\makeatletter
\@ifundefined{HCode}
{\documentclass[CRBIOL,Unicode,screen,biblatex,published]{cedram}
\addbibresource{crbiol20250221.bib}
\newenvironment{noXML}{}{}
\let\citep\parencite
\let\citet\textcite
\def\xcitealp#1#2{\citeauthor{#1}, \citelink{#1}{#2}}
\newcommand*{\citelink}[2]{\hyperlink{cite.\therefsection @#1}{#2}}
\def\defcitealias#1#2{}
\let\citepalias\parencite
\let\citetalias\textcite
\newenvironment{Table}{\begin{table}}{\end{table}}
\def\thead{\noalign{\relax}\hline}
\def\endthead{\noalign{\relax}\hline}
\def\tabnote#1{\vskip4pt\parbox{.83\linewidth}{#1}}
\def\tsup#1{$^{{#1}}$}
\def\tsub#1{$_{{#1}}$}
\RequirePackage{etoolbox}
\def\jobid{crbiol20250221}
%\graphicspath{{/tmp/\jobid_figs/web/}}
\graphicspath{{./figures/}}
\newcounter{runlevel}
\skip\footins30pt
\let\MakeYrStrItalic\relax
\def\refinput#1{}
\def\back#1{}
\def\tbody{\noalign{\relax}}
\newenvironment{inftab}{}{}
\def\hyphen{\text{-}}
\def\xmorerows#1#2{{#2}}
\def\0{\phantom{0}}
\def\xsection#1{}
\def\botline{\\\hline}
\DOI{10.5802/crbiol.182}
\datereceived{2025-03-11}
\daterevised{2025-06-16}
\dateaccepted{2025-07-01}
\ItHasTeXPublished
\def\og{\guillemotleft}
\def\fg{\guillemotright}
\makeatletter
\g@addto@macro{\UrlBreaks}{\UrlOrds}
\gappto{\UrlBreaks}{\UrlOrds}
\usepackage{hyperref}
\makeatother
}
{
\PassOptionsToPackage{authoryear}{natbib}
\documentclass[crbiol]{article}
\def\CDRdoi{10.5802/crbiol.182}
\let\newline\break
\def\selectlanguage#1{}
\usepackage[T1]{fontenc}
\def\xmorerows#1#2{\morerows{#1}{#2}}
 \def\citelink#1#2{\citeyear{#1}}
 \def\xcitealp#1#2{\citealp{#1}}
\def\href#1#2{\url[#1]{#2}}
\def\xsection#1{}
\makeatletter
\def\CDRsupplementaryTwotypes#1#2{}
}
\makeatother

\usepackage{upgreek}

\dateposted{2025-09-08}
\begin{document}

\begin{noXML}

\CDRsetmeta{articletype}{review}

\title{Risks associated with CRISPR homing gene drive}

\alttitle{Risques associ\'{e}s au for\c{c}age g\'{e}n\'{e}tique}

\author{\firstname{Virginie} \lastname{Courtier-Orgogozo}\CDRorcid{0000-0002-9297-9230}}
\address{Universit\'{e} Paris Cit\'{e}, CNRS, Institut Jacques Monod,
75205 Paris cedex 13, France}
\curraddr{Institut Jacques Monod, 15 rue H\'{e}l\`{e}ne Brion, 75205
Paris cedex 13, France}
\email{virginie.courtier@normalesup.org}

\keywords{\kwd{Gene drive}
\kwd{CRISPR}
\kwd{Risk}
\kwd{Biosecurity}
\kwd{Biosafety}
\kwd{Biotechnology}
\kwd{Bioweapon}}

\altkeywords{\kwd{For\c{c}age g\'{e}n\'{e}tique}
\kwd{CRISPR}
\kwd{Risque}
\kwd{Bios\'{e}curit\'{e}}
\kwd{Bios\^{u}ret\'{e}}
\kwd{Biotechnologie}
\kwd{Arme biologique}}

\begin{abstract} 
CRISPR homing gene drive is a disruptive biotechnology developed over
the past decade with potential applications in public health,
agriculture, and conservation biology. This technology relies on an
autonomous selfish genetic element able to spread in natural
populations through the release of gene drive individuals. However, it
has not yet been deployed in the wild. In this review, we examine the
key risks associated with CRISPR homing gene drives. First, we explore
technical limitations, where gene drives might not be as efficient as
intended, and cases where mitigation strategies may not be able to
block a gene drive. Then, we present four main categories of adverse
effects: (a) ecological risks, corresponding to unintended consequences
on ecosystems and non-target populations; (b) sociological risks, i.e. 
concerns over public perception, governance, and societal acceptance;
(c) risks associated with research activities; and (d) risks associated
with malevolent usage. Regulatory aspects are not addressed here. This
article provides a foundation for evaluating gene drive risks to ensure
responsible and informed decision-making.
\vspace*{2pt}
\end{abstract}

\begin{altabstract} 
Le for\c{c}age g\'{e}n\'{e}tique est une biotechnologie novatrice qui a
\'{e}t\'{e} d\'{e}velopp\'{e}e au cours de la derni\`{e}re d\'{e}cennie
et qui pr\'{e}sente des applications potentielles dans les domaines de
la sant\'{e} publique, de l'agriculture et de la biologie de la
conservation. Cette technologie repose sur un \'{e}l\'{e}ment
g\'{e}n\'{e}tique \'{e}go\"{i}ste autonome capable de se propager dans
les populations naturelles, suite \`{a} la lib\'{e}ration dans
l'environnement d'individus porteurs. Cependant, elle n'a pas
encore \'{e}t\'{e} d\'{e}ploy\'{e}e dans la nature. Dans cette
\'{e}tude, nous examinons les principaux risques associ\'{e}s au
for\c{c}age g\'{e}n\'{e}tique. Tout d'abord, nous explorons les
limites techniques susceptibles d'emp\^{e}cher le for\c{c}age
g\'{e}n\'{e}tique d'\^{e}tre aussi efficace que pr\'{e}vu, ainsi que
les cas o\`{u} les strat\'{e}gies d'att\'{e}nuation pourraient ne pas
\^{e}tre en mesure de bloquer un for\c{c}age g\'{e}n\'{e}tique.
Ensuite, nous pr\'{e}sentons quatre grandes cat\'{e}gories d'effets
ind\'{e}sirables : (a) les risques \'{e}cologiques, correspondant \`{a}
des cons\'{e}quences impr\'{e}vues sur les \'{e}cosyst\`{e}mes et les
populations non cibl\'{e}es ; (b) les risques sociologiques,
c'est-\`{a}-dire les pr\'{e}occupations li\'{e}es \`{a} la perception
du public, \`{a} la gouvernance et \`{a} l'acceptation par la
soci\'{e}t\'{e} ; (c) les risques associ\'{e}s aux activit\'{e}s de
recherche ; et (d) les risques associ\'{e}s \`{a} une utilisation
malveillante. Les aspects r\'{e}glementaires ne sont pas abord\'{e}s
ici. Cet article fournit une base g\'{e}n\'{e}rale pour \'{e}valuer les
risques associ\'{e}s au for\c{c}age g\'{e}n\'{e}tique et pour garantir
une prise de d\'{e}cision responsable et \'{e}clair\'{e}e.
\end{altabstract}

\thanks{European Research Council under the European Community's Seventh
Framework Program (FP7/2007-2013 Grant Agreement no. 337579), Agence
Nationale de la Recherche (ANR) under the project
``ANR-24-CE13-0018-01''}

\maketitle

\vspace*{1.2pc}

\twocolumngrid

\end{noXML}

\defcitealias{Champeretal2020b}{ibid.}
\defcitealias{Champeretal2018}{ibid.}
\defcitealias{Courtier-Orgogozoetal2020}{ibid.}

\section{Introduction}\label{sec1}

CRISPR homing gene drive belongs to a special category of technologies
that can potentially have widespread impacts on natural populations and
the environment. This makes it particularly difficult to fully
anticipate and manage the associated risks, as for other high-impact
technologies like nuclear physics, geo-engineering or the manipulation
of viruses with pandemic potential. Despite this conundrum, the present
context is highly conducive to the development of such high-impact
technologies, for at least three reasons. First, as research advances,
technologies are becoming more and more efficient and accessible, and
thus likely to have larger effects. Second, competitiveness between
teams and between countries, as well as the promotion of ``High Risk,
High Gain'' research, via funding incentives and \mbox{possibly} facilitated
access to top science journals such as \textit{Nature} and
\textit{Science}, encourage the development of such research. Third,
the climate change and the biodiversity crisis can lead some people to
place excessive hopes in such technologies to address the current
challenges and thus encourage further development of these approaches.
For the particular case of gene drive, the major arguments put forward
to promote the development of this technology relate to the fight
against mosquito-borne infectious diseases: failure of current insect
control methods to curb disease transmission and to stop the spread of
arboviruses (especially dengue, with record numbers of infections in
recent years), due in particular to the spread of insecticide
resistance, the expanding range of vector mosquitoes, increasing
pathogen resistance to medication and the absence of, or low
accessibility to, efficient vaccines \citep{Wengetal2024}.

Undertaking a benefit-risk assessment of CRISPR homing gene drive is
inherently difficult. It is impossible to fully anticipate all
possible impacts. Furthermore, there is no standard method to weigh up
the various pros and cons in a way that would satisfy everyone,
according to their various perspectives. Nevertheless, in order to
decide whether CRISPR homing gene drive should be applied to natural
populations, a reasoned decision must be based on a benefit-risk
assessment; there is no other path. According to the ``responsible
research and innovation'' (RRI) policy framework promoted by the
European Union \citep{Wittrocketal2021}, the mission of scientific
research is not limited to producing knowledge; it must also ensure
that the research and resulting applications are in line with society's
needs, interests and values. In this spirit, characterizing the various
risks and benefits associated with CRISPR homing gene drive is part of
the scientific enterprise. Identifying the risks can help to design
safer technologies and to inform public debate and decision-makers.
Risk considerations have been omnipresent for CRISPR homing gene
drives. As soon as the first papers describing this new technology were
published in 2015 \citep{Gantzetal2015,GantzBier2015}, risks were
highlighted by the researchers themselves who developed the technology,
as well as others \citep{Akbarietal2015,Esveltetal2014}. Since then,
the ethical questions raised by CRISPR homing gene drives have not
faded away. A quick PubMed search reveals that currently in the
scientific literature more papers are being published about the risks
associated with CRISPR homing gene drive than about new technological
developments.

In their report issued in 2016, the National Academies of Sciences,
Engineering, and Medicine of the United States of America concluded
that ``it is essential to examine each gene drive on a case-by-case
basis'' \citep{NationalAcademiesofSciencesEngineeringandMedicine2016}.
We agree with this statement but would like to add that, given the
issues at stake, it can also be useful to combine such case-by-case
assessments with a general overview of the risks and benefits
\citep{Moroetal2018}. In particular, the slippery slope argument
\citep{VanDerBurg1992}, that authorizing a specific application for
gene drive may then pave the way for others to be accepted, can be
framed more precisely when one has a clear view of the risks and
benefits that extend beyond the initial application in question.

Ultimately, the risks and benefits associated with gene drive
technologies should be compared to the ones associated with other pest
control methods (insecticides, wetland removals to control
mosquitoes\,\ldots), but this goes beyond the author's expertise and
would require a multidisciplinary study. The present article focuses on
the general risks associated with CRISPR homing gene drive technology.
The benefits are reviewed elsewhere
\citep{Champeretal2016,Esveltetal2014}. This article does not cover
existing regulations of CRISPR homing gene drive
\citep{GeneticLiteracyProject2025} nor ongoing discussions in various
countries on how to regulate this biotechnology. After a brief
description of the technology, we examine the limitations and risks
associated with CRISPR homing gene drive.

\section{CRISPR homing gene drive technology}\label{sec2}

\subsection{The term ``gene drive''}\label{sec21}

Gene drive is a broad term that can designate both natural processes
and biotechnologies that can bias genetic inheritance
\citep{Alpheyetal2020}. We note, however, that some scientists prefer
to restrict this terminology to human-made constructs, as in its
initial definition
\citep{AgrenClark2018,Esveltetal2014,WellsSteinbrecher2022}. Several
international bodies, such as the Convention on Biological Diversity
(CBD) or the International Union for Conservation of Nature and Natural
Resources (IUCN), clearly make the distinction with the use of the term
``Engineered Genes Drives''
\citep{CPCBD2024,IUCN2024}.

Interestingly, the French term for gene drive, ``for\c{c}age
g\'{e}n\'{e}tique'' (literally meaning ``genetic forcing'')---coined by
Eric Marois, a researcher who works on mosquito gene drives
\citep{Herzberg2016}---, 
only refers to the biotechnology, and not to natural phenomena. It also
conveys the idea of compelling and coercing. Meanwhile, the
communication services of the non-governmental organization ``Target
Malaria'' is currently promoting the use of an alternative, more
benign, French term, ``impulsion g\'{e}n\'{e}tique''
(\url{https://targetmalaria.org/fr/notre-mission/comment-cela-fonctionne/}).

Gene drive increases the probability that a particular genetic element
will be transmitted to the offspring, thus allowing the propagation of
a suite of genes and mutations throughout a population 
\citep{Esveltetal2014}. The advantage in transmission can also allow
genetic elements with fitness disadvantages (``selfish-DNA'') to spread
through the population. Several types of CRISPR homing gene drives have
been designed over the years
\citep{Rabanetal2023,Wangetal2024}. 
Here we focus on the technique that has received the most attention and
that is currently the most developed, CRISPR homing gene drive, named
here for short ``gene drive''.

\subsection{Molecular mechanisms}\label{sec22}

Gene drive technology relies on a piece of DNA, named ``gene drive
element'', that is inserted within a chromosome and that has the
ability to copy itself at the same location on the other paired
chromosome \citep{Esveltetal2014}. This DNA piece contains several
components that give rise to an active CRISPR-Cas9 system: a gene
encoding the Cas9 protein, a gene encoding a guide RNA that targets a
site located on the paired chromosome where the gene drive element will
insert, other genes if necessary (such as a gene conferring resistance
to a pathogen) and flanking regions that are identical to the ones
adjacent to the guide RNA target site. All these components allow the
formation of Cas9-guide RNA molecular complexes that can cut DNA at the
target site. Then, homology-directed repair can lead to a copy of the
gene drive element on the other paired chromosome. In most cases, the
\textit{Cas9} gene is expressed in the germline, so that germline cells
end up with a copy of the gene drive element on both of their paired
chromosomes, and thus produce gametes that all harbor the gene drive
element (Figure~\ref{fig1}). As a consequence, all the progeny of a
gene drive individual are expected to carry the gene drive element, as
opposed to standard Mendelian genetics where a given allele present in
one parent is received on average by 50\% of the progeny. Theoretical
modeling indicates that if gene drive individuals are released at a
small frequency in a wild population, about 10--15 generations are
sufficient to obtain a population contaminated at 100\%, which
corresponds to a couple of years for many insect species
\citep{Deredecetal2008}. \looseness=1

\begin{figure}
\vspace*{3pt}
\includegraphics{fig01}
\vspace*{3pt}
\caption{\label{fig1}CRISPR homing gene drive technology relies on a
piece of DNA, named ``gene drive element'', that has the ability to
copy itself onto the other paired chromosome. In diploid individuals 
(left), the two paired chromosomes are inherited by
both daughter cells after cell division, and segregate individually in
gametes after meiosis, so that germline cells produce 50\% of gametes
with one chromosome and 50\% of gametes with the other paired
chromosome. With CRISPR homing gene drive, the gene drive element (dark
blue rectangle) copies itself onto the other paired chromosome. The
germline cell thus generates 100\% of gametes that carry the gene drive
element.} 
\end{figure}

Gene drive can spread two types of mutations. On the one hand, it can
spread genes of interest that have been introduced into the gene drive
element, such as genes encoding antibodies directed against the malaria
parasite vector \textit{Plasmodium falciparum}, conferring resistance
to it \citep{Gantzetal2015}  or genes that would reverse pesticide
resistance \citep{Guichardetal2019}.  On the other hand, the insertion
of the gene drive element itself can create the desired mutation. For
example, gene drive elements were inserted at the \textit{doublesex}
locus in \textit{Anopheles gambiae} mosquitoes, making females sterile
while not affecting males. The later gene drives were found to
eradicate populations maintained in small cages \citep{Kyrouetal2018} 
as well as larger, age-structured populations in bigger indoor cages
within a year \citep{Hammondetal2021}.

\subsection{Applications}\label{sec23}

Two approaches are envisioned for gene drive, modification and
suppression \citep{NaidooOliver2024}.  The goal of the modification
approach is to spread a desired trait within a population. The expected
endpoint is a population where 100\% of its members carry the gene
drive element. In contrast, the goal of the suppression approach is to
reduce or eliminate a target population. Here the gene drive introduces
genetic changes that cause sex-specific sterility, sex-ratio
distortion, or other traits that decrease reproductive success, leading
to population collapse. The suppression approach is currently more
advanced than the modification approach, as it is easier to find
genetic changes that alter fertility and survival than some that
introduce desired, specific traits such as resistance to given
pathogens in host insects, preventing them from being disease carriers
\citep{Gantzetal2015},  or insecticide susceptibility
\citep{Kaduskaretal2022}.  While some researchers have compared the
modification approach with insecticides and the suppression approach
with vaccines \citep{NaidooOliver2024},  we would like to note here
that these analogies convey incorrect ideas about associated risks. In
contrast to vaccines, the final population after a modification drive
still carries a gene drive element that is susceptible to mutate and
incidentally pass onto another population. And in contrast to
insecticides, stopping the release of suppression drive individuals is
not sufficient to stop the gene drive. These points will be further
developed in Sections~\ref{sec53} and~\ref{sec4},\break respectively.

A wide range of applications are envisioned for gene drives, not only
in public health, but also in agriculture and conservation biology 
\citep{Esveltetal2014}.  Gene drive may help control or eliminate
vector-borne diseases by targeting disease-carrying organisms, in
particular mosquitoes for malaria, dengue, Zika, and chikungunya. This
technology could also offer potential solutions to control insect pests
(fruit flies, locusts, etc.) that damage crops or affect livestock.
Gene drive has also been proposed to reintroduce susceptibility genes
in weeds, to make herbicides effective again \citep{Neve2018}. Finally,
gene drives could help protect endangered species by controlling
invasive species or by introducing beneficial traits to at-risk
populations \citep{Rodeetal2019}.
\par\vspace*{1.5pt}\pagebreak

\subsection{Current status}\label{sec24}

In the laboratory, gene drive systems have been engineered in yeasts
and various sexually reproducing animals including flies, mosquitoes
and mice (Table~\ref{tab1}). Applying gene drives to plants is challenging for
several reasons. First, many plants reproduce vegetatively, via tubers,
cuttings, etc. or through self-pollination, thus limiting the spread of
a gene drive. Second, unlike insects or rodents, plants usually have
long life cycles, slowing down the spread of a gene drive. Third, many
plants have polyploid genomes, making gene drive inheritance harder to
control. Fourth, homology-directed repair, which is necessary for the
gene drive element to copy itself, does not appear to be as frequent in
plants as in \mbox{animals} \citep{GorbunovaLevy1999,Lietal2013}.  Recently,
other types of gene drives, which are not ``homing'' and do not rely on
homology-directed repair, have been developed in the model plant
\textit{Arabidopsis thaliana} \citep{Liuetal2024,Oberhoferetal2024}.
These are not examined in this study.

\begin{table*}
\caption{\label{tab1}List of species in which CRISPR homing gene drive
has been demonstrated}
\tabcolsep 11pt
\begin{tabular}{ccc}
\thead
{Taxonomic group} & {Species} & {References} \\
\endthead
Flies & \textit{Drosophila melanogaster} & \citet{GantzBier2015} \\
Mosquitoes & \textit{Anopheles stephensi} & \citet{Gantzetal2015} \\
Mosquitoes & \textit{Anopheles gambiae} & \citet{Hammondetal2016} \\
Yeasts & \textit{Saccharomyces cerevisiae} & \citet{Dicarloetal2015} \\
Yeasts & \textit{Candida albicans} & \citet{Shapiroetal2018} \\
Mammals & \textit{Mus musculus} & \citet{Grunwald2019} \\
Mosquitoes & \textit{Aedes aegypti} & \citet{Lietal2020} \\
Mosquitoes & \textit{Culex quinquefasciatus} & \citet{Harvey-Samueletal2023} \\
Flies & \textit{Drosophila suzukii} & \citet{Yadavetal2023} \\
Flies & \textit{Ceratitis capitata} & \citet{Meccarielloetal2024} 
\botline
\end{tabular}
\tabnote{Studies are presented in chronological order.
Only the first published study for each species is mentionned.}
\end{table*}

To date, no gene drive has been released in wild populations. In
November~2018, the 195 states that are signatories of the Convention on
Biological Diversity, a multilateral treaty (not ratified by the USA)
adopted a text indicating that the release of gene drive organisms in
the wild must occur after obtaining ``the free, prior and informed
consent of indigenous peoples and local
communities''\footnote{Decision 14/19. `Synthetic
Biology'. Decision adopted by the Conference of the Parties to the
Convention on Biological Diversity.'' (November 30, 2018). Online at
\url{https://www.cbd.int/doc/decisions/cop-14/cop-14-dec-19-en.pdf}.}.

``Risk'' and ``danger'' are two words that are often confused. Danger
refers to an inherent hazard or threat posed by a technology or
process, regardless of probability, while risk refers to the likelihood
of an adverse event occurring due to a specific action or technology,
often expressed as a combination of probability and impact. Danger
focuses on the potential severity of harm if the event occurs, while
risk accounts for the likelihood of the event. Regarding gene drive,
risk assessment should evaluate potential positive and negative
consequences at various levels, including the target population,
non-target populations, ecological effects, and socio-economic \mbox{impacts}.
Because likelihoods are extremely difficult to estimate for gene
drives, especially in the general case, we chose here not to evaluate
probabilities but to characterize the various possible unintended and
adverse outcomes of gene drive. We therefore use the term ``risk'' in a
broader sense. We first examine scenarios where the technology may fail
or may fail to be contained. Then we explore four types of adverse
scenarios: (a) ecological risks, where the gene drive would have
unintended consequences at the level of the target population, the
other species or the ecosystem; (b) sociological risks associated with
governance and public perception; (c) risks associated with
experimentation; and (d) risks associated with malevolent usage.

\section{Technical risks}\label{sec3}

As for technical risks, we consider here cases where the proposed
biotechnology would not be as efficient as planned, and would not lead
to the expected benefits. Such cases do not truly constitute adverse
events, but simply failures.

\subsection{Cryptic species}\label{sec31}

Cryptic species correspond to interbreeding individuals that are
morphologically similar to a given species but are unlikely to
hybridize with the members of this other species, for various reasons
\mbox{(incompatible} behaviors, different ecological niches, physiological
incompatibilities, etc.). In practice, this means that one species name
refers to two or more reproductively isolated species
\citep{Bickfordetal2007}.  As a consequence, a gene drive targeting a
particular species cannot reach the cryptic species. Even if the gene
drive element is designed to be able to insert into the genome of the
cryptic species, it will not do so because released animals---and their
progeny---will not hybridize with the individuals belonging to the
cryptic species. For gene drives to spread within cryptic species,
individuals able to mate with the cryptic species must be released. 

Cryptic species are not rare in the \textit{Anopheles} genus. In
Western Kenya, several cryptic species of \textit{Anopheles}
mosquitoes, which are susceptible to transmitting malaria, have been
recently identified using molecular markers \citep{Zhongetal2020}.  The
high number of cryptic species challenges vector control strategies
targeting \textit{Anopheles} mosquitoes. 

In the case of suppression drives in mosquitoes, the successful
elimination of a population targeted by a gene drive could leave an
empty ecological niche that a cryptic species of mosquitoes could fill.
This could thus lead to the expansion of this cryptic species. Under
such a scenario, the initial gene drive would fail. Additional gene
drives, constructed into the genomic background of the cryptic species
(so that they can reach them via mating), would have to be used on top
of the initial one to try to eliminate the cryptic species.
\mbox{Observations} in the southwest Pacific suggest that \textit{Anopheles}
species diversity can hamper vector control strategies
\citep{Russelletal2013}.  Repeated spray of insecticides led to the
\mbox{disappearance} of the most \mbox{biting} mosquitoes \textit{Anopheles
punctulatus} and \textit{Anopheles koliensis}. However,
\textit{Anopheles farauti} populations went back to pre-spray levels
within a few years, and replaced the populations of 
\textit{A.~punctulatus} and \textit{A.~koliensis}. 
This was apparently due to the
evolution of a new, adapted behavior: avoiding insecticide exposure, by
blood-feeding early in the evening and outdoors. Conversely, in another
setting, a two-year longitudinal study in urban Singapore found that an
intensive \textit{Wolbachia}-based suppression of \textit{Aedes
aegypti} populations did not lead to niche replacement by the other
mosquito species \textit{Aedes albopictus} \citep{Wongetal2025}. These
two examples highlight the difficulty of predicting the dynamics of
natural populations of cryptic species and closely related species in
response to control measures. To devise successful gene drive
strategies, it is important to examine breeding habitats and habits for
the target species and to look for possible cryptic species.

\subsection{Gene drive resistance}\label{sec32}

Target populations might evolve resistance to the gene drive, either
via resistance alleles already present (standing genetic variation) or
via new mutations \citep{Druryetal2017,Uncklessetal2017}.  The
molecular mechanism of gene drive propagation involves a Cas9-induced
DNA break that is repaired by homologous recombination (see
Section~\ref{sec22}). However, if the cut is not repaired by homology-directed
repair, non-homologous end-joining (NHEJ) and microhomology-mediated
end-joining (MMEJ) events can change the sequence of the target site.
As a result, the mutated target site is no longer recognized by the
guide RNA and thus constitutes a resistance allele. Additionally, when
a gene drive allele is inherited from the mother, the Cas9/guide RNA
complexes deposited in the egg can cut the wild-type male DNA before it
comes close to the female DNA, precluding homology-directed repair, and
this can generate resistant alleles \citep{Bishopetal2022}.  The
evolution of resistance alleles represents a major obstacle to
successful applications of gene drive technology
\citep{Carramietal2018,Hammondetal2017,Uncklessetal2017}. 

So far, four strategies have been explored to avoid gene drive
resistance. The first one is to use multiple guide RNAs within the gene
drive construct, each targeting a nearby site
\citep{Champeretal2018,Champeretal2020a}  or with guide RNAs 
{targeting}
both the wild-type and the most common resistance alleles
\citep{Bishopetal2022}.  Even if one site gets mutated, the other
one(s) or the mutated site can still be used by the gene drive element
to insert itself. Experiments using Drosophila show that the addition
of a second guide RNA can indeed diminish the resistance rates
\citep{Champeretal2018}.  The target sites should be chosen relatively
far apart to prevent mutations at one site from converting an adjacent
target site into a resistance allele. Interestingly, increasing the
number of guide RNAs cannot ameliorate gene drive efficiency
indefinitely. This is because guide RNA additions are associated with
larger gene drive elements, and as homology arms are further apart, the
efficiency of homology-directed repair is reduced
\citep{Champeretal2020b}. Theoretical work indicates that, depending on
the type of drive and various performance characteristics, the optimal
number of guide RNA targets varies from two to eight
\citepalias{Champeretal2020b}. Furthermore, finding several guide RNA target
sites (with no off-targets) within a short genomic region can be
difficult for populations with high genetic diversity, such as
\textit{A.~gambiae} \citep{Champeretal2018}.\looseness=-1

A second approach is to restrict the expression of the \textit{Cas9}
gene or the guide RNAs to a developmental stage when homology-directed
repair is predominant and end-joining pathways are nonexistent
\citepalias{Champeretal2018}.  This strategy requires specific promoters
that may not be available for all species \citep{Duetal2024}.
Furthermore, it does not fully eliminate resistance because incomplete
homology-directed repair can still create small insertions and
deletions that confer resistance \citep{Champeretal2018}.

A third strategy is to target a gene encoding an essential protein, and
rescue it by providing within the drive element a modified gene
encoding the same essential protein using different codons, so that it
is not cleaved by the drive. Resistance alleles are expected to disrupt
the function of the target gene, reduce fitness and, consequently, be
eliminated from the population over time. Targeting a haplolethal gene
allows the rapid elimination of resistance alleles, since heterozygous
individuals carrying the \mbox{resistance} \mbox{allele} are lethal. A rescue drive
with two guide RNAs targeting the haplolethal gene~\textit{RpL35A},
encoding for a ribosome protein, was successfully implemented in
Drosophila  \citep{Champeretal2020a}. {However}, engineering similar
rescue drives in non-model species is challenging for several reasons:
identifying a bona fide haplolethal target gene in a non-model species
genome can be fastidious; the rescue can be difficult to implement; and
recoded regions may not fully prevent homology-directed recombination
\citep{Chenetal2023}.  An easier path is to target a haplosufficient
gene. Compared to haplolethal genes, these are more numerous in genomes
\citep{Deutschbaueretal2005}.  In this case, the resistance alleles are
eliminated more slowly, as they can be maintained in heterozygous,
viable individuals. Such drives were constructed with one guide RNA in
\textit{Anopheles stephensi} \citep{Adolfietal2020}  and with one or
four guide RNAs in \textit{D.~melanogaster}
\citep{Houetal2024,Kanduletal2021,Terradasetal2021},  and all evolved
resistance. In the most recent study with four guide RNAs
\citep{Houetal2024}, resistant individuals were found to carry a large
deletion in the coding region of the targeted gene. This shows that
targeting an essential gene does not suffice to prevent resistance
alleles: the precise sequence targeted by the drive within the gene
should be essential for the function of the\break protein.

A fourth approach is to target a highly conserved site, so that
eventual mutations at the target site are deleterious and give no
progeny. Kyrou et al. devised a clever gene drive system in 
\textit{A.~gambiae} with a single guide RNA targeting a highly conserved sequence
that is essential for the function of the female-specific Doublesex
isoform protein \citep{Kyrouetal2018}.  Females that are homozygous for
the gene drive are sterile and unable to bite, whereas males are
fertile. This drive can thus disperse via males and via somatic
heterozygous females. This drive was able to spread rapidly in cage
populations, leading to the full elimination of all mosquitoes in 8-12
generations. Cas9-resistant variants appeared, but they did not block
the spread of the drive, which is consistent with the fact that these
Cas9-resistant alleles are expected to produce nonfunctional Doublesex
isoform proteins and thus sterility. The \textit{Doublesex} locus was
also targeted recently by gene drive in \textit{D.~suzukii}, and all
the few Cas9-resistant alleles that were detected similarly led to
sterile females, indicating that they would not prevent the spread of
the drive \citep{Yadavetal2023}.  So far, this last approach seems to
be the most effective one to prevent resistance. But it only applies to
suppression drives, and it is still unclear whether it would be
resistance proof in broader {settings} and native ecological conditions
\citep{Kyrouetal2018}.

Ten years after the first proof-of-principle experiments
\citep{Gantzetal2015,GantzBier2015},  gene drive technology is still
not ready for release. For a gene drive to be successfully implemented
in the wild, further technical work is required to make sure that
resistance will not evolve and that the mating system of the targeted
population will allow the drive to spread to all individuals.

\section{Risk of ineffective mitigation}\label{sec4}

In case of unintended effects following a gene drive release, one may
want to stop the drive. However, halting a drive is not as easy as with
insecticides, where stopping treatment will prevent further spreading
and should ultimately put an end to the damage. A released gene drive
is likely to continue to spread within the targeted population, even if
releases have stopped. If gene drives are to be implemented in the
wild, it is important to have effective methods to be able to stop them
if\break necessary. 

Two types of ``reversal drives'' or ``brakes'' capable of overwriting
or neutralizing an existing gene drive have been proposed
\citep{Wuetal2016,Xuetal2020}.  They are themselves gene drives that
encode guide RNAs but not the Cas9 protein. One type of reversal drive
inserts itself within the \textit{Cas9} gene of the initial gene drive
element and inactivates it. The inactivated drive thus remains in the
genome. The other type of reversal drive contains two guide RNA genes
that allow the excision of the entire initial gene drive element and
its replacement with the reversal drive that is devoid of the
\textit{Cas9} gene. Theoretical models show that the temporal dynamics
of gene drives and reversal drives are complex and depend on multiple
factors \citep{Girardinetal2019,Rodeetal2020,Vellaetal2017}.  In
certain conditions pertaining to the relative fitness of the drive, the
reversal drive and the wild-type alleles, the moment at which reversal
drives are released, and the \mbox{heterogeneity} in the spatial distribution
of the population, some drives may be\break unstoppable.

Another strategy is to use a transgene inherited in a Mendelian fashion
and encoding anti-CRISPR proteins
\citep{Basgalletal2018,DAmatoetal2024,Taxiarchietal2021}. Anti-CRISPR
proteins are natural molecules found in phages that are able to
\mbox{inhibit} Cas9 activity and can thus neutralize the
\mbox{activity} of a gene drive. This method does not involve DNA
breaks and repair events, so off-target effects are minimal and the
outcome is more predictable than with reversal drives. In large-size
cages containing age-structured populations of mosquitoes, the spread
of a suppression drive targeting the \textit{doublesex} locus was
stopped by releasing anti-drive males at a 30\% allelic frequency every
three to four days for several weeks, until the termination of the
experiment \citep{DAmatoetal2024}.  Sequencing at various time points
revealed that functionally resistant alleles were not selected over the
course of the experiment. However, drive alleles were still present
after more than 200 days, raising uncertainty about whether they would
manage to spread again had anti-CRISPR male releases been paused for
several weeks. These findings indicate that in a complex, near-natural
environment, continuous releases of anti-CRISPR transgenic individuals
would be required to effectively counteract the spread of a potent
suppression gene drive and prevent the elimination of the target
population. Further experimental validation and mathematical modeling,
particularly in spatially heterogeneous environments, will be necessary
to assess the range of conditions allowing this anti-drive strategy to
successfully mitigate gene drive propagation.

Overall, the anti-drive approach appears as a more effective
countermeasure against gene drives than the reversal drive strategy.
Further research will be needed to ensure its safety and efficacy
across diverse ecological and genetic contexts. An important issue with
all the proposed countermeasures so far is that they require the
release of new individuals and close monitoring of the target
population and ecosystem. This means that if an initial gene drive
release proves problematic, local communities are condemned to pursue
remediation actions with the research team. This may cause additional
risks, and highlights the fact that sociological factors play an
important role in the debate around gene drives
(see Section~\ref{sec6}).

\section{Ecological risks}\label{sec5}

Even if gene drives may not be as effective as anticipated (see
Section~\ref{sec3}), they may still have unintended effects on the
target population, on other species, or on entire ecosystems. While the
molecular-level \mbox{consequences} of gene drives can often be
predicted with relatively good precision, their impacts become
increasingly difficult to foresee as we move to higher levels of
biological organization---from phenotypic traits within the target
species to broader ecological interactions---, due to the growing
complexity and interconnectedness of these systems.

\subsection{Complexity of the environment and unpredictability of the
gene drive element}\label{sec51}

What makes ecological risks associated with gene drives especially
difficult to apprehend is two-fold. First, gene drives are designed to
act in native ecosystems, where multiple species interact with each
other, as opposed to ``classical'' genetic engineering of domesticated
species, which usually live in standardized environments isolated from
their wild counterparts. Second, the gene drive element is a
self-replicating genetic element that is extremely labile given its
design: many types of mutations can occur in the gene drive element and
transform an original drive into a new one, with novel potential
adverse effects on the phenotype of the drive-carrier animals and on
ecosystems. Indeed, a new cargo gene may insert into the element,
bringing new phenotypic potentialities such as insecticide resistance
or increased attraction to humans. Furthermore, mutations in the
sequence of the guide RNA gene can change the cut site, and the drive
may insert at a new position in the genome, using as homology arms
repeated sequences distributed across genomes. 

Given the infinitesimal size of the gene drive element compared to the
rest of the genome, mutations are expected to be more frequent outside
of the gene drive element than within. Nevertheless, if mutations occur
in the gene drive cassette, these mutations, compared to mutations
occurring at other positions in the genome, are likely to have larger
consequences at the level of mosquito populations. This is because gene
drives have a higher ability to spread in populations due to their
non-Mendelian inheritance. For example, insecticide resistance alleles
usually decrease mosquito fitness in the absence of insecticides, and
are thus rapidly eliminated in the absence of insecticide
\citep{KliotGhanim2012}. But if a resistance gene inserts within a
gene drive, then its non-Mendelian inheritance can compensate its
detrimental fitness effect in untreated regions, and so lead to
increased spread of the resistance.

Regarding the range of mutations, it is interesting to compare gene
drive technology with \textit{Wolbachia}-based methods for mosquito
control \citep{Wangetal2024}.  The latter strategies also rely on the
spread of self-replicating genetic elements---the bacterial symbionts
\textit{Wolbachia}---, which are susceptible to mutations as well.
Regarding mutations in the \textit{Wolbachia} genome, for them to
persist over multiple generations, they must be compatible with the
survival and function of the symbiont, which depends on complex
biological processes such as metabolism, cell division, and migration.
Therefore, the range of possible persisting mutations appears to be
more limited for \textit{Wolbachia} than for gene drives. With respect
to possible phenotypic effects of future mutations, comparing those in
\textit{Wolbachia} with the ones in gene drive elements is tricky.
\textit{Wolbachia} has been infecting insects for many millions of
years \citep{Sanaeietal2021} so that it has evolved strong abilities
to alter the insect's phenotype. But on the other side, insects have
also evolved means to counteract \textit{Wolbachia} effects. In
contrast, the gene drive element encoding the Cas9 protein and guide
RNAs is a new self-replicating entity, never encountered before by
insects.\looseness=-1

Overall, the risk of gene drives adopting new functions is higher for
modification drives than for suppression drives, because modification
drives are maintained within the target population whereas the
suppression drives are designed to be eliminated together with the
target population. Gene drive elements, whether in their intended form
or following mutations, may induce unexpected changes in the target
species, potentially impacting other species and the broader ecosystem.

\subsection{Disruption of ecosystems}\label{sec52}

The rapid spread of gene drives can lead to unanticipated ecological
impacts. Suppression drives are particularly worrisome because they
intend to eliminate an entire species from an ecosystem. The concern
here is the same as for the use of insecticides or other methods to get
rid of a given species. Suppressing one species may cause cascading
effects, altering predator--prey relationships and food web stability.
The removal of all the cats on Port-Cros island in France led to rat
proliferation, whereas the stabilization of the cat population at
around 250 individuals on the neighboring Le Levant island was
associated with controlled rat populations (A. Atlan, personal
communication). Furthermore, a species considered to be invasive or
detrimental by some communities may actually be regarded as valuable by
others \citep{Carroll2011,Davisetal2011}.  For example, kudzu in Japan
is a native plant considered beneficial for erosion control, animal
fodder, and even traditional medicine. In contrast, in the United
States of America it is most commonly seen as a notorious invasive
species, growing uncontrollably, although there are a few vocal kudzu's
supporters \citep{AldermansndAlderman2001}.  The target species may
have unknown ecological functions in the ecosystem, and its elimination
may thus fragilize the ecosystem. For example, invasive black rats may
be responsible for dispersing seeds of native plants, a function
previously undertaken by the native rodent species replaced by black
rats \citep{ShielsDrake2011}.  Eliminating black rats may then affect
the dissemination of native plants.

The goal of eliminating an invasive species might be to return to an
ancestral equilibrium similar to the one prior to the invasion. But
local eradication of an invasive species leaves an empty ecological
niche that can trigger diverse cascading trophic effects on the
distribution and abundance of multiple other species in direct and
indirect interaction with the eliminated species
\citep{Zavaletaetal2001}.  For instance, the removal of feral goats and
pigs from Sarigan Island, a US territory in the northwestern Pacific,
triggered the proliferation of the previously undetected invasive vine
\textit{Operculina ventricosa}, which subsequently spread within the
ecosystem \citep{Kessler2002}. In certain conditions, eradicating an
invasive species can even make the system more susceptible to new
invasions \citep{Davidetal2017}.

Currently, gene drives are primarily being studied by geneticists and
molecular biologists. However, given their potential impact on
ecosystems, it is crucial for ecologists and evolutionary biologists to
actively engage in research about the possible ecological consequences
of gene drive.


\subsection{Propagation to non-target populations}\label{sec53}

CRISPR homing gene drives are expected to be highly invasive in nature
\citep{Nobleetal2018}.  A well-documented example of global
dissemination is the natural transposable element named the P element,
which rapidly spread across all natural populations of the fruit fly
\textit{Drosophila melanogaster} worldwide during the mid-20th century,
probably following its horizontal gene transfer from another Drosophila
species \citep{Anxolabehereetal1988,ClarkKidwell1997}. A
primary concern is thus that gene drives designed to target a specific
population may contaminate other populations of the same species. The
longer the gene drive is present in the wild, the greater the risk.
Here again, the risk is higher for modification drives than for
suppression drives.

To control invasive black rats and house mice in New Zealand, gene
drive approaches have been considered
\citep{Leitschuhetal2018,Prowseetal2017}.  Islands are not as isolated
as they might appear. Genomic studies have shown that rodents can move
between islands, often facilitated by human activities such as maritime
transport \citep{Sjodinetal2020}.  To develop population-specific gene
drives, a potential strategy is to target unique DNA sequences present
exclusively in the intended population 
\citep{Esveltetal2014,Sudweeksetal2019}. However, this approach
presents significant challenges, as it requires capturing the full
extent of genetic variation across all natural populations, which is
nearly impossible. Additionally, gene drives are susceptible to
mutations (see Section~\ref{sec51}), and may thus gain the capacity to spread
in other populations, although they were not designed to do so. 

\subsection{Propagation to non-target species}\label{sec54}

Another risk is the propagation to closely related species. In fact,
some gene drive strategies targeting the \textit{Anopheles} species
complex are intentionally designed to propagate across multiple closely
related species, as several members of this complex serve as vectors
for malaria transmission \citep{Nolan2021}. Targeting the highly
conserved \textit{doublesex} locus is a promising strategy for
developing suppression drives in \textit{Anopheles} mosquitoes (see
Section~\ref{sec32}). However, this ultraconserved sequence is found in
several species of \textit{Anopheles} and is also likely to be
conserved in other closely related species. Using highly conserved
sequences as target sites increases the risks of propagation to closely
related species, even if they were not necessarily targeted initially.

Even if hybridization between two closely related species occurs very
rarely, as long as it happens at some appreciable frequency, the risk
of transmitting the gene drive to another species exists
\citep{Courtier-Orgogozoetal2020}.  The P element that had previously
contaminated all natural populations of \textit{D.~melanogaster} is now
invading wild populations of the closely related species 
\textit{D.~simulans}
\citep{Hilletal2016,Kofleretal2018,Kofleretal2015,Nascimentoetal2020}, 
and it probably started via an interspecific cross with 
\textit{D.~melanogaster}. Data indicate that hybridization and interspecific
genome mixing occur sporadically between species previously thought to
be reproductively isolated. For example, hybridization can occur
between the black rat \textit{Rattus rattus} and the Asian rat
\textit{R.~tanezumi} \citep{Lacketal2012},  as well as between the fly
pest species \textit{Drosophila suzukii} and its close relative
\textit{D.~subpulchrella}
\citep{Conneretal2017,Courtier-Orgogozoetal2020}.  Therefore, gene
drives targeting these species are likely to end up in the closely
related species.

In addition to hybridization, DNA can be naturally transferred from one
species to another through horizontal gene transfer, mostly via viruses
and microorganisms that can carry over pieces of DNA
\citep{GilbertCordaux2017}.  Gene drive elements resemble homing
endonuclease genes, which are natural transposable elements that can
bias their inheritance by cutting and inserting themselves at targeted
sites within genomes \citep{AgrenClark2018}.  A homing endonuclease
gene targeting the \textit{cox1} mitochondrial gene has been found in
multiple species of plants, fungi and green algae. Phylogenetic studies
revealed that it has been transferred independently 70~times between
162~plant species involving 45 different families
\citep{Sanchez-Puertaetal2008}.  This suggests that genetic elements
with a transmission advantage, such as gene drives, can possibly reach
distantly related species by horizontal gene transfer. Notably, for a
given gene drive construct to contaminate a distantly related species,
six conditions should be met: (1) horizontal transfer; (2) expression
of the \textit{Cas9} gene and the guide RNA gene in the new host; (3)
presence of a target site in the host genome; (4)~presence of flanking
sites to allow the insertion of the gene drive element; (5) resistance
of the gene drive to the host immune system; and (6) survival and
reproduction of the contaminated individuals
\citep{Courtier-Orgogozoetal2020}.  In-depth \mbox{examination} of each of
these conditions indicates that the probability of a gene drive element
to contaminate another species is not null, especially because of the
presence of repeated sequences in genomes that may facilitate
homology-directed repair \citepalias{Courtier-Orgogozoetal2020}.

In summary, the range of ecological risks highlighted here underscores
the need to thoroughly assess the ecological role of the target species
within its ecosystem. It also stresses the importance of involving
ecologists in the planning and decision-making process before any
intervention to try to minimize unintended consequences.

\section{Sociological risks}\label{sec6}

Gene drive technology presents significant sociological risks that go
beyond technical and ecological concerns. These risks stem from public
perception and governance issues. As previously seen with COVID-19
vaccines during the pandemic, negative public perception can rapidly
spread via social media \citep{Rodriguesetal2023}.  One of the greatest
risks is an unauthorized release of gene drive organisms into the wild,
whether accidental or deliberate, because this could severely damage
public trust in scientists, institutions or regulators
\citep{Esvelt2018,Minetal2018}.  If gene drives are overly hyped
\citep{Boete2025}, or perceived as being deployed recklessly or without
full transparency, this could lead to increased skepticism toward
genetic engineering and heightened public opposition to biology
research activities in general, even those with clear potential
benefits and little risk. Furthermore, this could undermine trust in
public health in general, and thus reduce the effectiveness of outbreak
prevention and control measures \citep{WorldHealthOrganization2012}.

If gene drives are released without proper engagement with local
populations, affected communities may see gene drives as an unnatural
disruption of their environment and may feel disempowered. This could
lead to accusations of biocolonialism, where genetic technologies are
imposed on vulnerable populations without adequate consultation or
compensation. This issue is particularly acute for malaria control,
where gene drive research is primarily conducted in wealthy countries,
but intended for deployment in the Global South and especially in
sub-Saharan Africa. Efforts are being made to involve local communities
in Burkina Faso and to foster inclusive, well-informed discussions
\citep{Sykesetal2024}.  The ``Mice Against Ticks'' project in Nantucket
and Martha's Vineyard islands represents an interesting example of how
residents identified potential ecological consequences that were
overlooked by researchers, highlighting the importance of community
input \citep{Buchthaletal2019}.  Non-scientists---such as local
communities, farmers, or citizen scientists---can sometimes be
extremely well informed about the various technical aspects and can
contribute valuable knowledge and perspectives to scientific
discussions. For example, to reduce the risk of catastrophic wildfires
in Australia, modern land management strategies benefited from the
integration of Aboriginal fire management practices, such as controlled
burns \citep{Ensetal2015}.

Furthermore, gene drive organisms do not respect national borders: one
country's decision to use the technology could be seen as imposing
risks on neighboring nations and could result in diplomatic incidents.
This adds complexity to the development of international regulation,
particularly with respect to decision-making and accountability
\citep{Beck2019}. Currently, no international regulatory framework
exists for gene drive technology.

To address the sociological risks associated with gene drive
technology, scientists and policymakers must prioritize transparency,
inclusivity, and active engagement with local populations. Public
participation in the debate is essential but not sufficient.
Researchers and policymakers should also be tolerant towards diverse
perspectives: they should respect and take into account the views of
others. 

\section{Risks associated with experimentation}\label{sec7}

No experimental manipulation is completely free of human error. The
accidental release of just a few gene drive individuals during testing
may lead to contamination in the wild. Biologists have been aware of
this issue since the early days of gene drive research and have
implemented a series of rules for laboratory research on gene drives
\citep{Akbarietal2015,Esveltetal2014,NationalAcademiesofSciencesEngineeringandMedicine2016,Oyeetal2014}. 
These include conducting research in regions where the target species
is not present and cannot survive and reproduce, and using barriers and
protocols more stringent than what is usually used for genetically
modified organisms to prevent escapes. While standard scientific
practice encourages preserving genetically modified organisms to allow
replication and further research, it is considered that gene drive
research makes an exception to this rule and that gene drive organisms
should be appropriately discarded once an experiment is over.

Recently, Zapletal et~al.~proposed to use ``biodegradable gene drives''
for research \citep{Zapletaletal2021}. These gene drives encode for
site-specific recombinase/transposases that will lead to excision of
the gene drive element and reversion of the modified chromosome to a
wild-type allele. Although these gene drives have not been tested
experimentally, mathematical modeling indicates that strong homing
biodegradable gene drives start to spread and then are eliminated.
However, since these drives do not fully correspond to the ultimate
gene drive applications, their practical relevance for laboratory
studies and field trials remains uncertain.

A more common and safer strategy for gene drive research is to use
split gene drives
\citep{Akbarietal2015,Champeretal2019,Chenetal2024,Terradasetal2021}.
Unlike standard gene drives, where all necessary genetic components are
present within a single construct, the \textit{Cas9} gene is inserted
into one genomic location and the guide RNA and repair template are
placed at another. Since the two components are not inherited together,
the split gene drive is self-limiting: it can function within a
controlled laboratory setting when both elements are present in an
organism but cannot spread indefinitely in wild populations. This
strategy of using elements with no autonomous replication ability
resembles the use in virology research of pseudoviruses, which can
infect only a single cycle of host cells \citep{Xiangetal2022}.

International discussions around gene drive research have not yet
converged on a set of rules and guidelines to be respected worldwide.
As of today, regulations vary between countries
\citep{GeneticLiteracyProject2025}.  To our knowledge, there is no
independent international body examining gene drive research projects
before and after they are implemented.

\section{Risks associated with malevolent usage}\label{sec8}
Let us now consider the possibility that gene drive could be
malevolently used by states or non-state groups, for military or
terrorist objectives. In theory, two types of biological weapons could
be built based on gene drive: suppression drives that would target
insect species necessary for agriculture (e.g.\ pollination) and
modification drives that would render pest or disease-carrying insects
resistant to insecticides or able to deliver toxins to humans.

The Biological Weapons Convention, which bans the development of
offensive biological weapons, was signed by most countries in 1972
\citep{UnitedNations1972}.  Ten states have neither signed nor ratified
the treaty: Chad, Comoros, Djibouti, Eritrea, Israel, Kiribati, the
Federated States of Micronesia, Namibia, South Sudan and Tuvalu
\citep{UnitedNations2019}.  Unfortunately, sporadic reports suggest
that clandestine biological weapon development programs may still be
underway in rival nations, despite their signature of the Biological
Weapons Convention. Current data show that several terrorist groups,
such as Aum Shinrikyo, the Islamic State and Al-Qaeda, have put efforts
into developing biological weapons in the past
\citep{Danzigetal2012,ParachiniGunaratna2022}.  Given the current
situation, the danger associated with malevolent uses of gene drive 
technology should be examined seriously.

The use of biological weapon is intrinsically linked to the development
of antidotes and vaccines, to make sure that the attackers and their
allies are not affected by the biological weapon itself. Compared to
standard biological agents that target humans directly---via pathogenic
agents, such as toxins, viruses and bacteria---, the targets of gene
drive technology in its current form are sexually reproducing animals
with short life cycles, mostly insects and rodents. For special cases
where gene drives would be designed to deliver toxins to humans,
vaccines or antidotes would be required as for standard bioweapons. For
other cases, it seems unrealistic to envisage trying to immunize
certain insect/rodent populations prior to the release of a nefarious
gene drive: as explained above, current available strategies to prevent
a given gene drive from spreading (before or after it is released) are
not spatially restricted and do not appear to be fully satisfactory.
Alternatively, a possible usage of a rogue gene drive would be to
target an insect species that is present in the enemy region, but not
in the attackers' region. Then, the main difficulties lie in using
CRISPR to genetically modify the target insect species, and in
{maintaining} and raising sufficient numbers of transgenic 
animals for
future release. Releases can easily go undetected for insects, since
they are very small. Overall, in the current state of knowledge,
potential applications of rogue gene drives concern a small number of
animal species, which display the following characteristics: they
reproduce sexually, have a short life cycle, are necessary for
agriculture (e.g.\ pollination) or act as pests or carriers of disease,
are not present in the territory of the attacker, can be reared, bred
and maintained in large number in the laboratory, and are amenable to
effective CRISPR technology. In 2017, the Defense Advanced Research
Products Agency (DARPA) in the United States of America allocated
funding to address potential malevolent uses of gene drive and to
develop defensive countermeasures. However, relatively few research
articles deal with this topic, partly to avoid the dissemination of
dangerous ideas. Esvelt and his colleagues have proposed to monitor
at-risk regions via environmental sequencing, in order to detect gene
drive elements in any species, and to launch immunizing reversal drives
to counteract any drive element that would be found
\citep{Esvelt2018,Minetal2018}.  As long as several locations are
sampled and a relatively large number of reads is obtained from the
species of concern, sequencing is indeed bound to detect the gene
drive. Additionally, a mosquito-borne toxin would probably be noticed
from the clinical cases first. Immediate defenses would probably be
insecticides, and medium-term ones may involve immunizing reversal
drives. Designing such reversal drives is straightforward for the
defender, but accomplishing transgenesis in the target species and
releasing multiple reversal drive organisms in the wild to successfully
stop the malevolent gene drive may still be burdensome
\citep{Esvelt2018}. Overall, the monitoring and mitigation processes
can be costly and difficult, especially in low-income countries and in
areas experiencing political instability. Furthermore, immunizing
reversal drives are not always guaranteed to work
\citep{Rodeetal2020}. \looseness=-1

Since one bottleneck in the building process of rogue gene drives is
the insertion of the gene drive element in non-model insect species via
CRISPR, a first step in limiting rogue gene drives would be to exclude
from scientific papers the methodological details for applying gene
drive to non-model species, as for the technical instructions to make
nuclear weapons \citep{Gurwitz2014}.

\section{Conclusions}

This paper outlines four categories of risks associated with CRISPR
homing gene drive, and reviews current perspectives on each. Given the
issues at stake, it is important to carry out a general assessment of
the risks and benefits associated with gene drive. As each gene drive
application comes with its own specificity, it must undergo rigorous
evaluation before any release is considered. When evaluating gene
drives for mosquito population control, it is essential to carefully
consider alternative technologies, such as \textit{Wolbachia}-based
approaches or long-lasting insecticidal nets
\citep{Boete2025,Wangetal2024}.  Importantly, the success of a future
gene drive release should not facilitate subsequent deployments of the
technology, a concern known as the slippery slope argument
\citep{VanDerBurg1992}.  Regarding the possibility of malevolent use of
gene drive, we emphasize the need for deeper theoretical exploration
and concrete preventive measures to enhance global security. As
exemplified by the 1975 Asilomar conference \citep{Cobb2025}, ethical
committees and self-regulations by scientists are probably not
sufficient to protect against misuse. Moving forward responsibly with
gene drive technology requires increased public awareness and strong
political engagement. We hope that this work will help decision-makers
to carefully weigh the potential benefits of this biotechnology against
its risks, and that it will serve as a basis for public debate.

\section*{Abbreviations}

\begin{inftab}
\begin{tabular}{lp{5.5cm}}
\tbody
CBD & Convention on Biological Diversity\\
CRISPR & Clustered Regularly Interspaced Short Palindromic Repeats\\
DNA & Deoxyribonucleic Acid\\
IUCN & International Union for Conservation of Nature and Natural Resources\\
RNA & Ribonucleic acid
\end{tabular}
\end{inftab}

\section*{Acknowledgements}
We acknowledge the COMETS for general discussions about risky research.
We thank C. Bo\"{e}te, F. Graner, K. Esvelt and the anonymous reviewer
for comments on the article. AI was used to improve the fluency of
certain sections of the text.

\section*{Declaration of interests}

The author does not work for, advise, own shares in, or receive funds
from any organization that could benefit from this article, and has
declared no affiliations other than their research organization.

\section*{Funding}
This work was supported by the European Research Council under the
European Community's Seventh Framework Program (FP7/2007-2013 Grant
Agreement no. 337579) and by Agence Nationale de la Recherche (ANR)
under the project ``ANR-24-CE13-0018-01''.

\CDRGrant[ERC]{337579}
\CDRGrant[ANR]{ANR-24-CE13-0018-01}

\back{}

\printbibliography
\refinput{crbiol20250221-reference.tex}

\end{document}