From reversal drives to anti-CRISPRs, scientists are designing molecular backstops to stop gene drives if needed

This version (Version 1) was created on 29 July 2025

1. Why we need to undo the drive

Once released into a population, a gene drive can spread far beyond its point of origin. That is precisely the goal of the technology, and its greatest risk. By biasing inheritance, gene drives can rapidly alter or suppress entire populations, raising concerns about long-term impacts on biodiversity, food systems and national sovereignty (Rode et al., 2020; Tajudeen et al., 2023). Considering these risks, safeguards are not optional, but essential.

The first line of defence relies on intrinsic genetic biocontainment strategies. These are built directly into the gene drive system and include approaches such as threshold-dependent drives, conditionally active constructs, or designs that lose function over generations. They aim to restrict gene drive spread in space or time, or to limit persistence in the absence of specific molecular or environmental cues. I explore these approaches in detail in two companion living literature reviews (here and here).

However, as for any technology, no safeguard is perfect. Once a gene drive has been released, unexpected effects or unintentional persistence of modified organisms may still occur. In such scenarios, second-line countermeasures would become necessary: strategies designed to stop, reverse or neutralise the activity of a gene drive already present in a wild population (Vella et al., 2017; Rode et al., 2020; Devos et al., 2022).

A wide range of countermeasures is under development, from genetic releases such as reversal, immunising and anti-drives (Taxiarchi et al., 2021; LeMieux et al., 2019), to molecular tools such as small RNAs, engineered proteins and small-molecule inhibitors that act without further releases (Chennuri et al., 2022). While these strategies face scientific and logistical challenges, such as ecological uncertainty (Devos et al., 2022) and difficulties with molecule delivery in the field, their development is seen as essential for responsible governance. Even if never deployed, having credible ways to regain control could support public and regulatory confidence in future gene drive applications (LeMieux, 2019; Chennuri et al., 2022).

In the following sections, I describe the current landscape of countermeasure strategies under development for CRISPR-based gene drives. These strategies fall broadly into two categories: secondary releases of modified organisms intended to genetically counteract the original drive, and targeted inhibitors that interfere with gene drive activity through molecular means (Figure 1). This classification is not based on a formal or accepted framework. Rather, it reflects one possible way to group and interpret the available scientific literature.

2. Secondary releases

Releasing a second genetically modified organism to counteract the effects of a first may sound like chasing fire with fire, but this is precisely the idea behind secondary releases. These countermeasures involve introducing engineered individuals of the same species as the original gene drive, with the goal of halting, replacing or outcompeting it. All such approaches rely on some form of genetic interference, and all face the fundamental challenge of achieving equal or greater coverage than the initial release. I group these strategies into five broad categories: neutralising drives, reversal drives, immunising drives, daisy restoration drives and anti-drives. The boundaries between them are not clear-cut. Functions overlap, definitions shift and technologies are evolving rapidly.

2.1. Neutralising Drives

Neutralising drives aim to stop an active gene drive system by disabling its core components, without necessarily restoring the original genome. This concept is sometimes described as a form of "functional reversal" where the gene drive no longer operates, even though its traces may remain in the genome (Giese, 2020).

Two prominent designs have been demonstrated in laboratory conditions: the e-CHACR (erasing Construct Hitchhiking on the Autocatalytic Chain Reaction) and the ERACR (Element Reversing the Autocatalytic Chain Reaction). Both are self-copying elements that carry guide RNAs (gRNAs) but no Cas9 (Figure 2). They are designed to hijack the Cas9 enzyme already present in the drive-carrying individuals to perform their function (Xu et al., 2020; Bier, 2022).

• e-CHACRs spread through a population and mutate the Cas9 gene, effectively disabling the drive system’s ability to cut DNA. In cage trials, they reached 100% introgression and fully eliminated Cas9 activity within nine to ten generations.
• ERACRs, on the other hand, aim to delete and replace the gene drive cassette itself. They have shown complete replacement of a simple GFP-marked gene drive in about six generations, though with less reliability than e-CHACRs and a tendency to form unintended hybrid elements (Xu et al., 2020).

Because these systems depend on the Cas9 protein expressed by the original drive, they spread only in the presence of the drive and revert to Mendelian inheritance once Cas9 is inactivated or removed. This targeting mechanism makes them relatively specific, though not foolproof: imperfect copying, chimeric recombination events and incomplete replacement have all been observed (Xu et al., 2020; Bier, 2022).

Beyond remediation, neutralising elements such as CHACRs are also being explored as potential update tools. For example, if a gene drive loses efficacy, CHACR could be introduced to add or replace effector genes. These updating constructs could be designed to preserve gene function while introducing new cargo and follow similar spread dynamics to the original drive (Bier, 2022).

2.2. Reversal Drives

Reversal drives, sometimes called overwriting drives, aim to undo the functional effects of a previously released gene drive by replacing it with a new genetic element. These constructs carry a rescue gene to restore the function of the disrupted locus, a guide RNA targeting the original drive, and a source of Cas9 to drive their own inheritance (Chennuri et al., 2022). Their action is restricted to individuals that already carry the original drive, ensuring that wild-type genomes are not affected.

Proof-of-principle experiments in yeast have shown that such overwriting drives can efficiently replace earlier gene drives and restore gene function, achieving over 99% replacement across generations (Chennuri et al., 2022). While this may sound like a full reversal, the outcome is still a genetically modified organism carrying new transgenic material. If the motivation for eliminating the original drive is to return to a non-modified state, this approach may fall short of expectations.

Another proposed reversal approach is the CATCHA (Cas9-triggered chain ablation) system. Unlike overwriting drives, CATCHA carries guide RNAs but no Cas9. It relies on Cas9 produced by the original drive to target and disrupt essential sequences within the drive itself, particularly the Cas9-coding region (Vella et al., 2017). This leads to a loss-of-gene drive function in subsequent generations, while also copying the CATCHA element into the same genomic location.

Modelling studies offer mixed outlooks for reversal drives. Deterministic models (I explore modelling in this article) show that when released at high frequencies, they can reduce or displace the original drive and coexist with wild-type alleles (Figure 3) (Vella et al., 2017). However, stochastic simulations suggest that reversal drives may only be effective in specific contexts, such as when the original drive is threshold-dependent or the fitness costs of competing elements are small (Rode et al., 2020).

2.3. Immunising Drives

Immunising drives are designed to do more than neutralise or replace a gene drive; they aim to make a population resistant to the original drive while restoring any lost gene function. These systems, sometimes called immunising reversal drives (IRDs), are full gene drives in their own right. They carry Cas9, multiple guide RNAs targeting both wild-type and drive alleles, and a recoded version of the disrupted gene, allowing them to spread autonomously and overwrite the genetic landscape created by the initial drive (Chennuri et al., 2022).

The immunising element’s recoded gene provides two functions: it restores normal gene activity (such as a reversal drive) and prevents the drive from targeting the new sequence (thus immunising the host). This dual function gives IRDs an advantage in population replacement scenarios. Deterministic models suggest that IRDs can reach fixation quickly, regardless of release frequency, because they outcompete both the original gene drive and wild-type alleles (Vella et al., 2017; Rode et al., 2020). Stochastic simulations also rank IRDs among the most efficient genetic countermeasures (Figure 4), especially when they confer a fitness benefit by restoring gene function.

However, this efficiency comes with warnings. Because IRDs eliminate both wild-type and drive alleles, the final population remains entirely transgenic. The Cas9 gene remains active in the population, raising concerns about long-term genetic stability and the risk of additional unintended modifications (Rode et al., 2020).

2.4. Daisy restoration drives

Countermeasures discussed so far aim to neutralise or overwrite the effects of a gene drive, but they would not fully restore a population to its original, unmodified genetic state. Daisy restoration drives would be different. Built on the principles of daisy systems, these tools could be designed not only to counteract existing gene drives but also to purge all engineered sequences from a population, returning it, at least in theory, to a wild-type genome (Min et al., 2017).

Daisy drive systems are modular gene drives made up of linked elements that lose components over generations, limiting their spread. In the case of a daisy quorum drive, an additional safeguard is introduced: a genetic element that only allows modified individuals to reproduce if a certain threshold of quorum elements is present in the population. This enables population-wide genetic changes to be reversed by driving the population below that threshold, either by introducing wild-type individuals or using targeted suppression constructs that selectively reduce the fertility or viability of edited individuals.

A daisy restoration drive builds on this concept. It carries guide RNAs designed to overwrite the unwanted gene drive, targeting unique sequences from the rogue construct and replacing them with a new cassette linked to the daisy system. Once the target sequence has been overwritten across the population, the remaining daisy elements, including the quorum components, can be eliminated through selective pressure or further releases, ultimately restoring wild-type genetics (Min et al., 2017).

In one daisy restoration strategy, engineered suppression elements could be introduced to impose a reproductive cost specifically on individuals expressing the Cas9 enzyme from an unwanted drive. These suppression elements would disrupt essential genes through homing, leading to non-viable offspring when two carriers mate. However, daisy restoration could also operate through a more precise mechanism: by replacing the unwanted sequence itself with a cassette encoding guide RNAs that target both the construct and the ribosomal genes used by the quorum system (Min et al., 2017). This overwriting cassette can be tightly linked to the daisy quorum effect and spreads through the population, replacing all copies of the unwanted edit (Figure 5). Once that replacement is complete, quorum-dependent suppression mechanisms would ensure that the remaining engineered elements are eliminated, restoring wild-type genetics.

Daisy restoration drives hold clear appeal: they offer a route to full genetic reversibility, something other drive-based countermeasures do not achieve. However, their complexity raises practical challenges. Recoding essential genes such as ribosomal targets may incur high fitness costs, and the system would require precise coordination of multiple genetic elements, each of which must function reliably at the population level.

2.5. Anti-Drives

Anti-drive systems using anti-CRISPR (Acr) proteins offer a distinct approach to countering gene drives: instead of rewriting DNA, they block Cas9 itself. These proteins, originally discovered in bacteriophages, bind to Cas9 and prevent it from guiding or cutting DNA (Chennuri, 2022). Acting as molecular decoys, they emerged from the evolutionary arms race between viruses and CRISPR-equipped bacteria.

Among all countermeasure strategies discussed, Acr-based anti-drives are the most experimentally advanced. While most others remain theoretical or have been tested only in Drosophila, at least two Acr systems have shown functional performance in Anopheles mosquitoes. To my knowledge, this makes them the only secondary release strategy to date with laboratory evidence in a target disease vector.

The first study, Taxiarchi et al. (2021), introduced a transgenic mosquito line carrying the AcrIIA4 protein, which was expressed constitutively from a neutral genomic locus. When released into cages with a doublesex-targeting homing drive, the Acr line blocked Cas9 activity, halted the drive’s spread, and reversed the suppression phenotype. Female fertility returned and population numbers recovered (Figure 6). No spatiotemporal control was needed for the Acr to function effectively.

In contrast, D’Amato et al. (2024) tested a germline-restricted Acr construct introduced as a separate secondary release into populations already carrying an active gene drive. Conducted in Anopheles coluzzii, the experiment followed how the Acr line spread and acted across generations, aiming to suppress drive activity without impacting wild-type individuals. The study focused explicitly on a risk-mitigation scenario, simulating a response to an unintended or undesired release (Figure 7).

Despite using the same Acr family, the two designs differed in timing, release strategy and regulatory control. Together, they mark the first steps from theoretical models to operational countermeasures in a disease-relevant species.

Unlike neutralising, reversal or immunising drives, Acr-based anti-drives do not create additional genome edits. By interfering with protein function rather than DNA, they avoid recombination and reduce the chance of off-target effects (Chennuri, 2022; Taxiarchi, 2021). However, without linkage to a gene drive element, these constructs follow Mendelian inheritance, meaning their persistence depends on fitness costs and ecological dynamics.

Current anti-drive systems rely on heritable expression in engineered organisms. Still, there seems to be a growing interest in using Acrs as non-genetic interventions, for example, as directly deliverable molecules or through transient expression systems. Although this remains untested in insects, the idea seems conceptually appealing because it could enable reversible, genetic-free countermeasures.

3. Targeted inhibitors

While most gene drive countermeasures rely on secondary releases, other strategies aim to inhibit gene drive activity without altering the genome. These approaches focus on disabling the molecular machinery of CRISPR itself, typically the Cas9 protein, using engineered inhibitors that act at the RNA, protein or chemical level. Unlike secondary releases, targeted inhibitors do not spread through populations and do not require heritable modification of released organisms.

Much of the work in this area comes from the therapeutic field, where researchers have been developing tools to fine-tune or halt CRISPR activity in gene-editing treatments. As a result, this is a rapidly growing space, with ongoing efforts to discover small molecules, short RNAs and nanomaterials capable of blocking Cas9 function across different systems. While not all developments were designed with gene drive containment in mind, they offer insights and potential tools for biosafety applications.

In a notable example, LeMieux (2019) reported on a researcher who advocated for the use of fast-acting, low-cost chemical inhibitors as a means to counteract gene drives. While no such system has yet been built, the researcher expressed interest in developing simple, mass-producible devices to distribute these compounds where needed.

These inhibitors are still largely experimental, but they offer promising features for future biosafety tools: fine control over timing, reversibility and, in some cases, delivery without transgenic release.

3.1. RNA-based inhibitors

Some viruses and plasmids have evolved a subtler strategy to evade CRISPR: small RNA molecules that mimic components of the host’s own immune system. These RNA-based anti-CRISPRs, recently described as Racrs, represent a distinct mechanism for disabling CRISPR–Cas systems; one that does not rely on protein-protein interactions (Camara-Wilper, 2023).

Racrs (repeat-associated small RNAs) were identified in phages and plasmids. Bioinformatic analysis suggested they resembled CRISPR repeats, but their biological roles were unclear until recently. Camara-Wilper and colleagues demonstrated that many of these RNA sequences can functionally suppress a range of CRISPR–Cas subtypes. In functional assays, Racrs inhibited Cas activity by mimicking host sequences and interfering with CRISPR assembly or function.

The mechanistic details remain under investigation, but the observed inhibition and broad subtype targeting suggest that Racrs are a widespread and effective countermeasure in nature (Camara-Wilper, 2023).

From a biosafety and gene drive perspective, Racrs offer intriguing potential. They could be engineered or adapted to selectively inhibit Cas systems at the RNA level, possibly allowing reversible or transient control. Unlike proteins, these molecules might be easier to synthesise or modify, and they avoid some of the delivery and immunogenicity issues associated with larger molecules. However, their application in eukaryotic systems and their stability outside of bacterial environments remain open questions.

3.2. Nanopolymer-based inhibitors

A less conventional, but promising, approach to inhibiting gene drives involves biocompatible nanopolymers that can directly interfere with Cas9 activity. Originally explored as delivery platforms for CRISPR components, some of these polymers were unexpectedly found to inhibit Cas9function themselves, a discovery that adds a new layer to gene drive control research (Chepurna, 2025).

Chepurna and colleagues tested several nanopolymers, including polymalic acid (PMLA), polyglutamic acid (PGA) and polyaspartic acid (PLD), each conjugated with a trileucine (LLL) moiety. These modified polymers showed dose-dependent inhibition of Cas9 nuclease activity. At sufficient concentrations, they were capable of completely blocking gene editing in vitro.

The mechanism of inhibition is not yet fully elucidated, but the findings suggest that the polymers interfere with Cas9's structural conformation or access to target DNA. Unlike protein- or RNA-based inhibitors, these polymers do not rely on sequence recognition or direct mimicry of CRISPR components. Instead, they appear to act through physicochemical interference, potentially offering broader applicability across CRISPR systems.

3.3. Small molecule inhibitors

Small molecules offer a flexible and scalable way to modulate gene drive systems. These low molecular weight compounds can bind directly to the Cas9 protein, blocking its ability to form active complexes or cleave DNA. Their fast kinetics, reversibility and ease of synthesis make them attractive candidates for precision control of CRISPR-based gene editing in both research and applied settings.

Initial efforts to identify Cas9 inhibitors yielded compounds such as BRD0539 (Figure 7), which was shown to reversibly inhibit SpCas9 in biochemical and cell-based assays (Maji, 2019). These inhibitors were effective at controlling Cas9 activity in a dose- and time-dependent manner, offering fine-tuned temporal regulation. This property could be particularly valuable in gene drive containment scenarios, for example, pausing a drive’s spread to allow controlled breeding or delaying its action until a specific developmental window.

Building on these findings, newer screening platforms have identified more potent inhibitors. Lim et al. (2022) reported BRD7586 (Figure 8), discovered through a high-throughput assay. BRD7586 exhibited greater potency than previous compounds and represents a step toward generalisable small-molecule inhibitors for a wider range of CRISPR systems.

Lee (2022) further refined this class of inhibitors, culminating in the development of compound 85 (Figure 9), which showed a 35-fold improvement in potency compared to earlier hits. Structural studies confirmed that these compounds act by preventing the formation of the Cas9:gRNA complex, a mechanism that offers a predictable and targeted mode of inhibition.

From a biosafety perspective, small-molecule inhibitors bring several advantages. They are cell-permeable, proteolytically stable and non-immunogenic. They also avoid the risks associated with introducing new genetic material into the environment. However, practical challenges remain, especially around delivery, dose control in open systems and degradation under environmental conditions.

4. The road to reversibility

The field of gene drive countermeasures has advanced at a steady but determined pace. What once seemed like an abstract need for emergency brakes has given rise to a variety of tangible, testable strategies, from genetic releases that overwrite or suppress a drive, to molecules that stop CRISPR at the protein or RNA level. Today, we are no longer asking if gene drives can be countered, but how, when and with what level of confidence.

Secondary release systems offer some of the most visible and well-characterised solutions. Yet each of these approaches depends on complex dynamics (timing, spread, fitness costs) and none are guaranteed to succeed under real-world conditions. As Devos et al. (2022) and Xu et al. (2020) caution, the release of a second engineered organism into an already modified population can introduce further uncertainty. These are tools to mitigate risk, not erase it.

Anti-drive systems based on anti-CRISPR proteins have emerged as a promising alternative. They are mechanistically simple, do not rely on additional genome editing, and some of them have already prevented population collapse in controlled settings (Taxiarchi, 2021; D’Amato, 2024). Still, their effectiveness hinges on population structure, mating probabilities and ecological context, variables that are difficult to control outside the lab.

Meanwhile, targeted inhibitors represent a shift toward molecular containment: interventions that don’t alter the genome at all. While none are field-ready, they hold promise as tools for laboratory safety, confined field trials, or even emergency interventions. These inhibitors could someday, perhaps, be paired with sensor-triggered systems or localised delivery platforms, offering reversible, dose-dependent and even spatially controlled regulation of CRISPR-based drives.

What emerges from this landscape is not a single solution, but a toolbox; one that is growing, diversifying and still being tested. Developers are building countermeasures alongside gene drives, not as an afterthought but as part of the core design. Regulatory frameworks, too, will need to evolve, shifting from a binary logic of “safe or unsafe” to one that considers degrees of controllability.

In the end, the presence of countermeasures should not be seen as a licence to proceed without caution. But they do offer something that was missing not long ago: a sense that gene drive interventions, if pursued, might not be irreversible by definition. That is not the same as being safe. But it is, perhaps, the beginning of a safer path forward.

5. References

The studies cited in this version of the living literature review are visually summarised in Figure 10, which maps their citation links, publication dates, and thematic focus.

Ethan Bier (2022). Gene drives gaining speed. https://doi.org/10.1038/s41576-021-00386-0

Sarah Camara-Wilpert, David Mayo-Muñoz, Jakob Russel, Robert D. Fagerlund, Jonas S. Madsen, Peter C. Fineran, Søren J. Sørensen, Rafael Pinilla-Redondo (2023) Bacteriophages suppress CRISPR–Cas immunity using RNA-based anti-CRISPRs. https://doi.org/10.1038/s41586-023-06612-5

Pratima R. Chennuri, Zach N. Adelman, Kevin M. Myles (2022). Genetic approaches for controlling CRISPR-based autonomous homing gene drives. https://www.doi.org/10.3389/fbioe.2022.897231

Oksana Chepurna, Avradip Chatterjee, Yuanqing Li, Hong Ding, Ramachandran Murali, Keith L. Black, Tao Sun (2025). Nano-polymers as Cas9 inhibitors. https://doi.org/10.3390/polym17030417

Rocco D’Amato, Chrysanthi Taxiarchi, Marco Galardini, Alessandro Trusso, Roxana L. Minuz, Silvia Grilli, Alastair G. T. Somerville, Dammy Shittu, Ahmad S. Khalil, Roberto Galizi, Andrea Crisanti, Alekos Simoni, Ruth Müller (2024). Anti-CRISPR Anopheles mosquitoes inhibit gene drive spread under challenging behavioural conditions in large cages. https://doi.org/10.1038/s41467-024-44907-x

Yann Devos, John D. Mumford, Michael B. Bonsall, Debora C.M. Glandorf, Hector D. Quemada (2022). Risk management recommendations for environmental releases of gene drive modified insects. https://doi.org/10.1016/j.biotechadv.2021.107807

Bernd Giese, Arnim von Gleich, Johannes L. Frieß (2020). Alternative techniques and options for risk reduction of gene drives. https://doi.org/10.1007/978-3-030-38934-5_7

Sang-Woo Lee, Kim Tai Tran, Ruben Vazquez-Uribe, Charlotte Held Gotfredsen, Mads Hartvig Clausen, Blanca Lopez Mendez, Guillermo Montoya, Anders Bach, and Morten Otto Alexander Sommer (2022). Identification and optimization of novel small-molecule Cas9 inhibitors by cell-based high-throughput screening. https://doi.org/10.1021/acs.jmedchem.1c01834

Julianna LeMieux (2019). CRISPR-accelerated gene drives pump the brakes. https://www.doi.org/10.1089/crispr.2019.29069.jlm

Donghyun Lim, Qingxuan Zhou, Kurt J. Cox, Benjamin K. Law, Miseon Lee, Praveen Kokkonda, Vedagopuram Sreekanth, Rajaiah Pergu, Santosh K. Chaudhary, Soumyashree A. Gangopadhyay, Basudeb Maji, Sophia Lai, Yuka Amako, David B. Thompson, Hari K. K. Subramanian, Michael F. Mesleh, Vlado Dančík, Paul A. Clemons, Bridget K. Wagner, Christina M. Woo, George M. Church, Amit Choudhary (2022). A general approach to identify cell-permeable and synthetic anti-CRISPR small molecules. https://doi.org/10.1038/s41556-022-01005-8

Basudeb Maji, Soumyashree A. Gangopadhyay, Miseon Lee, Mengchao Shi, Peng Wu, Robert Heler, Beverly Mok, Donghyun Lim, Sachini U. Siriwardena, Bishwajit Paul, Vlado Dancık, Amedeo Vetere, Michael F. Mesleh, Luciano A. Marraffini, David R. Liu, Paul A. Clemons, Bridget K. Wagner, Amit Choudhary (2019). A high-throughput platform to identify small-molecule inhibitors of CRISPR-Cas9. https://doi.org/10.1016/j.cell.2019.04.009

John Min, Charleston Noble, Devora Najjar, Kevin M. Esvelt (2017). Daisy quorum drives for the genetic restoration of wild populations. https://doi.org/10.1101/115618

Nicolas O. Rode, Virginie Courtier-Orgogozo, Florence Débarre (2020). Can a population targeted by a CRISPR-based homing gene drive be rescued? https://doi.org/10.1534/g3.120.401484

Yusuf Amuda Tajudeen, Habeebullah Jayeola Oladipo, Iyiola Olatunji Oladunjoye, Muhammad Kamaldeen Oladipo, Hameedat Damilola Shittu, Imam-Fulani Abdulmumeen, Abdullateef Opeyemi Afolabi, Mona Said El- Sherbini (2023). Transforming malaria prevention and control: the prospects and challenges of gene drive technology for mosquito management. https://doi.org/10.1080/07853890.2024.2302504

Chrysanthi Taxiarchi, Andrea Beaghton, Nayomi Illansinhage Don, Kyros Kyrou, Matthew Gribble, Dammy Shittu, Scott P. Collins, Chase L. Beisel, Roberto Galizi, Andrea Crisanti (2021). A genetically encoded anti-CRISPR protein constrains gene drive spread and prevents population suppression. https://doi.org/10.1038/s41467-021-24214-5

Michael R. Vella, Christian E. Gunning, Alun L. Lloyd, Fred Gould (2017). Evaluating strategies for reversing CRISPR-Cas9 gene drives. https://www.doi.org/10.1038/s41598-017-10633-2

Xiang-Ru Shannon Xu, Emily A. Bulger, Valentino M. Gantz, Carissa Klanseck, Stephanie R. Heimler, Ankush Auradkar, Jared B. Bennett, Lauren Ashley Miller, Sarah Leahy, Sara Sanz Juste, Anna Buchman, Omar S. Akbari, John M. Marshall, Ethan Bier (2020). Active genetic neutralizing elements for halting or deleting gene drives. https://doi.org/10.1016/j.molcel.2020.09.003