A target product profile for a rapid diagnostic test to monitor mosquito gene drive presence and frequency

Use case 1: Detecting the presence of gene drive mosquitoes

This use case focuses on the timely detection of the gene drive construct in a mosquito population at a time when its presence has not yet been confirmed. The RDT would serve as a complement to core monitoring activities of field trials, and as a low-cost tool to enable easy participation of communities in trial and monitoring areas, as well as engagement of specific stakeholder groups, in the context of structured and active engagement activities. Locations where the gene drive has not yet been detected would likely include designated release and post-release monitoring zones, as well as locations beyond the mosquito’s predicted dispersal range. We anticipate the RDT will often be used in locations where gene drive mosquitoes are expected to be absent, and are not part of the core monitoring pipeline or its buffer zones. In this context, RDTs offer distinct advantages, foremost among them the capacity to enable active participation of minimally-trained (simple taxonomic skills) local communities and stakeholders in surveillance activities within defined monitoring areas. Such capacity is essential for confirming designated control areas as true negative controls, ensuring their validity for comparative analyses and for detecting faster-than-expected spread of the gene drive. RDTs provide clear utility in regions lacking access to centralized laboratory facilities or staff trained in molecular assays. Collections would be made through larval habitats, traps placed in houses overnight, or indoor aspiration performed in a house or animal shed. If samples are found to be positive for the gene drive, these should be confirmed through additional RDT tests and/or laboratory-based assays, and these locations may then be included in an existing surveillance pipeline, likely including sample analysis in a centralized laboratory. Last but not least, RDTs could be used in various developers’ stakeholder activities, such as laboratory visits, and various meeting formats including village community engagement. In summary, for use case 1, the RDT would: (i) facilitate detection in low-resource and isolated settings, (ii) enable participation of communities in trial and monitoring areas in the context of structured engagement activities, (iii) provide an initial test for inclusion of a zone as part of the core monitoring system, (iv) provide an approach for detecting the gene drive after the initial trial phase, and (v) be used by developers as a key stakeholder engagement tool.

Use case 2: Estimating gene drive carrier frequency

This is an extension of use case 1, and takes advantage of the simplicity and flexibility of the RDT to estimate gene drive carrier frequency by performing a limited number of RDTs on mosquito pools of various sizes. This series of RDTs can provide an estimate of gene drive carrier frequency, especially when sampled mosquitoes are representative of the population, and distributed optimally among pools. It is assumed that, when more accurate gene drive carrier or allele frequency estimates are needed, these will involve more extensive and representative sampling, and processing via molecular assays in the laboratory. The RDT-inferred carrier frequency estimates will therefore be provisional, and only seek to provide a rough estimate of gene drive carrier frequency in the population. In this context, for use case 2, the RDT would: (i) facilitate estimation of gene drive carrier frequency in newly-identified positive sites, and (ii) facilitate estimation of gene drive carrier frequency in known positive sites that may not be actively monitored, or would like to be audited by a third party. Collectively, these use cases place certain requirements and restrictions on the design and performance of the test.

Test sensitivity for use case 1

In this use case, the RDT would be used to detect the presence of the gene drive cassette in a new population. The false positive rate should be low, as detection of the gene drive will likely trigger costly follow-up collections and analyses. In the minimally essential case, we would like to detect the presence of the gene drive in a population with 95% confidence when the gene drive carrier frequency reaches 5%. Of note, we consider a 5% carrier frequency as a pragmatic design point at which detection is likely even for small sample sizes (e.g., > 20), and at which gene drive spread could still be considered in its early stages. For a gene drive carrier frequency of x and a test sensitivity per mosquito of s, the probability of detecting the gene drive in a sample of n mosquitoes, p_GD, is given by:

For a gene drive carrier frequency of 5% and a minimally essential probability of detection of 95%, Fig. 2A depicts the required sample size as a function of test sensitivity per individual mosquito. Required sample sizes are all in the vicinity of 100. Taking this as a default sample size that is achievable in the field in the rainy season, the minimally essential test sensitivity per individual mosquito is ≥ 60%. We then consider an ideal test sensitivity per individual mosquito of ≥ 90%, which leads to a probability of detection of 99% for a sample of 100 mosquitoes.

At lower gene drive carrier frequencies, larger sample sizes and/or higher test sensitivities are required to achieve comparable detection probabilities, and so the values presented here should be interpreted as illustrative rather than prescriptive. For instance, for a gene drive carrier frequency of 2% and a sample size of 200, a test sensitivity of ≥ 75% is required to achieve a detection probability of 95%. Exploration of alternative scenarios across a range of true carrier frequencies is supported through our open-source R-Shiny application, “TPP Explorer” (available at https://pverma.shinyapps.io/tpp_explorer).

Test specificity for use case 1

Now let us consider a fully wild-type mosquito population. For a false positive rate per mosquito of f (corresponding to a test specificity of one minus f), the probability of detecting a false positive in a sample of n mosquitoes tested one-by-one, p_FP, is given by:

Let us continue with a total sample size of 100, per the calculations for test sensitivity. Considering a minimally essential sample-wide false positive rate of 5% (Table 1), the minimally essential test specificity per mosquito is ≥ 99.95% (Fig. 2B). Increasing the sample size to 200 mosquitoes, a test specificity of ≥ 99.98% produces a sample-wide false positive rate of 5%. We consider this to be an ideal test specificity per mosquito. Note that, although the minimally essential test specificity of ≥ 99.95% for use case 1 is quite demanding, for a multiplexed diagnostic test, the composite false positive rate, f_MP, is equal to the multiplication of false positive rates for its M constituent tests, i.e.:

Here, f₁ through f_M refer to the false positive rates of constituent tests 1 through M. I.e., for two multiplexed tests of equal specificity, the minimally essential specificity for each test is ≥ 97.8%, and for three multiplexed tests, the minimally essential specificity for each test is ≥ 92.1%.

Pooled analysis of samples for use case 1

Testing for the presence of a gene drive in a new population will ideally be conducted with pooled samples for speed and cost-efficiency. In this case, the TPP parameters will vary depending upon the number of mosquitoes included in each pool. For a pool size of m, the minimally essential and ideal test sensitivity, s_m, and test specificity, (1—f_m), follow a similar derivation. We first consider the test specificity for a pool of m mosquitoes. Given a false positive rate per pool of m mosquitoes, f_m, the probability of detecting a false positive in a sample of n mosquitoes (where n is a multiple of m) is given by:

Let us consider a total sample size of 100 and a pool size of 10, as an example. Considering a minimally essential sample-wide false positive rate of < 5%, the minimally essential test specificity per pool of 10 mosquitoes is ≥ 99.5%. If we increase the total sample size to 200 mosquitoes, a test specificity per pool of 10 mosquitoes of ≥ 99.75% maintains a sample-wide false positive rate of < 5%. We consider this to be an ideal test specificity for a pool size of 10. Minimally essential and ideal test specificities for a range of pool sizes are depicted in Fig. 2D.

Next, we consider the test sensitivity for a pool of m mosquitoes. Again, we would like to detect the presence of the gene drive in the population with 95% confidence when the gene drive carrier frequency, x, reaches 5%. The chance that a gene drive carrier is present in a pool of m mosquitoes, x_m, is given by:

We define the test sensitivity for a pool of m mosquitoes, s_m, as the probability of detecting at least one gene drive mosquito in a pool of this size. The probability of detecting a gene drive mosquito in a sample of n mosquitoes (where n is a multiple of m) is then given by:

Let us consider the example of a total sample size of 100 and a pool size of 10. For a gene drive carrier frequency of 5%, the probability that a gene drive mosquito is present in a pool of 10 mosquitoes is 40.1%. Considering a minimally essential probability of detection of 95%, the minimally essential test sensitivity per pool of 10 mosquitoes is then ≥ 65%. We consider an ideal test sensitivity per pool of 10 mosquitoes to be ≥ 92%, which would lead to a probability of detection of 99% for a sample of 100 mosquitoes. Minimally essential and ideal test sensitivities for a range of pool sizes are depicted in Fig. 2C. Minimally essential and ideal parameter values for all scenarios explored for use case 1 are summarized in Table 1.

Test sensitivity and specificity for use case 2

In this use case, the RDT would be used to provide an estimate of gene drive carrier frequency in a population, which could be followed up with more extensive sampling and laboratory processing to provide a more accurate frequency estimate. False positives and false negatives will increase the variance in estimates of gene drive carrier frequency. They will also introduce bias into these estimates; but this can be corrected if the false positive and false negative rates are reasonably well-known in a field setting prior to use. For this use case, we consider two scenarios: (i) measuring gene drive carrier frequency as the drive spreads through a population (e.g., for a carrier frequency of ~ 50%), and ii) characterizing persistence of the drive at high frequency in a population (e.g., for a carrier frequency of ~ 90%). Here, we consider individual tests of mosquitoes; but the case of pooled samples is described in Additional file 1. We are primarily interested in the standard deviation (SD) of our estimate of gene drive carrier frequency, and use a binomial distribution with a known true positive rate, s, and a known false positive rate, f, to calculate this. For a measured proportion of samples positive for the gene drive, y, the estimator for gene drive carrier frequency, $\widehat{p}$, is given by:

Considering a sample size, n, the SD of the estimator, $\widehat{p}$, is given by:

Per the previous use case, imagine a sample of 100 mosquitoes, and consider a minimally essential SD of the estimator, $\widehat{p}$, of 10%, and an ideal SD of 7% (of note, the lowest possible SD for this case is 5%, which occurs when the test sensitivity and specificity are both 100%). As the gene drive spreads through the population (i.e., for a carrier frequency of ~ 50%), substituting the minimally essential test sensitivity (60%) and specificity (99.95%) for use case 1 into Eq. 8 yields an SD of $\widehat{p}$ of 7.6%, and substituting the ideal test sensitivity (90%) and specificity (99.98%) for use case 1 into this equation yields an SD of 5.5%. Since these values are less than 10% and 7%, respectively, this implies that the minimally essential and ideal conditions for use case 1 automatically satisfy the conditions for use case 2 as the gene drive spreads through the population. SDs of the estimator, $\widehat{p}$, for a range of test sensitivities and specificities for this scenario are depicted in Fig. 3A. E.g., in the case where sensitivity is equal to specificity, the minimally essential test sensitivity/specificity is ≥ 75% and the ideal test sensitivity/specificity is ≥ 86%, and in the case where test specificity is set to 99.95%, the ideal test sensitivity is ≥ 68%.

Secondly, we are interested in characterizing gene drive carrier frequency as the drive persists at high frequency in a population. In this case, we consider a carrier frequency of ~ 90% and reapply Eq. 8 for the SD of the estimator, $\widehat{p}$. Substituting the minimally essential test sensitivity and specificity for use case 1 into this equation yields an SD of $\widehat{p}$ of 8.3%, and substituting the ideal test sensitivity and specificity for use case 1 yields an SD of 4.4%. Since these values are again less than 10% and 7%, respectively, this implies that the minimally essential and ideal conditions for use case 1 automatically satisfy the conditions for both scenarios for use case 2. SDs of the estimator, $\widehat{p}$, for a range of test sensitivities and specificities for the scenario where the drive persists at high frequency are depicted in Fig. 3B. E.g., in the case where sensitivity is equal to specificity, the minimally essential test sensitivity/specificity is ≥ 73% and the ideal test sensitivity/specificity is ≥ 81%, and in the case where test specificity is set to 99.95%, the ideal test specificity is ≥ 69%.

Pooled analysis of samples for use case 2

Given the intended use of the RDT to obtain an approximate estimate of gene drive carrier frequency in a field setting where resources are limited, it is likely that pooled rather than individual mosquito tests will commonly be used. Pooling of mosquitoes will increase the SD of the estimator, $\widehat{p}$; but this may be acceptable when the goal is to obtain an estimate of carrier frequency that can be followed up with more extensive and rigorous molecular assays in the laboratory, when needed. For a given sample size and available number of test kits, the optimal distribution of pool sizes depends on gene drive carrier frequency, and the sensitivity and specificity of the RDT for each pool size. A maximum likelihood-based approach to designing optimal pooling strategies for given parameter values and subject to given sample size and test kit constraints is described in Additional file 1. Here, to evaluate each pooling design, mosquito sampling and pooled testing is simulated n_rep times, and carrier frequency is estimated using a maximum likelihood framework applied to the replicate-level pooled test outcomes. The SD of the estimator, $\widehat{p}$, is then derived from the inverse of the observed Fisher information, and pooling designs are ranked in terms of their SD.

We developed an R-Shiny application, “TPP Explorer”, to explore optimal pooling designs to minimize the SD of the estimator, $\widehat{p}$, for a given sample size and available number of RDTs. The optimal distribution of pool sizes is highly dependent on gene drive carrier frequency—at high frequencies, larger pools become saturated with positive tests and hence individual tests or smaller pools are more informative; while at low frequencies, larger pools are informative when paired with pools of different sizes. Consider, for instance, a sample size of 100, 20 test kits, a test sensitivity (independent of pool size) of 90% (the ideal value for use case 1), a test specificity (independent of pool size) of 99.95% (the minimally essential value for use case 1), and possible pool sizes of one, two, five and 10. For a carrier frequency of 10%, the optimal pooling distribution is a single test of 10 pooled mosquitoes, 17 tests of five pooled mosquitoes, and two tests of two pooled mosquitoes, which results in an SD of 3.7%. For a carrier frequency of 90%, the optimal pooling distribution is 20 tests of individual mosquitoes, which only makes use of a fifth of the sampled mosquitoes, but results in an SD of 9.7%.

A trade-off exists between the desired precision of the carrier frequency estimate and the number of test kits used; but 20 test kits are generally sufficient to obtain an SD in the vicinity of 10% for the above-listed parameters (a sample size of 100, sensitivity of 90%, and specificity of 99.95%). E.g., for a carrier frequency of 50%, the optimal pooling with 20 test kits results in an SD of 11.6%, which, if desired, can be reduced to 10.0% by increasing the number of test kits to 27.

These results assume that mosquitoes are sampled randomly from a well-mixed population, and that individuals within each pool represent independent samples from the underlying population. In practice, mosquito populations are spatially structured, and collections are often concentrated at a small number of trapping locations. Under such conditions, detection probability and frequency estimates depend on how sampling locations relate to the underlying population distribution. In spatially heterogeneous populations, achieving a desired detection probability may require larger or repeated samples, or sampling distributed across several locations rather than relying on pools constructed from a small number of collection sites. Where collections are clustered, constructing pools from mosquitoes obtained across multiple traps or households within a site will provide a more representative estimate of population frequency. These considerations do not alter the target sensitivity and specificity values outlined in the TPP; but they do affect how these performance characteristics are achieved in practice. Pooling designs also have direct implications for the number of tests required, and hence overall monitoring costs, which we consider in the next section.

Cost considerations and monitoring effort

While precise costing is beyond the scope of this study, it is useful to consider the scale of monitoring effort implied by the proposed use cases of RDT. As an illustrative scenario, consider monitoring across 10 sites with sampling of 100 mosquitoes per site, which is consistent with achievable field collections. These estimates represent a simplified scenario of RDT requirements only, rather than the full monitoring effort, which may include additional sampling and confirmatory laboratory testing, with RDT-based testing complementing rather than replacing centralized laboratory analyses. For use case 1, which focuses on detection at low or unknown frequency, pooled testing with pools of 10 mosquitoes implies approximately 10 tests per site per sampling round, corresponding to on the order of 100 tests across all sites. For use case 2, applied during targeted follow-up sampling campaigns after detection, approximately 20 pooled tests per site are sufficient to achieve a standard deviation of about 10% for a sample size of 100 mosquitoes. This corresponds to on the order of 100–200 tests per sampling round across all 10 sites.

Overall, these estimates suggest that even a relatively small trial may require on the order of 10³–10⁴ RDTs per year, assuming monthly sampling for use case 1 and more frequent sampling for use case 2. These requirements would scale with sampling intensity and the number of monitored sites. Experience from Wolbachia deployment programs suggests that diagnostic and monitoring requirements can become substantial in area-wide vector interventions, with monitoring intensity depending on trial endpoints, spatial scale, and duration of follow-up [42, 43]. At the ideal cost target of USD 1–2 per test specified in Table 2, these testing requirements corresponds to several thousand to tens of thousands of dollars annually, excluding personnel, transport, and confirmatory laboratory analyses. These calculations highlight that even modest per-test costs can translate into substantial programmatic expenses when scaled across space and time, making pooling and low-cost RDTs essential for maintaining feasible monitoring budgets.

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.