Specific sequences in the 3′ untranslated region (UTR) of CaMXMT1 messenger RNA were selected for construction of RNAi short and long fragments (Fig. 1). We transformed Agrobacterium tumefaciens EHA101 cells with these constructs and then used them to transform Coffea canephora5. After 2–4 months of culture, most infected tissues turned brown and necrotic; however, it was possible to regenerate hygromycin-resistant cells from these tissues. Seedlings were then cultured as described5.

Figure 1: Properties of decaffeinated transgenic coffee leaves.
figure 1

a, One-year-old somatic seedlings of Coffea canephora from wild-type (left) and RNAi-transgenic (right) plants. b, c, Endogenous theobromine and caffeine, respectively, in mg per g of fresh plant tissue, in different somatic seedlings of C. canephora, as detected by high-performance liquid chromatography2. Mean values were calculated from six independent measurements per line. Short RNAi fragments (RNAi-S) were constructed using 139-base-pair (bp; corresponding to nucleotide positions 1,139–1,277) and 161-bp (positions 1,117–1,277) sequences of CaMXMT1 (GenBank accession number AB048794), with an intervening 517-bp β-glucuronidase (GUS) fragment as spacer; long RNAi fragments (RNAi-L) contained two identical sequences of 332 bp (positions 946–1,277) separated by a 517-bp GUS fragment. The resulting constructs were inserted into a pBIH1-IG vector7; the control construct contained a green fluorescent protein gene (GFP). Somatic embryos of C. canephora were grown on modified half-strength Murashige and Skoog medium containing 20 µM 2-isopentenyladenine.

More than 35 transgenic somatic seedlings were obtained from each transformant, each containing short or long RNAi fragments or a control gene encoding green fluorescent protein (GFP). The phenotypes were apparently normal when compared with the wild-type plant (Fig. 1a).

Young leaves of one-year-old seedlings were collected 2–3 weeks after flushing and their purine alkaloid content was measured. The wild-type and transgenic lines that expressed GFP contained similar amounts of endogenous theobromine and caffeine (about 1 and 8 mg per g of fresh plant tissue, respectively; Fig. 1b, c). By contrast, young leaves of transgenic lines expressing RNAi showed a 30–80% reduction in theobromine content (Fig. 1b) and a 50–70% reduction in caffeine content (Fig. 1c) in comparison with the controls.

At present, coffee is decaffeinated industrially, but the process is expensive and the flavour of the product is poor6 — problems that could potentially be overcome by the genetic engineering of coffee plants3,6. As CaMXMT1 is expressed in young leaves, buds and immature fruits4, the transgenic plants described here should yield coffee beans that are essentially normal apart from their low caffeine content at maturity.

We are now applying this RNAi-based technique to C. arabica, which produces high-quality Arabica coffee and accounts for roughly 70% of the world market. Our method not only shortens the breeding period, which is more than 25 years for conventional crossing, but also opens the way to develop new species of coffee plant.