Tag Archives: WAGO

Interacting small RNA pathways in worms 5: Global Genome Surveillance

As discussed previously, in C. elegans more than 15,000 21U-RNAs are expressed from two large clusters on chromosome IV. As very few of these piRNAs exhibit perfect sequence complementarity with other endogenous sequences within in the C. elegans genome, it’s been difficult to deduce the targets and functions of this system. New studies from the labs of Eric Miska and Craig Mello have significantly advanced our understandings of piRNAs in C. elegans. Bagijn et al. and Lee et al. have confirmed a previous supposition that 21U-RNAs can base-pair with imperfectly complementary sequences. This less stringent base-pairing opens up the entire transcriptome to possible piRNA-mediated regulation. Bagijn et al, Ashe et al, and Shirayama et al. have further characterised the effector pathways downstream of PRG-1/piRNA targeting. Together these papers outline a finely tuned genomic surveillance mechanism capable of discerning self and non-self transcripts.


Previous studies had shown that 21U-RNAs act upstream of 22G-RNAs to repress the activity of Tc3 transposons. Lee et al. therefore investigated whether the expression of 22G-RNAs was altered in worms mutant for the worm Piwi protein PRG-1. When all 22G-RNAs were considered together, they couldn’t observe a correlation between PRG-1 activity and 22G-RNAs. However, 22G-RNAs can be divided into four different pathways defined by the different AGOs with which they complex. When those associated with 2 different ‘Eri’ pathways, and the CSR-1 22G-RNAs were discounted, Lee et al. observed that WAGO-associated 22G-RNAs tended to be depleted in prg-1 mutants. mRNAs targeted by WAGO 22G-RNAs showed a tendency towards upregulation in prg-1 mutants, consistent with a repressive role for these small RNAs.

22G-RNAs are not evenly distributed on their target mRNAs. Instead, hotspots occur where 22G-RNA species are far more common than at other points on the same mRNA. Lee et al. postulated that these hotspots could be regions where PRG-1/21U-RNA complexes recruit RNA-dependent RNA polymerases to generate 22G-RNAs. If this is as widespread a phenomenon as suggested by the depletion of 22G-RNAs in prg-1 mutants, imperfect base-pairing of 21U-RNAs to their targets must be occurring, as only 29 WAGO targets show perfect complementarity to 21U-RNAs. Lee et al. therefore asked whether 22G-RNAs are enriched in mRNA coding regions with the potential for energetically favourable but imperfect base pairing to 21U-RNAs. Basing their parameters on the imperfect base-pairing observed for miRNA target interactions (in which strong base-pairing in a ‘seed’ region of nucleotides 2-8, facilitates less perfect pairing between the rest of the sequences), Lee et al. determined the level of 22G-RNAs within a 100nt window around potential 21U-RNA binding sites in wild-type and prg-1 mutant worms.  Allowing 2 mismatches and 1 G:U wobble pair outside the seed region, and at most 1 G:U pair within the seed (parameters that meant that more than 50% of genes contained potential 21U-RNA binding sites), the researchers found that 22G-RNAs mapping to within ± 50nt of the 21U-RNA binding sites on WAGO mRNA targets, were 3 fold enriched in wildtype worms relative to the prg-1 mutants. A smaller enrichment of 1.4 fold was also observed for CSR-1 mRNA targets. The level of 22G-RNA enrichment correlated with the expression levels of 21U-RNAs. Lee et al. confirmed the importance of pairing within the seed region by comparing this data with another set of potential 21U-RNA targets with similar total mismatches but poor seed pairing, which showed little 22G-RNA enrichment.

Bagijn et al performed slightly different analyses, which came to similar conclusions. They identified 681,746 potential binding sites for 16,003 21U-RNAs if 3 mismatches are allowed (not biased away from seed region). They then looked to see what proportions of these sites 22G-RNAs also mapped to, finding 1.6, 1.4, 1.5 and 1.2 fold 22G-RNA enrichments relative to matched control sequences, at 0, 1, 2 and 3 mismatch sites respectively. As with Lee et al. the levels of 22G-RNAs correlated with 21U-RNA abundance.

These analyses suggest that 21U-RNAs are targeting a large proportion of the germline transcriptome, including both protein-coding genes and transposons. In relation to the repression of Tc3 (discussed in the previous post), Bagijn et al. identified three piRNAs with imperfect complementarity to sequences in the TIRs, perhaps clearing up some of the ambiguities from previous studies. By using their data on 22G-RNA enrichments at potential 21U-RNA binding sites, they were able to rank the likelihood of piRNA regulation for transcripts. Six of eleven candidates that showed strong reductions in 22G-RNAs in prg-1 mutants also showed statistically significant transcriptional upregulation (including both transposons and protein-coding genes).

piRNA sensors

Most of the papers under discussion used transgenic ‘sensor’ lines to dissect the mechanisms of PRG-1/21U-RNA mediated silencing. Bagijn et al generated a ‘piRNA sensor’ in which a GFP – histone H2B fusion gene with a sequence complementary to a known 21U-RNA is inserted into the C. elegans genome. In the control sensor line (in which the reverse complement of the 21U-RNA is included instead), the GFP reporter protein is expressed in germline nuclei. In the piRNA sensor line the transgene is silenced. This silencing is dependent on PRG-1, as the sensor was desilenced in prg-1 mutants. Bagijn et al detected 22G-RNAs mapping to the piRNA sensor mRNA close to the piRNA target site, that weren’t present in the control sensor or in prg-1 mutants. As levels of both the piRNA sensor pre-mRNA and mRNA were raised in prg-1 mutants, it appears that silencing occurs at the level of transcription, and possibly post-transcriptionally as well.

By crossing the piRNA sensor line into various mutant worm strains, Bagijn et al showed that the silencing involved many of the 22G-RNA pathway components discussed previously, including the helicase DRH-3, the RdRPs EGO-1 and RRF-1, and various WAGOs. As has been discussed previously, many of the WAGO proteins have overlapping functions and hence display partial redundancy when impaired. However, wago-9 (also known as hrde-1) appears to be an especially important AGO in the silencing process; the sole single WAGO mutant to cause desilencing phenotypes. All these 22G-RNA components act downstream of PRG-1, as 21U-RNAs are still present in these mutants while 22G-RNAs fail to accumulate. In prg-1 mutants both classes of small RNAs are affected.

Although in other animals Piwi proteins are known to act by slicing it’s targets, transgenes encoding PRG-1 with a mutated endonuclease motif could rescue the desilencing observed in prg-1 mutants. By mutating the piRNA target sequence in the sensor, Bagijn et al. then showed that 2 mismatches are tolerated for piRNA mediated silencing to occur. The residues changed could be anywhere in the sequence including at positions 10 and 11 – the normal site for Piwi slicing activity.

Lee et al. undertook very similar experiments to Bagijn et al, generating a piRNA sensor line and confirming that mismatches with the piRNA target site were tolerated, including at the normal slicing site. However there is a crucial difference between the two sets of experiments using piRNA sensors. In Bagijn et al’s study, PRG-1 was required continuously for silencing to be maintained through the generations. That is to say that an already silenced sensor would be desilenced when crossed into a prg-1 mutant line. Lee et al’s piRNA sensor only required PRG-1 for silencing to be initiated, but not for it to be maintained. When Lee et al’s silenced sensor was outcrossed into prg-1 mutants, GFP expression was not activated, whereas if the transgene was introduced into prg-1 mutants it was not silenced. This important disparity may well be caused by differences in the (broadly similar) compositions of the sensor transgenes, and will be discussed later.

Long-term heritability.

The PRG-1-dependent silencing of the piRNA sensors is incredibly long lived, being transmissable through many generations. This type of epigenetically heritable effect has been previously observed with various RNAi paradigms in C. elegans, with disagreement over whether epigenetic transmission is primarily through inheritance of small RNAs, or via chromatin modifications, or both (see these earlier posts: 1, 2 ). Ashe et al. generated another transgenic sensor line to monitor the heritability of dsRNA-induced RNAi. They found that both the piRNA-mediated silencing pathway and the heritable RNAi silencing pathway converge on a common group of nuclear factors responsible for 22G-RNA mediated silencing. These include two proteins of unknown function, NRDE-1 and NRDE-2, known to mediate nuclear RNAi via interactions with nascent transcripts; the heterochromatin protein 1 orthologue, HPL-2; the nuclear AGO WAGO-9; and a SET-domain protein, SET-25, thought to be a methyltransferase responsible for histone H3 lysine-9 trimethylation (a repressive chromatin modification). This collection of downstream effectors suggests that the heritability of both RNAi and piRNA mediated silencing rests on epigenetically stable chromatin marks.

However, if a silenced piRNA sensor strain is crossed with a strain expressing a independent GFP transgene, a dominant silencing of both transgenes occurs. This trans­-acting silencing effect is most likely mediated by secondary siRNAs (ie 22G-RNAs). When small RNA populations are assayed after the induction of RNAi silencing, a large diversity of small RNAs with little 5’ bias (ie Dicer products) were found to target the sensor. However, by the 4th generation, targeting small RNAs had clarified themselves into antisense 22G-RNAs. These 22G-RNAs appeared to be generated de novo in each generation. It therefore appears likely that epigenetic silencing is achieved by the inheritance of both small RNAs and chromatin marks, is reaffirmed in each generation, and acts at both transcriptional and post-transcriptional levels.


In the previously discussed papers, transgenes that were engineered to contain 21U-RNA binding sites were actively silenced in C. elegans. Shirayama et al. describe how transgenes containing non-endogenous genes can be silenced even in the absence of perfect piRNA recognition sequences. They found that when transgene fusions of gfp  and endogenous genes were inserted into the genome, they were occasionally completely silenced, whilst sometimes exactly the same construct, inserted into the same genomic site, was expressed. When a silent line was crossed to an expressing line, the transgene was invariably silenced in 100% of progeny. This dominant trans-acting silencing was heritable through many generations.  Shirayama et al termed this phenomenon RNA-induced epigenetic silencing (RNAe)(rather forestalling there own explanation of it’s causes).

By analysing levels of pre-mRNAs and mRNAs in silenced and active lines, Shirayama et al. found that silencing occurs at both the transcriptional and post-transcriptional levels. Silenced alleles were activated when crossed into lines mutant for various factors involved in repressive chromatin formation (the polycomb group proteins MES-3 + 4, and heterochromatin protein HPL-2). Whilst ChIP experiments demonstrated that silenced transgenes were enriched for histone H3K9 trimethylation. Hence RNAe involves transcriptional silencing at the level of chromatin formation.

The trans- acting nature of RNAe suggested a small RNA component. Crossing silenced alleles into lines mutant for factors known to function in the WAGO 22G-RNA pathway (rde-3,+mut-7) resulted in desilencing. Similarly transgenes were desilenced in worms mutant for the nuclear AGO encoding wago-9; effects which could be enhanced by mutations in additional wago genes. Sequencing of small RNAs from silenced worm gonads revealed a strong accumulation of 22G-RNAs targeted against gfp. The gfp genes were always combined with endogenous genes in the transgene constructs. Interestingly, 22G-RNAs were not produced against this part of the transgene, suggesting a mechanism protecting against the silencing of endogenous genes.

When a silenced transgene is crossed into prg-1 mutants it does not become reactivated, however when Shirayama et al performed transgenesis directly into prg-1 mutants silencing failed to occur. This demonstrates that PRG-1 is required for the initiation of RNAe and not for it’s maintenance (in agreement with the findings of Ashe et al and Lee et al). Although Shirayama et al did not find the exact piRNA recognition sites triggering this silencing, they did apparently identify a number of candidate piRNAs, the recognition sites of which displayed heightened expression of 22G-RNAs. It seems therefore that the recognition of foreign nucleic acids and their suppression via RNAe is the primary function of the PRG-1/21U-RNA system.


All of these studies are in general agreement about the basic dynamics of piRNA silencing in C. elegans. piRNA target recognition is mismatch tolerant. PRG-1 does not act by endonuclease activity. Instead, upon recognition it recruits a RdRP complex to generate 22G-RNAs against sequence adjacent to the 21U-RNA binding site. These 22G-RNAs complex with WAGOs (especially WAGO-1 and WAGO-9) that effect genetic silencing at both the level of repressive chromatin and post-transcriptionally. PRG-1/21U-RNA recognition triggers a self-sustaining WAGO-22G-RNA dependent silencing.

However, there is still a question about the targets of this system. With a few mismatches tolerated the known repertoire of 21U-RNAs can target the whole genome. Obviously, it’s not all repressed, so just what is happening? The computational analyses of Bagijn et al suggested that protein-coding mRNA sequences showed signs of bias against potential piRNA recognition sequences, whilst transposon and pseudogene sequences did not.  They also found evidence to suggest that transposon insertions between the conserved 21U-RNA promoter motif and the 21U-RNA gene itself generate new transposon-targeting piRNAs. These findings lead to a model of C. elegans piRNA biology with similarities to that observed in Drosophila; piRNA clusters rich in transposon sequence, generating piRNAs primarily used to repress transposon mobilisation in the germline, with endogenous genes being selected to avoid piRNA mediated repression. Lee et al. did find similar trends but were not convinced that the numbers were strong enough, instead emphasising that many of the findings in these experiments suggest the existence of an anti-silencing pathway.


When Shirayama et al. used dsRNA to silence gfp containing transgenes that hadn’t been silenced by RNAe, they were surprised to find the silencing induced was inherited very stably, whereas RNAi against gfp transgenic lines that had been produced by earlier methods never silenced as heritably. They then crossed these earlier transgenic lines to silenced gfp lines to test whether they were susceptible to trans-silencing. Instead they found that these lines would dominantly activate gfp expression. This suggests the presence of a dominant trans-acting mechanism competing against the trans-acting silencing mechanism.

Another line of evidence suggesting the existence of an anti-silencing pathway is the difference in the requirement for prg-1 for silencing between the piRNA sensors of Bagijn et al and Lee et al.  In the Bagijn et al sensor the 21U-RNA target site was flanked by endogenous sequences that are known targets of CSR-1 22G-RNAs. In contrast, in Lee et al’s sensor the piRNA recognition site was surrounded by sequences not targeted by this small RNA population.

A likely explanation for why the silencing of the Bagijn et al sensor required prg-1 in each generation, was that a CSR-1 based anti-silencing pathway counteracts the silencing induced by PRG-1 triggered WAGO mediated silencing. Likewise, CSR-1 22G-RNAs could be the trans-acting anti-silencing agents suggested in Shirayama et al’s experiments. Shirayama et al’s fusion transgenes were more or less susceptible to silencing depending on the endogenous gene fused to gfp, and as noted earlier, in general silencing associated 22G-RNAs were only enriched for the gfp genes and not against the endogenous sequences in the transgenes.

It therefore seems that an anti-silencing pathway exists that licenses ‘self’ transcripts, protecting them from WAGO 22G-RNA silencing. The CSR-1 pathway is the perfect candidate for this activity, as its’ associated 22G-RNAs are known to target thousands of germline expressed transcripts without inducing silencing. Further dissection of these pathways is obviously necessary to prove CSR-1’s role, and whether the anti-silencing pathway also acts downstream of PRG-1, or just antagonistically to it.

The Miska lab agrees about the existence of the licensing pathway, however they suggest that some of the results indicate the existence of another mechanism. Ashe et al. reported that their piRNA sensor could be stably silenced in a prg-1 independent manner, but only if it had been present in a heterozygous state for multiple generations. This is evidence of a mechanism for detecting unpaired chromatin during meiosis. Shirayama et al’s findings too are indicative of a mechanism for detecting unpaired alleles. This could have significant implications evolutionarily, as it may facilitate the phenotypic expression of recessive traits.

One can describe this self and non-self recognition system as a type of acquired genetic immune system. Although, as yet, the details are only understood in outline, it appears to be a exquisitely finely tuned, and effective system. Until recently no viruses were known to infect C. elegans. This system must be part of the reason why. There are still major questions about to what extent transposons are controlled by the piRNA system in C. elegans and how much of its’ functionality is devoted to the regulation of endogenous genes. As is evident from this series of posts, C. elegans has so many different AGOs, and so many different facets to it’s RNAi systems (germline + soma RNAi, nuclear + cytoplasmic RNAi, RNAe etc) that it can be very difficult to dissect them from each other and then be able to see them in the round. Perhaps the most interesting aspect of this work is the light shed on piRNA and RNAi systems in vertebrates and Drosophila. Secondary siRNA systems have not been found in these organisms, so this system of genetic immunity does not work in the same way in these other clades. The 21U-RNA triggering 22G-RNA generation mechanism has been likened to the amplification of secondary piRNAs by the ping-pong amplification system in Drosophila (a system missing in C. elegans). However, the most fascinating aspect of this work could be the light shed on vertebrate piRNA systems. Hundreds of thousands of ‘pachytene’ piRNAs, without known targets, are expressed in the mammalian germline. Do these piRNAs also tolerate mismatches when binding? And do they also mediate a genome surveillance system capable of detecting non-self transcripts? or unmatched chromatin? Are mammalian transcripts marked in some way to avoid this silencing? No doubt these C. elegans studies will invigorate research in mammalian piRNAs, and no doubt I’ll revisit it soon.

See also: Epigenetic Licensing of a Sex Determination Gene
The CSR-1 siRNA pathway gets more enigmatic

Bagijn MP, Goldstein LD, Sapetschnig A, Weick EM, Bouasker S, Lehrbach NJ, Simard MJ, & Miska EA (2012). Function, targets, and evolution of Caenorhabditis elegans piRNAs. Science (New York, N.Y.), 337 (6094), 574-8 PMID: 22700655

Lee HC, Gu W, Shirayama M, Youngman E, Conte D Jr, & Mello CC (2012). C. elegans piRNAs Mediate the Genome-wide Surveillance of Germline Transcripts. Cell, 150 (1), 78-87 PMID: 22738724

Ashe A, Sapetschnig A, Weick EM, Mitchell J, Bagijn MP, Cording AC, Doebley AL, Goldstein LD, Lehrbach NJ, Le Pen J, Pintacuda G, Sakaguchi A, Sarkies P, Ahmed S, & Miska EA (2012). piRNAs Can Trigger a Multigenerational Epigenetic Memory in the Germline of C. elegans. Cell, 150 (1), 88-99 PMID: 22738725

Shirayama M, Seth M, Lee HC, Gu W, Ishidate T, Conte D Jr, & Mello CC (2012). piRNAs Initiate an Epigenetic Memory of Nonself RNA in the C. elegans Germline. Cell, 150 (1), 65-77 PMID: 22738726

Interacting small RNA pathways in worms 2: 22G-RNAs

In two papers published in 2009 (Gu et al. and Claycomb et al), Craig Mello’s group characterised 22G-RNAs. They found that they could be divided into two different functional classes based on the Argonaute proteins with which they are complexed – CSR-1 or WAGOs. The biogenesis of both groups utilises common factors, but each targets different classes of loci and fulfil different roles.

The starting point for the discovery of 22G-RNAs was the analysis of various mutant alleles of the gene drh-3 (which encodes a dicer-related helicase). Homozygous drh-3 alleles result in infertile worms, whilst RNAi targeting leads to worms dying in embryogenesis with defects in chromosome segregation and in the production of small RNA populations. Gu et al identified three partial loss of function drh-3 alleles in which the homozygous worms were viable at 20°C but infertile at 25°C. When analysing the small RNA populations present in these worms, they found that a prominent 22nt species of RNA was absent, whilst others were unaffected. Chemical analysis showed that these RNAs had 5’ guanosine residues, and were 5’ triphosphorylated (most small RNAs are 5’ monophosphorylated – as produced by the activity of the Dicer endonuclease). The 22nt RNAs were not sensitive to periodate – suggesting that, unlike piRNAs they are not 3’ modified.

Gu et al. then made libraries of all small RNAs from wild type and drh-3 mutant worms, and deep sequenced them. 21nt and 22nt RNAs accounted for 25% and 36% of the wild type reads, respectively. The 21nt RNAs were divided equally between those with 5’U and those with 5’G. ~ 60% of the 22nt reads had a 5’G. Both 21nt and 22nt 5’G containing RNAs were strongly depleted in the drh-3 mutants. Of all the endogenous siRNA reads in the wild type library (64% of the total), ~53% were antisense to protein-coding genes, whilst ~16% were derived from transposons and repetitive sequences and ~31% were from non-annotated loci. All of these endo-siRNAs were depleted in drh-3 mutants. To try and clarify matters, Gu et al. termed this population of 22nt drh-3-dependent endo-siRNAs with a bias towards 5’G, 22G-RNAs. (note: I’m not quite clear as to whether this included 21G and/or 22U populations as well).

By analysing various mutant lines, Gu et al. found that 22G-RNAs are present in the soma and the germline. However, they were especially enriched in the germline, and in oocytes (ie. maternally derived). In the soma 22G-RNAs appear to act downstream of the exogenous-RNAi pathway (which won’t be discussed further – I’ll concentrate on their roles in the germline).

Germline 22G-RNAs are independent of the exo-RNAi pathway. For instance, they were not depleted in dcr-1 (dicer) mutants, suggesting that their biogenesis is not triggered by dsRNA. The triphosphorylated 5′ end of 22G-RNAs, and the independence from dcr-1, suggested that their biosynthesis was dependent on RNA-dependent RNA polymerases (RdRPs).  Single mutants for the known RdRPs still expressed 22G-RNAs. However, in worms mutant for both rrf-1 and ego-1, the researchers found that they failed to accumulate. Immunoprecipitation experiments showed that DRH-3 interacted biochemically with both RRF-1 and EGO-1, as well as a tudor-domain containing protein, EKL-1. These four proteins make up the core RdRP complex responsible for the biosynthesis of 22G-RNAs in the germline.

Depletion of Argonaute proteins leads to a reduction of the small RNAs with which they complex. Gu et al. used worm lines mutant for multiple WAGO genes to get a picture of which AGOs mediated 22G-RNA function. They found that worms deficient in wago-1 showed a major reduction in germline 22G-RNAs, whilst a worm strain lacking all 12 wago genes (MAGO12) showed an even greater deficit.

Deep sequencing from worms mutant for drh-3, ekl-1, or from the rrf-1 ego-1 double mutants had near complete germline 22G-RNA deficits. However in the MAGO12 worms, only a subset of 22G-RNAs matching repeat elements, as well as some coding and non-annotated loci were absent. Gene-targeted 22G-RNAs were far less likely to be affected. Immunoprecipitation of WAGO-1 complexes revealed an enrichment for the repeat element biased subset, whilst the AGO CSR-1 was found to interact with a subset of  22G-RNAs that are antisense to germline-expressed protein coding genes (Claycomb et al. discussed next post). This bimodal distribution of 22G-RNA targets revealed that two distinct 22G-RNA pathways functioned in the germline. They both share a common biosynthesis pathway but differ in the AGOs with which they complex.

The WAGO-associated 22G-RNA pathway appears to act by silencing it’s targets. Those loci targeted by the most highly expressed 22G-RNAs were derepressed in drh-3 mutants. Transposons are a major target for the WAGO mediated system. 22G-RNAs matching repetitive elements were depleted in MAGO12 worms. By assaying the reversion rate of mutations caused by the insertion of the transposon Tc5, and by monitoring the transcription of the Tc1 and Tc3 transposons, Gu et al showed that transposons are derepressed in drh-3 mutants (I’d have preferred to see these effects in the MAGO12 mutants to definitively show that it’s only the WAGO associated subset required).

The WAGOs are a worm specific clade of AGOs which don’t seem to act by ‘Slicer’ endonuclease activity. This study showed a lot of redundancy amongst WAGOs with regard to their 22G-RNA associated roles. However, the authors expect there to be a number of distinct roles within this family of factors. WAGO-1 appears to be a crucial factor in these systems. Importantly, this paper showed the existence of a major Dicer-independent RNA based genome surveillance system. This system has the ability to silence transposons and other repetitive sequences. It also appears to act upon pseudogenes and ‘cryptic loci’, preventing detrimental transcription/translation. However, the details of this system’s targets were beyond of the scope of this first paper.

The next post will discuss CSR-1 associated 22G-RNAs, before we come to 21U-RNAs and the links between the three systems.

Gu W, Shirayama M, Conte D Jr, Vasale J, Batista PJ, Claycomb JM, Moresco JJ, Youngman EM, Keys J, Stoltz MJ, Chen CC, Chaves DA, Duan S, Kasschau KD, Fahlgren N, Yates JR 3rd, Mitani S, Carrington JC, & Mello CC (2009). Distinct argonaute-mediated 22G-RNA pathways direct genome surveillance in the C. elegans germline. Molecular cell, 36 (2), 231-44 PMID: 19800275

Claycomb JM, Batista PJ, Pang KM, Gu W, Vasale JJ, van Wolfswinkel JC, Chaves DA, Shirayama M, Mitani S, Ketting RF, Conte D Jr, & Mello CC (2009). The Argonaute CSR-1 and its 22G-RNA cofactors are required for holocentric chromosome segregation. Cell, 139 (1), 123-34 PMID: 19804758