It has recently become clear that in some circumstances Lamarck may have been right. There are a growing number of examples where a clear case can be made for the inheritance from parent to offspring of environmentally acquired gene expression changes. We have developed a sensor in C. elegans in which RNAi-induced silencing of a GFP transgene is robustly inherited for multiple generations in the absence of the initial RNAi trigger. The visible nature of this phenotype provides an exquisitely sensitive system, whereby we can separate individual animals according to their silencing status and measure effects in these distinct groups. Using this approach, we show that the histone H3 lysine 9 (H3K9) trimethyltransferase SET-25 and the putative histone methyltransferase SET-32 are required in the RNAi-exposed generation for effective transmission of transgene silencing, but are dispensable for silencing inheritance in subsequent generations. This indicates that SET-25 and SET-32 are required for the establishment of a transgenerational silencing signal but not for long-term maintenance of this signal, suggesting that transgenerational epigenetic inheritance is a multi-step process with distinct genetic requirements for establishment and maintenance of heritable silencing. We further characterize SET-25 and SET-32 to gain insights into the nature of the silencing establishment signal. We show that SET-25 is expressed in nuclei in the mitotic zone and embryos, whilst SET-32 is expressed in both the cytoplasm and nuclei throughout the entire germline. Additionally, immunofluorescence and quantitative mass spectrometry reveal aberrant H3K9me3 in
set-25 and
set-32 mutants. Previous studies have implicated secondary siRNAs as the effector molecules of heritable silencing. We separated animals according to their inherited silencing status (GFP-silenced or GFP-expressing) and sequenced small RNA populations in
set-25 and
set-32 mutants, and in mutants of the nuclear RNAi inheritance machinery. We found that the abundance of secondary siRNAs does not correlate well with silencing phenotypes in the sequenced generation or in the next generation, suggesting that the current mechanistic models of epigenetic inheritance are incomplete.