OX3604C males decreased, but did not eliminate target populations in the field-cage experiment. Experimental protocols were similar between the previous laboratory experiment 
and our field cage experiment, including high 10
1 release ratios that in both cases were expected, based on modeling, to favor population extinction. Over the course of the field cage study, the OX3604C
target male ratio increased in all cages and reached the highest value of 1,000
1 in cage 1 during week 17 post-release, but this did not result in extinction of the target population. These results also were consistent with the data on the frequency of the DsRed2 marker detected in larvae from treatment cages, which ranged from 20–54% during week 17 post-release when the trial was terminated. The high estimated OX3604C male fitness cost calculated from the field cage trial (97%), the lower than predicted population reduction, and the long estimated extinction times (average 23–65 weeks) lead to the expectation that male OX3604C may be less effective for population reduction under open field conditions than predicted from results of the laboratory cage experiment.
Output from a simulation model predicted that the lack of a homozygous OX3604C strain did not contribute significantly to the absence of population extinction, because of the high OX3604C
target male release ratio. Presence of wild-type individuals in the transgenic population required that we manually sex pupae in order to avoid introducing wild-type females into treatment cages. This resulted in a difference in management of the transgenic strain relative to the target strain and relative to the handling of the transgenic strain by Wise de Valdez et al
. The mortality observed among the transgenic males before introduction into treatment cages was low (~5%), suggesting that this additional handling did not cause substantial harm. Furthermore, transgenic male survival in the cages was high. As can be seen in , there were on average about 5 times as many males in treatment cages as control cages on day six after each release of OX3604C. This is equivalent to a daily survival rate of ~0.91, which is similar to published values for Ae. aegypti
survival in houses in the field 
. The potential effect of differential handling was not addressed directly by the mating competitiveness experiments because the need to separate males from females added an additional handling step for GDLS2 males and females, such that the treatment of the two types was matched more closely in these experiments than in the earlier field cage experiment. Because differences were not apparent in the field cage trial, male survival was not evaluated in mating competitiveness experiments.
Mosquito size is sometimes but not always 
associated with fitness. For Ae. aegypti
, Ponlawat and Harrington 
reported greater mating success by larger than smaller males. Measurements made in the first two weeks of the field cage experiment indicated that OX3604C males (median wing length
2.24) were slightly larger than GDLS2 males (median wing length
2.17, Mann-Whitney U test p
<0.01), so this is unlikely to have contributed to the field cage outcomes.
from mating competitiveness experiments with laboratory-reared mosquitoes in field cages have generally been found to underestimate fitness costs found when the same types of mosquitoes are released in the field 
. While the mating disadvantage of OX3604C males observed in experiment 6 (59.1%) appears to be one factor explaining the lack of extinction in our field cage trial, if there had only been a fitness cost of 59%, some extinctions would have been expected (Figures S4, S5). Therefore, although short-term mating competitiveness experiments are useful in assessing one major component of fitness, they are not designed to measure as many aspects of fitness as are measured in long-term studies.
Adaptation of genetically-engineered mosquitoes and target populations to laboratory and field cages always needs to be taken into account when moving from the laboratory to the field. In the field cage trial, OX3604C was derived from introgression of the OX3604C construct into a GDLS1 genetic background, reared in laboratory conditions (i.e., stable temperature, relative humidity, and photoperiod) and mated in small, crowded laboratory cages for more than 20 generations, all of which potentially selected for capacity to mate in a small spaces, and other adaptations for increased fitness in a laboratory environment. Conversely, GDLS2 originated from mosquitoes collected from the same locations 2 years after those used to create GDLS1, and GDLS2 target populations were maintained in large outdoor field cages for 16 generations before the start of the experiment. During the prerelease period, GDLS2 populations experienced natural variation in daily temperature and relative humidity and mated successfully for ~4 months in their large outdoor enclosures. Adaptation to field cages may have been an advantage for GDLS2 males when competing with OX3604C males for GDLS2 females.
Our results are consistent with those from previous mosquito studies 
indicating that colony maintenance and mass rearing should be planned prior to field-cage or open-field trials. Rearing large numbers of transgenic mosquitoes in large outdoor, semi-field enclosures for several generations may help avoid undesirable laboratory adaptation and reduce fitness differences between transgenic mosquitoes and conspecifics in their natural, target populations. Short-term mating competition experiments in large field cages could be an efficient way to gather preliminary information on genetically-engineered mosquito fitness relative to local wild-type mosquitoes, but they only measure one important fitness component while field cage trials include additional components.
We emphasize the potential impact of differential strain adaptation to the field or laboratory, but it is also possible that the fitness difference was due to the transgenesis process. Although insertion of the transgene did not affect the ability of the OX3604C to cause extinction in the laboratory system, it is feasible that some negative pleiotropic effect of the gene insertion was manifested only under field cage conditions. Precautions were taken to avoid some negative effects that are often associated with transgenesis. Most importantly, the originally engineered strain was backcrossed for five generations to a strain for the local area where the experiment was conducted. This was expected to replace over 96% of the genes from the engineered strain with local strain genes, except for genes linked to the transgene. If the transgene had been inserted within a transcribed gene, it could have disrupted gene function that affected fitness in outdoor field cages, but not in the laboratory. Attempts to fine-scale map the location of the transgene indicated that the insertion was in a genomic area with repetitive DNA, indicating the transgene was not inserted within a transcribed gene.
Although an argument can be made for not pursuing an open field evaluation of OX3604C males based on our field cage results, the best way to resolve the discrepancy between laboratory and field cage results would be to assess them under uncontained, open field conditions. Because this has not been done, data do not exist to determine whether laboratory or field cage experiments are most informative about how this strain will perform under natural conditions. A different genetic background, different chromosomal location of the transgene or different rearing procedures could separately or in combination affect the competitiveness of transgenic mosquitoes. Evaluation of other Ae. aegypti strains carrying the female-flightless transgene would help determine if results observed in this trial apply to this genetic modification in general or are specific to the OX3604C strain we studied.
Our results support inclusion of large outdoor field cage experiments in the systematic, phased evaluation of GE Ae. aegypti
, including those with transgenes like OX3604C that are self-limiting. Details of field cage construction and the level of containment needed will depend on the nature of genetic modification in the strain being evaluated as well as general requirements of the relevant regulatory authorities. If genetic modifications include the potential of elevated pathogen transmission or non-Mendelian inheritance (i.e.
, genetic drive systems), strain evaluation will require higher security caging than those used in our experiments 
. We stress that short-term mating competition experiments in large field cages could be used to obtain predictive information on mating competitiveness and fitness costs, but it is not clear that by themselves these would be sufficient substitutes for longer-term field cage tests. Results
of appropriately planned, executed and analyzed open-field releases of the OX3604C would be useful in addressing this issue.
All of the work described here was conducted within ethical, social and cultural guidelines for community engagement activities 
. We found that this approach helped us to develop respect and trust, basic ingredients for strong working relationships with local residents living near the field site, and for appropriate dialogue with state and national health and environmental authorities, scientists, and local and international press. Although the containment measures and communication activities taken in this work were greater than expected for research with natural strains of mosquitoes, we feel that this precautionary approach could have long-term benefits by decreasing suspicion that transgenic mosquito technology is being applied carelessly