On 2016 Feb 26, Martin Holcik commented:

we posted the following comment on the Science website, but reposting it here as well:

Internal initiation of translation writ large. S.D. Baird, Z. King and M. Holcik

In the article “Systematic discovery of cap-independent translation sequences in human and viral genomes” (15 January 2016 )(1) Weingarten-Gabby et al. surveyed a large number of mRNA-derived 210 bp segments for their ability to internally initiate translation. Our own search for IRESes with a secondary structure similar to that of a known IRES from XIAP mRNA identified two new IRESes but, surprisingly, with little notable structure similarity (2). Was that serendipity? We therefore wondered how prevalent IRESes are within the human transcriptome, and asked how many IRESes are there if we test 10 random UTRs. We selected 10 random 5'-UTRs from the UTRdb (3) using a perl script with a randomization function and tested their ability to initiate internal initiation using a previously characterized gal/CAT bicistronic reporter system (4). We identified one UTR from ZNF584 that showed bona fide IRES activity (without any cryptic promoter or splicing activity) but missed two IRESes discovered by the systematic survey (TEX2 and ZNF146). This is because the UTRs were improperly annotated at the time of our cloning; theoretically we had 3 genes out of 10 with IRES elements, or 30% of transcriptome, which is a bit higher than the 10% reported by the survey. While Weingarten-Gabby et al.'s test of small segments may have missed large structural IRESes like HCV, their discovery of so many IRES elements shows the IRES mechanism as a common feature, and that it frequently occurs within the coding sequence points to the ability of a cell to selectively express a more complex proteome than is evident by the transcriptome. Should the IRES elements be annotated into the RefSeq sequence features? Are there sequences that will recruit the ribosome, and coupled with RNA modification (such as recently described adenosine methylation or hydroxymethylation of cysteine (5) also regulate translation? The transcriptome to the proteome is not a simple step as ribosomal proteins and RNA binding proteins which control IRES activity will control which proteins are expressed. Nevertheless, these new insights suggest that internal initiation on cellular mRNAs writ large.

1. S. Weingarten-Gabbay et al., Comparative genetics. Systematic discovery of cap-independent translation sequences in human and viral genomes. Science. 351(6270). aad4939. doi: 4910.1126/science.aad4939. Epub 2016 Jan 4914. (2016).
2. S. D. Baird, S. M. Lewis, M. Turcotte, M. Holcik, A search for structurally similar cellular internal ribosome entry sites. Nucleic Acids Res. 35, 4664-4677. (2007).
3. F. Mignone et al., UTRdb and UTRsite: a collection of sequences and regulatory motifs of the untranslated regions of eukaryotic mRNAs. Nucleic Acids Res. 33, D141-146. (2005).
4. M. Holcik et al., Spurious splicing within the XIAP 5' UTR occurs in the Rluc/Fluc but not the {beta}gal/CAT bicistronic reporter system. RNA 11, 1605-1609 (2005).
5. B. Delatte et al., RNA biochemistry. Transcriptome-wide distribution and function of RNA hydroxymethylcytosine. Science. 351, 282-285. doi: 210.1126/science.aac5253. (2016).

On 2016 Feb 04, Ivan Shatsky commented:

It is a fruitful idea to use a high- throughput assay to fish out sequences that regulate translation initiation. I like this idea. It may result in very useful information provided that the experimental protocol is correctly designed to reach the goal of a study. However, while reading the text of the article I had an impression that the authors did not make a clear difference between the terms “IRES-driven translation” and “cap-independent” translation. In fact, cap-independent mechanisms may be of two kinds: a mechanism that absolutely requires the free 5’ end of mRNA (see e.g. Terenin et al. 2013. Nucleic Acids Res. 41(3):1807-16 and references therein and Meyer et al. 2015. Cell 163(4): 999-1010 ) and that which is based on internal initiation. Only in the latter case a 5’ UTR starts penetrating the mRNA binding channel of ribosomes with an internal segment of the mRNA rather than with a free 5’ end. Consequently, the experimental design should be distinct for these two modes of cap-independent translation. The method of bicistronic constructs used by the authors is suitable exclusively to identify IRES-elements. However, this approach is sufficiently reliable when it is employed in the format of bicistronic RNAs transfected into cultured cells. It is repeatedly shown that the initial format of bicistronic DNAs is extremely prone to almost unavoidable artifacts (for literature, see ref. 48 in the paper and the review by Jackson, R.J. The current status of vertebrate cellular mRNA IRESs. Cold Spring Harb Perspect Biol, 2013; 5). The control tests to reveal these artifacts which are still used (unfortunately!) by many researchers are not sensitive enough to detect formation of few percents of monocistronic mRNAs. (To this end, one should perform precise and laborious experiments which are not realistic in the case of high-throughput assays.). The capping of these aberrant mono- mRNAs can produce a dramatic stimulation of their translation activity (20-100 fold, depending on cell line). Therefore, even few percents of capped mono-mRNAs may result in a high activity of the reporter as compared to an almost zero activity of empty vector (see Andreev et al. 2009. Nucleic Acids Res.37(18):6135-47 and references therein). Real-time PCR assessment of mRNA integrity (Fig. S4) is an easy way to miss these few percents of aberrant transcripts. The other concern is genome-wide cDNA/gDNA estimation. The ratio for e.g. “c-myc IRES” is 2^-1.6 which is roughly 1/3 (Fig. S3). Does this mean that 2/3 of c-myc transcripts are monocistronic rather than bicistronic? I had a general impression that the authors were not aware of serious pitfalls inherent to the method of bicistronic DNA constructs and simply adapted this method to their high throughput assay. At least, I did not find citations of papers that discussed this important point.

 The data in section Supplementary materials (Figs. S5 and S6) give us expressive and compelling  evidence of such kind of artifacts: indeed, some  174 nt long fragments from the EMCV IRES possessed an IRES activity. Moreover, one of them with GNRA motif had the activity similar to that for the whole EMCV IRES (!?).  This result is in an absolute contradiction with our current knowledge on this picornaviral IRES, one the best studied IRES elements! Parts of the EMCV IRES are known to have no activity at all! Thus, the most plausible explanation is that the EMCV fragments harbor cryptic splice sites. The same is true for other picornavirus IRESs examined in these assays. The HCV IRES tested by the authors in the same experiments worked only as a whole structure (Fig.S6B), in a full agreement with data of literature. However, this result may not encourage us as it just means that the data obtained in this study may be a mixture of true regulatory sequences with artifacts.   
  We should keep in mind that the existence of viral IRES-elements is a firmly established fact. They have a complex and highly specific organization with well defined boundaries and THEY ARE ONLY ACTIVE AS INTEGRAL STRUCTURES. The minimal size of IRESs from RNAs of animal viruses is  >300 nts. Their shortening inactivates them and therefore, they cannot be studied with cDNA fragments of 200 nt long or less. Thus, I think it was a mistake to mix viral IRESs with cellular mRNA sequences. As to cellular IRESs, none of them has been characterized and hence we do not know what they are and whether they even exist. For none of them has been shown that they do not need a free 5’ end of mRNA to locate the initiation codon. Some of them have already been disproved (c-Myc, eIF4G, Apaf-1 etc.).  By the way, I do not know any commercial vector that employs a cellular IRES. Thus, I think that we should first find adequate tools to identify cellular IRESs, characterize several of them, and only afterwards we may proceed to transcriptome-wide  searching for cellular IRESs.

On 2016 Feb 04, Ivan Shatsky commented:

It is a fruitful idea to use a high- throughput assay to fish out sequences that regulate translation initiation. I like this idea. It may result in very useful information provided that the experimental protocol is correctly designed to reach the goal of a study. However, while reading the text of the article I had an impression that the authors did not make a clear difference between the terms “IRES-driven translation” and “cap-independent” translation. In fact, cap-independent mechanisms may be of two kinds: a mechanism that absolutely requires the free 5’ end of mRNA (see e.g. Terenin et al. 2013. Nucleic Acids Res. 41(3):1807-16 and references therein and Meyer et al. 2015. Cell 163(4): 999-1010 ) and that which is based on internal initiation. Only in the latter case a 5’ UTR starts penetrating the mRNA binding channel of ribosomes with an internal segment of the mRNA rather than with a free 5’ end. Consequently, the experimental design should be distinct for these two modes of cap-independent translation. The method of bicistronic constructs used by the authors is suitable exclusively to identify IRES-elements. However, this approach is sufficiently reliable when it is employed in the format of bicistronic RNAs transfected into cultured cells. It is repeatedly shown that the initial format of bicistronic DNAs is extremely prone to almost unavoidable artifacts (for literature, see ref. 48 in the paper and the review by Jackson, R.J. The current status of vertebrate cellular mRNA IRESs. Cold Spring Harb Perspect Biol, 2013; 5). The control tests to reveal these artifacts which are still used (unfortunately!) by many researchers are not sensitive enough to detect formation of few percents of monocistronic mRNAs. (To this end, one should perform precise and laborious experiments which are not realistic in the case of high-throughput assays.). The capping of these aberrant mono- mRNAs can produce a dramatic stimulation of their translation activity (20-100 fold, depending on cell line). Therefore, even few percents of capped mono-mRNAs may result in a high activity of the reporter as compared to an almost zero activity of empty vector (see Andreev et al. 2009. Nucleic Acids Res.37(18):6135-47 and references therein). Real-time PCR assessment of mRNA integrity (Fig. S4) is an easy way to miss these few percents of aberrant transcripts. The other concern is genome-wide cDNA/gDNA estimation. The ratio for e.g. “c-myc IRES” is 2^-1.6 which is roughly 1/3 (Fig. S3). Does this mean that 2/3 of c-myc transcripts are monocistronic rather than bicistronic? I had a general impression that the authors were not aware of serious pitfalls inherent to the method of bicistronic DNA constructs and simply adapted this method to their high throughput assay. At least, I did not find citations of papers that discussed this important point.

 The data in section Supplementary materials (Figs. S5 and S6) give us expressive and compelling  evidence of such kind of artifacts: indeed, some  174 nt long fragments from the EMCV IRES possessed an IRES activity. Moreover, one of them with GNRA motif had the activity similar to that for the whole EMCV IRES (!?).  This result is in an absolute contradiction with our current knowledge on this picornaviral IRES, one the best studied IRES elements! Parts of the EMCV IRES are known to have no activity at all! Thus, the most plausible explanation is that the EMCV fragments harbor cryptic splice sites. The same is true for other picornavirus IRESs examined in these assays. The HCV IRES tested by the authors in the same experiments worked only as a whole structure (Fig.S6B), in a full agreement with data of literature. However, this result may not encourage us as it just means that the data obtained in this study may be a mixture of true regulatory sequences with artifacts.   
  We should keep in mind that the existence of viral IRES-elements is a firmly established fact. They have a complex and highly specific organization with well defined boundaries and THEY ARE ONLY ACTIVE AS INTEGRAL STRUCTURES. The minimal size of IRESs from RNAs of animal viruses is  >300 nts. Their shortening inactivates them and therefore, they cannot be studied with cDNA fragments of 200 nt long or less. Thus, I think it was a mistake to mix viral IRESs with cellular mRNA sequences. As to cellular IRESs, none of them has been characterized and hence we do not know what they are and whether they even exist. For none of them has been shown that they do not need a free 5’ end of mRNA to locate the initiation codon. Some of them have already been disproved (c-Myc, eIF4G, Apaf-1 etc.).  By the way, I do not know any commercial vector that employs a cellular IRES. Thus, I think that we should first find adequate tools to identify cellular IRESs, characterize several of them, and only afterwards we may proceed to transcriptome-wide  searching for cellular IRESs.

On 2016 Feb 26, Martin Holcik commented:

we posted the following comment on the Science website, but reposting it here as well:

Internal initiation of translation writ large. S.D. Baird, Z. King and M. Holcik

In the article “Systematic discovery of cap-independent translation sequences in human and viral genomes” (15 January 2016 )(1) Weingarten-Gabby et al. surveyed a large number of mRNA-derived 210 bp segments for their ability to internally initiate translation. Our own search for IRESes with a secondary structure similar to that of a known IRES from XIAP mRNA identified two new IRESes but, surprisingly, with little notable structure similarity (2). Was that serendipity? We therefore wondered how prevalent IRESes are within the human transcriptome, and asked how many IRESes are there if we test 10 random UTRs. We selected 10 random 5'-UTRs from the UTRdb (3) using a perl script with a randomization function and tested their ability to initiate internal initiation using a previously characterized gal/CAT bicistronic reporter system (4). We identified one UTR from ZNF584 that showed bona fide IRES activity (without any cryptic promoter or splicing activity) but missed two IRESes discovered by the systematic survey (TEX2 and ZNF146). This is because the UTRs were improperly annotated at the time of our cloning; theoretically we had 3 genes out of 10 with IRES elements, or 30% of transcriptome, which is a bit higher than the 10% reported by the survey. While Weingarten-Gabby et al.'s test of small segments may have missed large structural IRESes like HCV, their discovery of so many IRES elements shows the IRES mechanism as a common feature, and that it frequently occurs within the coding sequence points to the ability of a cell to selectively express a more complex proteome than is evident by the transcriptome. Should the IRES elements be annotated into the RefSeq sequence features? Are there sequences that will recruit the ribosome, and coupled with RNA modification (such as recently described adenosine methylation or hydroxymethylation of cysteine (5) also regulate translation? The transcriptome to the proteome is not a simple step as ribosomal proteins and RNA binding proteins which control IRES activity will control which proteins are expressed. Nevertheless, these new insights suggest that internal initiation on cellular mRNAs writ large.

1. S. Weingarten-Gabbay et al., Comparative genetics. Systematic discovery of cap-independent translation sequences in human and viral genomes. Science. 351(6270). aad4939. doi: 4910.1126/science.aad4939. Epub 2016 Jan 4914. (2016).
2. S. D. Baird, S. M. Lewis, M. Turcotte, M. Holcik, A search for structurally similar cellular internal ribosome entry sites. Nucleic Acids Res. 35, 4664-4677. (2007).
3. F. Mignone et al., UTRdb and UTRsite: a collection of sequences and regulatory motifs of the untranslated regions of eukaryotic mRNAs. Nucleic Acids Res. 33, D141-146. (2005).
4. M. Holcik et al., Spurious splicing within the XIAP 5' UTR occurs in the Rluc/Fluc but not the {beta}gal/CAT bicistronic reporter system. RNA 11, 1605-1609 (2005).
5. B. Delatte et al., RNA biochemistry. Transcriptome-wide distribution and function of RNA hydroxymethylcytosine. Science. 351, 282-285. doi: 210.1126/science.aac5253. (2016).