Cells and viruses

Human foreskin fibroblast cells were obtained from the ATTC (#SCRC-1041) and cultured in DMEM (10-013-CV, Corning, Corning, NY, USA), supplemented with 10% FBS (Sigma-Aldrich, St-Louis, MO, USA) and 100 μm/L penicillin/streptomycin (Corning). Cells were regularly tested negative for mycoplasma and used between passages 3 and 13.

hCMV TB40/E-Bac4¹⁷ and Towne-eGFP (T-BACwt)¹⁸ were kindly provided by Edward Mocarski (Emory University, USA). To prepare viral stocks, cells were infected at low MOI (0.001–0.01) and kept in culture until 100% cytopathic effect was observed, usually after 10–15 days. Cells were then scraped out of the plate and centrifuged together with the supernatant (12,000×g, 1 h, 4 °C), resuspended in medium containing 5% milk, and sonicated to release cell-bound virions. Viral titers were assessed by plaque assay. Except when otherwise specified, subsequent infections were performed for 1 h at a MOI = 0.1, before replacing inoculum with fresh medium. Susceptibility to IFN-γ was assayed by virus growth in the presence of human recombinant IFN-γ (R&D, Minneapolis, MN, USA) after preincubation with IFN-γ for 2 h before infection.

Viral titers were assayed by plaque assay with 10-fold serial dilutions. 24-well plates were inoculated for 1 h and overlaid with 0.25% agarose. After 7–10 days, eGFP or mCherry fluorescent plaques were manually counted using an inverted microscope. Every viral plaque was analyzed on both green and red channel. 5–100 plaques were counted per well, and each data-point was the average of 3–4 technical replicates (i.e., 3–4 different wells).

Images of fluorescent viral plaques were acquired with a Nikon Eclipse Ti2 inverted microscope and Nikon acquisition software (NIS-Element AR 3.0). Channels for GFP and mCherry were merged and adjusted for contrast and exposure with ImageJ (v2.1.0).

Coinfection experiments were performed by coinfecting with wildtype Towne-eGFP and gene drive viruses for 1 h, with a total MOI of 0.1–0.2. For time-course experiments over multiple weeks, supernatants were used to inoculate fresh cells for 1 h before changing media.

Cloning and generation of gene drive viruses

A donor plasmid containing the gene drive cassette against UL23 (GD-mCherry) between homology arms was generated by serial modifications of pX330, a codon-optimized SpCas9 (from Streptococus pyogenes) and chimeric gRNA expression plasmid developed by the Zhang lab⁵⁰ (Addgene #42230). All modifications were carried out by Gibson cloning (NEB, Ipswich, MA, USA), using PCR products from other plasmids or synthesized DNA fragments (GeneArt™ String™ fragments, ThermoFisher, USA). Briefly, a fragment with a SV40 polyA terminator, a SV40 promoter and an mCherry fluorescent reporter was inserted between SpCas9 and betaGlobin polyA signal. The PciI-AgeI fragment upstream of SpCas9 was removed and replaced by UL23 left homology arm (amplified from TB40/E-bac4). A fragment with UL23-5′ gRNA under a U6 promoter and UL23 right homology arm was finally inserted downstream of betaGlobin polyA signal between the NotI and Xmai restriction sites.

GD-∆Cas9 donor construct was subsequently generated by removing SpCas9 by digestion and ligation. A donor construct to insert a SV40-driven mCherry reporter into the BAC cassette of hCMV TB40/E-bac4 was built similarly.

To build gene drive viruses, 1.5 million fibroblast cells were transfected with the homologous recombination donor plasmid and a helper plasmid (Addgene #64221)⁵¹. Transfection was performed by Nucleofection (Kit V4XP-2024, Lonza, Basel, Switzerland). 48 h after transfection, cells were infected for 1 h with hCMV TB40/E-bac4 at a low MOI (<0.1). After 7–10 days, viral plaques of mCherry-expressing cells were observed, suggesting successful integration of the gene drive sequence by homologous recombination (Supplementary Fig. 1a). mCherry-expressing viral plaques were isolated and purified by several rounds of serial dilutions and plaque purification. Purity and absence of unmodified TB40/E viruses were assayed by PCR after DNA extraction (DNeasy kit, Quiagen). PCR and Sanger sequencing across homology arms and cut sites confirmed that mCherry-expressing viruses contained the full gene drive sequence (Supplementary Fig. 1b). GeneRuler 1 kb DNA ladder (ThermoFisher) was used in agarose gels as a size marker. The ladder is shown in the gel images, with the stronger band corresponding to 500 bp.

Viral stocks were produced as specified above and titered by plaque assay.

Deconvolution of Sanger sequencing in Supplementary Fig. 7 was performed using Synthego ICE online tools (https://ice.synthego.com).

HIRT DNA extraction and analysis of recombinant BAC clones

hCMV episomal DNA was recovered from the whole population of coinfected cells by HIRT DNA extraction¹⁹ 48 h after infection at a high MOI. Infected cells (grown in 1–2 T175 plates) were scraped-off, washed in PBS and resuspended in HIRT resuspension buffer (10 mM Tris–HCl pH 8.0, 10 mM EDTA in 100 μL for 1 million cells). An equal volume of HIRT lysis buffer (10 mM Tris–HCl pH 8.0, 10 mM EDTA, 1.2% SDS) was added and gently mixed. NaCl was added (1 M final concentration) before incubating overnight at 4 °C. Supernatant was collected after centrifugation (15,000×g, 20 min, 4 °C). Contaminating RNA was removed with RNAse A and hCMV DNA purified by double phenol–chloroform–isoamyl alcohol (25:24:1) extraction. DNA was precipitated with two volumes of pure ice-cold ethanol, washed with 70% ethanol, dried and resuspended in 10 mM Tris–HCl pH 8.0.

2–3 μL of recovered DNA was electroporated into NEB® 10-beta electrocompetent cells (C320K, NEB) and plated on chloramphenicol LB plates. BAC DNA were purified using ZR BAC DNA Miniprep Kit (Zymo Research, Irvine, CA, USA).

Presence of mCherry or GFP on recombinant BAC clones was confirmed by PCR. Homology arms of mCherry-eGFP clones were analyzed by PCR and Sanger sequencing using tools available on Benchling.com. Primers are given in Supplementary Table 1.

Virion purification and Oxford nanopore sequencing

Twelve large flasks were infected with recombinant viruses and cultured until the monolayer reached 100% cytopathic effect. Cells debris was pelleted away by centrifugation (20 min, 500×g, 4 °C) and supernatants were recovered. Virions present in the supernatant were pelleted by ultracentrifugation (22 krpm, 90 min, 4 °C, Beckman-Coulter rotor SW28) on a 5-mL cushion of 30% sucrose. Supernatants were discarded, and the pellets containing virions were resuspended in PBS (1 h at room temperature) and pooled in a final volume of 500 μL.

DNase I (20U, 30 min, 37 °C, NEB M0303S) and RNAse A (100 μg, 15 min, ThermoFisher EN0531) treatment removed contaminating human nucleic acids unprotected by the virus envelope and capsid. Viral envelopes were lysed and DNase I was inactivated by adding 5× lysis buffer for 10 min (0.5 M Tris pH 8, 25 mM EDTA, 1% SDS, 1 M NaCl. Incubation for 1 h at 55 °C with 8U of proteinase K (NEB P8107S) finally disrupted viral capsids. To recover full-length hCMV genomes, subsequent pipetting and centrifugation steps were performed extremely carefully with wide-bore pipet tips. hCMV DNA was purified by double phenol–chloroform–isoamyl alcohol (25:24:1) extraction and centrifugation (4000×g, 3 min). DNA was precipitated for 1 h at −80 °C with 1/20 volume of 5 M NaCl and 1.5 volume of cold ethanol, centrifugated (6800×g, 30 min, 4 °C), washed with 70% ethanol, dried and resuspended in 10 mM Tris–HCl pH 8.0 for 24 h at 4 °C without pipetting. DNA was quantified by Nanodrop.

Libraries were prepared using SQK-LSK109 ligation sequencing kit from Oxford Nanopore Technology (Oxford, UK), without any fragmentation steps, using wide-bore pipet tips and careful pipetting steps to minimize DNA shearing. Libraries for the first biological replicate (two technical replicates) were prepared following the manufacturer instructions, using 2 μg of starting material. Lambda DNA control was added in the first technical replicate. In an attempt to maximize read length, the second biological replicate was prepared with 15 μg of DNA, omitting the first AMPure XP bead clean-up. Sequencing was performed on two FLO-MIN106-R9 Flow Cells on a MinION Mk1B device following manufacturer instructions.

Oxford Nanopore sequencing analysis

Nanopore FAST5 raw data was converted into FASTQ files using Albacore v2.3.3 basecalling, and runs statistics were obtained using Nanoplot v1.0.0 (https://github.com/wdecoster/NanoPlot)⁵². Adaptors were removed with Porechop v0.2.4 (https://github.com/rrwick/Porechop). Reads with quality Q > 6 were filtered using Nanofilt v2.5.0 (https://github.com/wdecoster/nanofilt)⁵² and Lambda DNA reads excluded with NanoLyse v1.0.0 (https://github.com/wdecoster/nanolyse)⁵². Technical replicates were merged for the rest of the analysis. Read lengths were filtered with Nanofilt.

Reference sequences for Towne-eGFP (GenBank KF493877) and TB40/E-Bac4 (GenBank EF999921) were downloaded. We first inserted the gene drive cassette into Towne-eGFP and TB40/E reference sequences. Reads were then mapped on a composite human hg38-Towne genome using Minimap2 v2.14 (https://github.com/lh3/minimap2)⁵³ and mapping statistics (Supplementary Fig. 4) were obtained with samtools v1.10⁵⁴ after filtering of secondary and supplementary reads (samtools -F 2048 -F 256).

To create of map of SNPs between Towne and TB40 strains, we created a FASTA file composed of multiple copies of the two genomes. Using Minimap2, this fasta file was mapped onto Towne-eGFP reference. SNPs were then called using bcftools mpileup and bcftools call, generating a BCF file with the SNPs coordinates:

minimap2 -a Ref_Towne.fasta multiple_Towne-TB40.fasta -cs -N 0 -2 | samtools view -S -b| samtools sort -o mapped.bam

bcftools mpileup -Ou -f Ref_Towne.fasta mapped.bam | bcftools call -mv -Ob -o map_snp.bcf

hCMV genome exists in four different configurations depending on the respective orientation of UL and US segments. A Towne-eGFP composite genome composed of Towne sequences in the four configurations was finally created, and a complete map of SNPs in the four configurations was also generated.

After comparing different mappers, we mapped sequencing reads (length > 10 kb) on the composite Towne genome using Graphmap v0.3.0 (https://github.com/isovic/graphmap)⁵⁵, keeping only the best mapped reads (default, returning uniquely mapped read) or allowing multiple read mapping (option -Z). Proportion of variants for each SNP coordinate was then calculated using Nanopolish (https://github.com/jts/nanopolish)⁵⁶. Nanopolish was run successively on the four subgenomes as follows, using the SNP map generated above (example for Towne in the SS: Sense–Sense configuration):

nanopolish variants—reads data.fastq.gz–bam Mapped_with_graphmap.bam–genome Reference_Towne_4config.fasta -p 2 -w Towne_config_SS:1-239862 -m 0.15 -x 2000 -c map_snp_4config.vcf -o Call_SS.vcf

Read coverage and Support fractions were then extracted from the VCF files and plots were generated using R. Of note, for multiple mapping reads, we had to artificially increase the mapQ fields of SAM/BAM files by 20, because Nanopolish automatically discard such reads.

Finally, to reconstruct the recombination history of individual genomes, the longest reads (>200 kb) were mapped on the composite Towne genomes using Graphmap and visualized with IGV⁵⁷. The recombination map of each individual read was then reconstructed manually using the SNP map.

The reference genome of GD-mCherry virus was inferred from reads that contained no Towne fragments. Importantly, we detected two large deletions in GD-mCherry not present in the original TB40E-bac4 genome: One 5.6-kb deletion in UL ranging from RL12 to UL8 genes and, therefore, including the frequently mutated RL13 gene, and a second 4.6-kb deletion in US from US15 to US19.

Statistics and reproducibility

Plaque assay data did not appear to satisfy the normality condition required for parametric tests. Due to small sample sizes, normality and lognormality tests could however not be performed. We therefore chose to run in parallel both parametric tests on log-transformed data, and non-parametric test on untransformed data, and reported results when both type of tests gave significant values. For groups analysis, we performed two-way ANOVA with Sidak’s multiple comparison test on log-transformed data, and Kruskal–Wallis test with Dunn’s multiple comparison test on untransformed data. Analysis were run using GraphPad Prism version 8.1.1 for macOS (GraphPad Software, San Diego, CA, USA, www.graphpad.com).

Examples of plaques shown in Fig. 3a and Supplementary Fig. 2 are representative of every plaque assay (n > 100) performed in the study.

Numerical simulations

Numerical simulations of viral gene drive were computed using a simplified viral replication model. Shortly, in each viral generation, N virtual cells were randomly infected and coinfected by N*MOI viruses, producing a new generation of viruses. In this new generation, wildtype viruses coinfected with drive viruses were converted to new gene-drive viruses. Gene drive virus replicate with a fitness cost f, and the coinfection rate is calculated from the MOI assuming a Poisson distribution. The code and a more thorough description are available at https://github.com/mariuswalter/ViralDrive.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.