Anti-viral RNAi transgenes are typically expressed constitutively ('always on') in all the plant's tissues - yet in plants and animals, injecting short interfering RNA (siRNAs) induces a similar protective effect that spreads throughout the plant's cells.

In 1998, Waterhouse and Wang provided one of first detailed descriptions of the machinery of RNA interference in plants.

Protein-RNA assemblages called RNA-induced silencing complexes (RISCs), mediate RNAi by cleaving double-stranded RNA molecules into fragments that vary from 21 to 24 nucleotides in length.

RISCs retain the fragments, which serve as templates for identifying complementary RNA sequences from viruses, or mRNAs from genes.

In animals, the RNA-cleaving endonucleases are encoded by one, or at most two, Dicer genes.

Waterhouse and Wang showed that dicotyletods like Arabidopsis have a basic complement of four specialised Dicer-like genes; poplar has five. Monocots typically have five; rice has six.

Waterhouse says the length of each fragment is a clue to which Dcl endonuclease produced it, and its role. Dcl1 yields 21nt microRNAs that regulate development. Dcl2 produces 22nt RNAs for antiviral defence.

Dcl3 makes 24nt RNAs that direct the methylation and histone deacetylase reactions that regulate gene activity by remodeling chromatin. And Dcl4, like Dcl1, makes 21nt RNAs that play the major role in virus defence. It is the Dcl gene that processes synthetic, hairpin RNAs to prime plant cells to resist viruses.

Waterhouse's team located Arabidopsis mutants for each gene, then crossed them to produce double mutants with all possible permutations of Dcl2, Dcl3 and Dcl4, including a triple-mutant knockout. Multiple developmental abnormalities left Dicer-like1 knockouts sickly and infertile.

Waterhouse and Carroll then performed grafting experiments with various stock-scion combinations to identify the molecule that spreads the gene-silencing effect from tissue to tissue, and to determine its mode of transmission.

They created rootstocks by inserting an RNAi 'hairpin' gene into each combination mutant line, targeting expression of green fluorescent protein (GFP), the standard marker of gene activity in plants and animals. They also inserted a GFP transgene into the rootstock plants, but the anti-GFP transgene blocked its expression, producing a normal phenotype.

The scions were taken from transgenic plants expressing GFP, whose tissues glow green under blue light - normal, green tissues lacking GFP appear red under blue light.

As rootstocks, they used the four combination mutants, including the triple mutant, after engineering each with a hairpin RNAi transgene designed to knock down expression of GFP in the scion.

Mystery messenger

Before experimenting with the transgenic stock-scion combinations, they performed a 'dry run' to determine how long it took the grafted plants to re-establish fluid flow through the graft.

By injecting a fluorescent green dye into the rootstocks of normal grafted plants, they established that it took five days for severed phloem tubes to reconnect, allowing the dye to flow into the graft.

Assuming the mystery RNAi 'messenger' was transported in the same manner, the earliest any GFP-silencing effect should be detected was five days post-grafting.

They left the scions on the rootstocks for varying lengths of time after phloem tube reconnection occurred at five days, then beheaded the grafted plants and implanted the scions on nutrient media.

They also made time-lapse videos of the action. "If you grow the plants without roots, you still see GFP silencing, showing that the signal continues to propagate through the tissues, even when it is disconnected from its source in the rootstock," Waterhouse says. "So the signal doesn't need to be constantly provided - it self-perpetuates after the initial pulse."

The experiment confirmed that the signal travels via the vasculature, not the plasmodesmata - microscopic channels in cell membranes, through which adjacent cells exchange ions and small molecules.

"This was reminiscent of work by Herve Vaucheret in 1997," Waterhouse says. "He showed that if you took the top part of an unsilenced tobacco plant, and put it on top of a silenced plant, the top part would be silenced.

"So it was predicted that the signal consisted of small RNAs moving through the vasculature, although it was not known at the time that there were four Dicer-like genes, producing different-sized small RNAs. The mutants allowed us to ask the right questions."

It had been suggested that the 24nt RNA from Dcl3 was the messenger. The Dcl3 mutant rootstock produces 21 and 22nt RNAs, but not 24nt RNAs - yet GFP silencing still occurred.

Rather than test the other double-mutants, the CSIRO researchers used the triple mutant, which expresses only Dcl1. To their enormous surprise, it still switched off GFP in the scion.

The experiment didn't rule out the possibility that Dcl1's 21nt molecule was the message-bearer, but that seemed unlikely, because the tissues of the triple-mutant rootstocks continue to fluoresce bright green, showing that no small RNAs - including Dcl1 - were being made against GFP

Yet the triple mutant rootstocks were still sending a GFP-silencing signal that switched off fluorescence in the scion. If the signal wasn't a small RNA, what could it be?

"Our working model is that it's not a small RNA, but some longer RNA made from the hairpin. Whether it's the entire hairpin is unclear.

"And that's the current state of play: we know a hairpin creates it, and it's almost certainly not a 'Diced' product."