Taken as a whole, the experiments reported here suggest that homologous recombination may be an effective mechanism in mosquito cells and encourage the further exploration of gene targeting strategies for the generation of transgenic mosquitoes. The use of different substrate topologies allowed us to investigate the process (or processes) through which homologous recombination occurs and revealed marked differences in the efficiency with which luciferase activity was restored. The data indicate that linear targeting vectors are better substrates than circular ones and that double-strand breaks can stimulate homologous recombination frequencies if they occur near, and preferably within, the region of homology. In this respect, the observed results are consistent with similar findings in mammalian cells, where double-strand breaks are thought to be effective because the broken strands can act as recipients in non-reciprocal exchanges and where their effects appear to be cumulative such that breaks in both substrates are much more efficient than breaks in only one [12, 14, 8, 9].
The way in which extrachromosomal substrates recombine to restore luciferase activity in mosquito cells can tell us much about the cellular processes involved. Circular molecules are generally held to represent poor substrates for homologous recombination and this is clear from the results presented here. When one or both substrates are linearized, luciferase activity can be restored by one of three mutually exclusive pathways for double-strand break repair [16, 17]. These are recombinational repair, single-strand annealing and non-homologous end-joining. The requirements for each pathway and, indeed, the choice of pathway, are beginning to be unravelled in mammalian and yeast systems. For example, in mouse ES cells the Ku70 and Ku80 proteins are believed to shift repair dynamics in favour of non-homologous end-joining whereas in yeast, the Rad52 and Rad51 proteins appear to shift towards recombinational repair . Little is known of the presence or action of potential homologues in insects but, from the experiments described here, it would appear that more than one recombination pathway might be involved and there is evidence that particular pathways are favoured by specific substrate topologies. There is also evidence that the choice of repair pathway may be influenced by the stage of the cell cycle at the time of repair, with a bias towards non-homologous end-joining during G1 – early S phase and towards recombinational repair during late S phase – G2.
During recombinational repair , initiation of homologous recombination involves a double-strand break that is enlarged to a gap by exonucleolytic degradation of both strands. The gap is then repaired by copying corresponding sequences from the homologous partner, with both sides of the gap invading the donor duplex leading to an intermediate structure with two Holliday junctions and heteroduplex DNA flanking the gap. Resolution of this structure gives equal numbers of crossover and non-crossover products and, since there is no loss of sequences, the process is described as conservative. Recent evidence has challenged the relative importance of this pathway, in particular the observation that both spontaneous and induced double-strand breaks are typically processed to long 3' single-stranded ends, rather than gaps . Our data suggest that recombinational repair is not the predominant pathway in mosquito cells since, according to this model, co-transfections involving one linearized and one circular molecule (DLC + DRLS and DLLB + DRC respectively) should have restored comparable luciferase activities.
The observation that restored luciferase activity depends on the choice of substrate used as chromosomal mimic could be interpreted by a modification of recombinational repair known as one-sided invasion . In this model, only one side of a double-strand break in the recipient invades the unbroken donor and primes DNA synthesis in the homologous region to generate a functional recombinant. During this process synthesis can extend beyond the region of homology and into the flanking DNA. In our experiments, linearization of DR with SacI releases a free 3' end immediately downstream of the homologous region. One-sided invasion of DL, with synthesis extending towards the end of the luciferase gene, would recover an intact coding sequence and restore activity. In the opposite configuration, linearization of DL with BamHI would release a free 3' end immediately upstream of the region of homology. One-sided invasion of DR with synthesis extending towards the start of the luciferase gene could also recover a functional coding sequence, although apparently at lower efficiency. This model could therefore provide an adequate explanation of the results obtained, provided only that repair efficiency is greater in one direction. In this situation, it is possible that the strong constitutive actin 5c promoter biases the repair when initiated away from the promoter (DLC + DRLS).
Single-strand annealing  has been described as the most efficient pathway for extrachromosomal homologous recombination in mammalian cells  and plant cells [25–27]. It is also known to be involved in double-strand break repair in yeast, where it is dependent on direct repeats at either side of the break  but the requirements in higher eukaryotes are not known. Recent data suggest that single-strand annealing is the predominant pathway for double-strand break repair in mouse oocytes but that this declines by the embryonic stage . The essential feature of the model is that DNA ends at a double-strand break are rendered single-stranded by a 5'-3' exonuclease or as a result of unwinding . When homologous sequences are present, this process ends with complementary single-strands that are capable of re-annealing. Repair synthesis and ligation then complete the formation of the non-reciprocal homologous junction. Our data show that single-strand annealing can also be an effective pathway for double-strand break repair in mosquito cells, given appropriate substrate topology. The model is favoured by linearization of both substrates within the region of homology and this configuration (DLLB + DRLS) does provide the greatest restoration of luciferase activity. Predictably, linearization outside of the region of homology is less effective since it does not provide homologous single-strands for the annealing reaction.
Non-homologous end-joining is a pathway for the repair of double-strand breaks in which the broken DNA ends are simply re-ligated, without the need for a template molecule or region of homology. This phenomenon has been highlighted recently  where it was described as the predominant mechanism in zygotes and early embryos of the zebrafish, Danio rerio as well as in D. melanogaster. It is also thought that this may be a common pathway in mammalian cells, although it is error prone and may introduce small deletions at the joining site . In the context of the experiments described here, end-joining would not restore a functional luciferase gene but the resulting duplication of sequence would be detectable as higher molecular weight fragments on Southern blotting.