced translation in general, but also to specific translational regulation of proteins participating in the antigen presentation pathway, such as Tap1, Erap1/Arts and constituents of the proteasome.
Three genes activated by v-Abl caught our attention due to their involvement in leukemia and cancer: Hmga1 (NM_016660), hnRNP A1 (NM_010447), and Hsp90ab1. Both hnRNP A1 and Hsp90ab1 have previously been reported to play important roles in BCR-ABL1-mediated leukemogenesis 
. Although HMGA1 is a known oncogene, to our knowledge, it has not previously been implicated in v-Abl–mediated transformation. All three of these genes displayed almost complete translational arrest upon imatinib treatment (, Fig. S1
). The abundance of these mRNAs also decreased, but only by roughly 1.5 to 2 fold (, Fig. S1
, Fig. S3B
). These data suggest that v-Abl activates the expression of these genes predominantly by promoting their translation. Translation of hnRNP A1 and Hsp90ab1 was also significantly translationally repressed by rapamycin treatment, but the change in Hmga1 translation upon rapamycin treatment was not significant by SAM (local FDR 88%, , Fig. S3A
), and was less than one standard deviation from the mean, suggesting that mTOR activation is responsible at least in part for the v-Abl dependent translational induction of Hsp90ab1 and hnRNP A1, but not significantly for translational induction of Hmga1 (). The abundance of all three transcripts changed only slightly in response to treatment (, Table S3B
, Fig. S1
Although the predominant effect of v-Abl inhibition was to decrease translation, ~160 genes were translationally activated upon imatinib treatment with a striking enrichment for genes encoding nuclear proteins (p<10−11, GO term enrichment) and proteins involved in cytoskeletal reorganization, particularly integrin activation.
Nuclear proteins translationally repressed by v-Abl included transcription factors implicated in repressing the IgH locus (Ezh2 (NM_007971), Cux1/Cutl1 (NM_009986), Bach2 (NM_007521), perhaps with MafG (NM_010756)) 
, activating RAG and/or promoting B cell differentiation (Foxp1 (NM_053202), Lef1 (NM_010703), Bcl6 (NM_009744)) 
, and tumor suppression (Ssbp2 (NM_024186), Mll5 (NM_026984)) 
. Others had no previously identified role in B cell development but could putatively enhance κlocus activation due to their ability to interfere with Id protein inhibition of E2A products (Mef2a (NM_013597), Tcf12 (NM_011544), Zfp238 (NM_013915)) (see Table S2
for a complete list).
Discordant yet significant regulation of multiple genes with shared functional roles was observed for genes involved in actin cytoskeleton reorganization and the phosphoinositide 3-kinase (PI3K) pathway, both of which were enriched in the v-Abl-repressed gene set (p<0.037, KEGG pathways, Table S4
). The PI3K pathway (induced downstream of the pre-BCR) was over-represented in the v-Abl-repressed gene set; they included Itpkb (Inositol 1,4,5-trisphosphate 3-kinase B) (NM_001081175), Dgk δ (DAG kinase δ) (NM_177646), Pik3ca (PI3K catalytic subunit 110α) (NM_008839), and Calm1 (Calmodulin 1) (NM_009790), although a number of molecules in convergent pathways were translationally induced by v-Abl, including PLCγ2 (NM_172285), Rac2 (NM_009008), Nek3 (NM_011848), and AI586015/Stap1/BRDG1 (BCR Downstream Signaling 1) (NM_019992) 
. The set of genes repressed by v-Abl was enriched for involvment in actin cytoskeleton reorganization, represented by Ppp1cb (NM_172707), Cyfip2/Pir121 (NM_133769), Iqgap1 (NM_016721), Itga4 (NM_010576), Pik3ca (NM_008839), Wasf2/WAVE2 (NM_153423), Rassf5/RapL (NM_018750). Additional v-Abl-repressed genes potentially involved in cytoskeletal reorganization but not included in the KEGG annotations include Stk4/Mst1 (NM_021420) 
, Ptpn12 (NM_011203) 
, Pacsin2 (NM_011862) 
, Triobp (NM_138579) 
, and Sipa1l1 (NM_172579) 
(). A few regulators of actin reorganization, Was/Wasp (NM_009515) 
, Coro1A (Coronin-1) (NM_009898) 
, D10Wsu52e/FAAP (NM_145422) 
, and Pstpip1 (NM_011193) 
, were translationally induced by v-Abl.
A summary of the contribution of v-Abl translational re-programming to differentiation arrest and leukemogensis.
The significant repression of Cyfip2/Pir121, Wasf2/WAVE2, Rassf5/RapL, Stk4/Mst1 and Itga4 by v-Abl is notable due to their concordant involvement in TCR-mediated integrin activation 
. Conversely translation of another molecule important for hematopoietic cell adhesion, Cd44 (NM_009851), is induced by v-Abl 
. The discordant regulation of functionally related genes demonstrates the gene-specific nature of v-Abl’s translational regulation, and suggests its involvement in re-wiring pathways affecting BCR signaling and adhesion.
In contrast to genes encoding mitochondrial proteins and the translation apparatus, those involved in cytoskeletal reorganization, BCR signaling, and transcription factors had more varied responses to rapamycin ( and Fig. S3A
). None of the genes translationally activated by imatinib treatment were significantly induced by rapamycin treatment at the 10% FDR cutoff. The gene whose translation was most differentially regulated in response to imatinib was Cyfip2, a gene whose overexpression single-handedly increases TCR-mediated adhesion 
Using the same significance criteria (10% FDR, SAM), more than twice as many genes showed significant changes at the level of mRNA abundance than at the level of translation, yet there is limited overlap between these two groups. Fewer than half of the translationally regulated genes distinguished by GO term or KEGG pathway analysis discussed above also significantly changed at the level of mRNA abundance, stressing the large amount of information missed by only considering changes in mRNA abundance (, Fig. S3B
). Over half of the transcription factors with significant translational changes did not significantly change at the level of mRNA abundance. Moreover, GO term and KEGG pathway analysis failed to identify antigen processing, lysosome constituents, the phosphatidylinositol signaling system or regulation of actin cytoskeleton as enriched categories using genes significantly changing only at the level of mRNA abundance (using either the same number of genes as used for translation, or the same 10% FDR cutoff, SAM). A table of all combined data is provided (Data S1
v-Abl orchestrates a dynamic, gene-specific program of post-transcriptional regulation of the basic cell machinery necessary for rapid proliferation, increasing expression of oncogenes, as well as re-wiring signaling pathways that govern cell fate decisions (). The selective translational activation or repression of diverse genes in response to imatinib treatment suggests that extensive reprogramming of translation is likely to play an important part in the ability of activated Abl to arrest B cell differentiation and promote leukemic transformation.