Within-host diversity of SARS-CoV-2 in cats is limited

Recently, members of our team inoculated three domestic specific-pathogen free cats with a second-passage SARS-CoV-2 human isolate from Tokyo (hCoV-19/Japan/UT-NCGM02/2020) [33]. Each index cat was co-housed with a contact cat beginning on day 1 post-inoculation (DPI). No new cat infections were performed for this study. Nasal swabs were collected daily up to 10 days post-inoculation, Fig 1. Viral RNA burden is plotted in S1A Fig and infectious viral titers are shown in S1B Fig.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Experimental timeline.

Schematic representing the sampling timeline for the three transmission pairs. Index cats were inoculated on day 0 with 5.2e5 PFU of a human isolate (hCoV-19/Japan/UT-NCGM02/2020) and were co-housed with a naive cat starting on day 1. Within each transmission pair, the top row of circles represents the index cat and the bottom row represents the contact cat. Open circles represent days on which there was no detectable infectious virus as indicated by plaque assay, and closed circles highlight days when live virus was recovered. Circles with a red outline indicate timepoints which were used in the beta-binomial estimate to calculate transmission bottleneck sizes.

https://doi.org/10.1371/journal.ppat.1009373.g001

Using conservative frequency thresholds previously established for tiled-amplicon sequencing, we called within-host variants (both intrahost single-nucleotide variants “iSNVs” and short insertions and deletions “indels”) throughout the genome against the inoculum SARS-CoV-2 reference (Genbank: MW219695.1) [34,35]. Variants were required to be present in technical replicates at ≥3% and ≤97% of sequencing reads [36] (all within-host variants detected at >97% frequency were assumed to be fixed; see Methods for details). iSNVs were detected at least once at 38 different genome sites. Of the 38 unique variants, 14 are synonymous changes, 23 are nonsynonymous changes, and one occurs in an intergenic region; this distribution is broadly similar to recent reports of SARS-CoV-2 variation in infected humans [30]. Similarly, we detected indels occurring at 11 different genome sites across all animals and timepoints. We identified 6–19 distinct variants per cat, of which 4–7 were observed on two or more days over the course of the infection within each cat (S2 Fig). All variants (iSNVs and indels) are plotted by genome location and frequency in Fig 2A.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Within-host diversity of SARS-CoV-2 viruses in domestic cats.

A) Plot representing all variants (iSNVs and indels) detected in any cat at any timepoint. Variant frequencies are plotted by genome location and are colored by gene. Circles represent synonymous iSNVs, squares represent nonsynonymous iSNVs, and stars represent indels. B) iSNV frequency spectrums with error bars showing standard deviation for index cats plotted against a “neutral model” (light gray bars) which assumes a constant population size and the absence of selection.

https://doi.org/10.1371/journal.ppat.1009373.g002

Genetic drift and purifying selection shape within-host diversity

To probe the evolutionary pressures shaping SARS-CoV-2 viruses within hosts, we first evaluated the proportion of variants shared between cats. Eighty-nine percent of variants (34 of 38 iSNVs and 8 of 11 indels) were found in a single cat (42/49), 8% of variants were found in 2–5 cats (4/49), and the remaining 6% of variants were found in all 6 cats (3/49).

Purifying selection, which acts to purge deleterious mutations from a population, is known to result in an excess of low-frequency variants. In contrast, positive selection results in the accumulation of intermediate- and high-frequency variation [37]. Especially in the setting of an acute viral infection, exponential population growth is also expected to result in an excess of low-frequency variants [38]. To determine the type of evolutionary pressure acting on SARS-CoV-2 in cats, we plotted these distributions against a simple “neutral model” (light grey bars in Fig 2B), which assumes a constant population size and the absence of selection [37]. This model predicted that ~43% of polymorphisms would fall in the 3–10% frequency bin, ~25% into the 10–20% bin, ~14% into the 20–30% bin, ~10% into the 30–40% bin, and ~8% into the 40–50% bin. The frequency distribution of variants detected in each index cat across all available timepoints did not differ significantly from this “neutral” expectation (p = 0.265, p = 0.052, p = 0.160, respectively; Mann Whitney U test).

Next we compared nonsynonymous (πN) and synonymous (πS) pairwise nucleotide diversity to further evaluate the evolutionary forces shaping viral populations in index and contact animals [39]. Broadly speaking, excess nonsynonymous polymorphism (πN/πS > 1) points toward diversifying or positive selection while excess synonymous polymorphism (πN/πS < 1) indicates purifying selection. When πN / πS is approximately 1, genetic drift, i.e., stochastic changes in the frequency of viral genotypes over time, can be an important force shaping genetic diversity. We observe that πS exceeds or is approximately equal to πN in most genes, although there is substantial variation among genes and cats (S1 Table, S10 and S11 Figs). πS is significantly higher than πN in all 3 index cats in Spike (p = 0.005, p = 0.004, p = 0.019, unpaired t-test) and ORF1ab (p = 2.11e-05, p = 1.84e-06, p = 1.99e-06, unpaired t-test) and in index cats 2 and 3 in ORF8 (p = 0.03, p = 0.04, unpaired t-test). πS and πN are not significantly different in at least one index cat in ORF3a, envelope, and nucleocapsid. There was not enough genetic variation to measure nucleotide diversity in the remaining four genes (S1 Table). Taken together, these results suggest longitudinal genetic variation within feline hosts is principally shaped by genetic drift with purifying selection acting on individual genes, particularly ORF1ab and Spike.

Longitudinal sampling reveals few consensus-level changes within hosts

The consensus sequences recovered from all three index cats on the first day post-inoculation was identical to the inoculum or “stock” virus. This consensus sequence remained largely unchanged throughout infection in all index cats with the notable exception of two variants: H655Y in Spike (nucleotide site 23,525) and a synonymous change at amino acid position 67 in envelope (nucleotide site 26,445; S67S), which arose rapidly in all 3 index cats and rose to consensus levels (≥50% frequency) at various timepoints throughout infection in all index cats. Neither of these iSNVs were detected above 3% frequency in the inoculum, but when we mined all sequencing reads, S H655Y and E S67S could be detected at 0.85% and 0.34%, respectively. S H655Y was the consensus sequence on days 2–5 and days 7–8 in index cat 1, as well as on days 4 and 8 in index cat 2, and remained detectable above our 3% variant threshold throughout infection (Fig 3). Similarly, envelope S67S (E S67S) was the consensus sequence on day 8 in index cat 1 and day 1 in index cat 2. S H655Y and E S67S were detectable on days 1–7 in cat 3 but stayed below consensus level.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Frequency of iSNVs over time in each index and contact cat.

The frequency of iSNVs discussed in the results over time in all six cats are shown. All iSNVs over time are shown in S2 Fig and all indels over time are shown in S3 Fig. Each variant is colored by gene location. Nonsynonymous variants are plotted with solid lines and synonymous variants are plotted with dashed lines. Variants detected in index cats are denoted with squares and variants detected in contact cats are denoted with circles. Timepoints with viral loads too low to yield high quality sequences are shown by the gaps in data, but iSNVs are connected across these gaps using light lines for readability (i.e. cat 1 day 9). The dotted line at 50% frequency represents the consensus threshold.

https://doi.org/10.1371/journal.ppat.1009373.g003

Interestingly, S H655Y and E S67S became fixed together following transmission in two transmission pairs (contact cats 4 and 6) and were lost together during transmission to contact animal 5. In cat 5, however, two different variants in ORF1ab, G1756G and L3606F, became fixed after transmission. ORF1ab G1756G was not detected above 3% and L3606F was found at 17.2% in the day 5 sample from the index cat 2 (the cat transmitting to cat 5); it was not found in the inoculum at any detectable frequency. The categorical loss or fixation of these variants immediately following transmission, and in particular the fixation following transmission of a variant that was undetectable before, are highly suggestive of a narrow bottleneck [40].

In addition, a synonymous variant in an alanine codon at amino acid position 1,222 in Spike (nucleotide site 25,174) was found at >50% frequencies on days 4 and 8 in index cat 3, but was not detected above 3% on any other days. All iSNVs over time are shown in S2 Fig and all indels over time are shown in S3 Fig. These within-host analyses show that genetic drift appears to play a prominent role in shaping low-frequency genetic variation within hosts.

SARS-CoV-2 transmission in domestic cats is defined by a narrow transmission bottleneck

To estimate the size of SARS-CoV-2 transmission bottlenecks, we investigated the amount of genetic diversity lost following transmission in cats. We observed a reduction in the cumulative number of variants detected in each contact cat compared to its index: 7 fewer variants in cat 4 (n = 9) compared to cat 1 (n = 16), 9 fewer in cat 5 (n = 10) than cat 2 (n = 19), and 10 fewer in cat 6 (n = 16) than cat 3 (n = 6). Likewise, the frequency distribution of variants in all three contact cats following transmission differed from the distribution of variants in all three index cats prior to transmission (p-value = 0.052, Mann Whitney U test). Following transmission, variant frequencies became more bimodally distributed than those observed in index cats, i.e., in contacts, most variants were either very low-frequency or fixed (S2 Fig).

To quantitatively investigate the stringency of each transmission event, we compared the genetic composition of viral populations immediately before and after viral transmission. We chose to use the first timepoint when infectious virus was recovered in the contact cat coupled with the timepoint immediately preceding this day in the index cat, as has been done previously [17]. We used days 2 (index) and 3 (contact) in pair 1, days 5 and 6 in pair 2, and days 4 and 5 in pair 3 (these sampling days are outlined in red in Fig 1). We applied the beta-binomial sampling method developed by Sobel-Leonard et al. to compare the shared set of variants (≥3%, ≤97%) in the pre/post-transmission timepoints for each pair [21]. Maximum-likelihood estimates determined that a mean effective bottleneck size of 5 (99% CI: 1–10), 3 (99% CI: 1–7), and 2 (99% CI: 1–3) best described each of the three cat transmission events evaluated here (Fig 4). This is in line with previous estimates for other respiratory viruses, including airborne transmission of seasonal influenza viruses in humans [40]. It is important to note, however, that the cat transmission pairs evaluated here shared physical enclosure spaces so the route of transmission could be airborne, direct contact, fomite, or a combination of these. Additionally, it has been shown that the route of influenza transmission can directly impact the size of the transmission bottleneck; for example, in one study airborne transmission of influenza viruses resulted in a narrow bottleneck, whereas contact transmission resulted in a wider bottleneck [16].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. SARS-CoV-2 transmission is defined by a narrow bottleneck.

Variant frequencies in the index cats (x-axis) compared with frequencies of the same variants in the corresponding contact cats (y-axis) that were used in the beta-binomial estimate are shown on the left. Estimates of SARS-CoV-2 transmission bottleneck with 99% confidence intervals shown on the right.

https://doi.org/10.1371/journal.ppat.1009373.g004

Ethics statement

No animal experiments were specifically performed for this study. We used residual nasal swabs collected from domestic cats as part of a previously published study [33]. Animal studies were approved prior to the start of the study by the Institutional Animal Care and Use Committee and performed in accordance with the Animal Care and Use Committee guidelines at the University of Wisconsin-Madison.

Domestic cat experiments

No animal experiments were specifically performed for this study. We used residual nasal swabs collected from domestic cats as part of a previously published study [33]. Animals used in this study were specific-pathogen-free animals from a research colony maintained at the University of Wisconsin-Madison and were negative for feline coronavirus. As previously described by Halfmann et al, domestic cats were housed in 0.56 m x 0.81 m x 1.07 m cages in a laboratory with 65% humidity at 23°C, and with at least 15.2 air exchanges per hour. Weight and body temperature (through implanted transponders) were measured daily (days 1–14). Under ketamine and dexdomitor anesthesia, three cats were inoculated with 5.2 x 105 plaque-forming units (PFU of SARS-CoV-2 given by a combination of inoculation routes for every animal (nasal [100 μl per nare], tracheal [500 μl], oral [500 μl], and ocular [50 μl per eye]). To reverse the effects of the anesthesia, antisedan was administered to the animals after completion of the inoculation. Nasal swabs were collected daily during the study (days 1–10).

Nucleic acid extraction

For each sample, approximately 140 μL of viral transport medium was passed through a 0.22 μm filter (Dot Scientific, Burton, MI, USA). Total nucleic acid was extracted using the Qiagen QIAamp Viral RNA Mini Kit (Qiagen, Hilden, Germany), substituting carrier RNA with linear polyacrylamide (Invitrogen, Carlsbad, CA, USA) and eluting in 30 μL of nuclease-free H₂O.

Complementary DNA (cDNA) generation

Complementary DNA (cDNA) was synthesized using a modified ARTIC Network approach [34,35]. Briefly, RNA was reverse transcribed with SuperScript IV Reverse Transcriptase (Invitrogen, Carlsbad, CA, USA) using random hexamers and dNTPs. Reaction conditions were as follows: 1 μL of random hexamers and 1 μL of dNTPs were added to 11 μL of sample RNA, heated to 65°C for 5 minutes, then cooled to 4°C for 1 minute. Then 7 μL of a master mix (4 μL 5x RT buffer, 1 μL 0.1M DTT, 1μL RNaseOUT RNase Inhibitor, and 1 μL SSIV RT) was added and incubated at 42°C for 10 minutes, 70°C for 10 minutes, and then 4°C for 1 minute.

Multiplex PCR for SARS-CoV-2 genomes

A SARS-CoV-2-specific multiplex PCR for Nanopore sequencing was performed, similar to amplicon-based approaches as previously described [34,35]. In short, primers for 96 overlapping amplicons spanning the entire genome with amplicon lengths of 500bp and overlapping by 75 to 100bp between the different amplicons were used to generate cDNA. Primers used in this manuscript were designed by ARTIC Network and are shown in S3 Table. cDNA (2.5 μL) was amplified in two multiplexed PCR reactions using Q5 Hot-Start DNA High-fidelity Polymerase (New England Biolabs, Ipswich, MA, USA) using the following cycling conditions; 98°C for 30 seconds, followed by 25 cycles of 98°C for 15 seconds and 65°C for 5 minutes, followed by an indefinite hold at 4°C [34,35]. Following amplification, samples were pooled together before TrueSeq Illumina library prep.

TrueSeq Illumina library prep and sequencing

Amplified cDNA was purified using a 1:1 concentration of AMPure XP beads (Beckman Coulter, Brea, CA, USA) and eluted in 30 μL of water. PCR products were quantified using Qubit dsDNA high-sensitivity kit (Invitrogen, USA) and were diluted to a final concentration of 2.5 ng/μl (150 ng in 50 μl volume). Each sample was then made compatible with deep sequencing using the TruSeq sample preparation kit (Illumina, USA). Specifically, each sample was enzymatically end repaired. Samples were purified using two consecutive AMPure bead cleanups (0.6x and 0.8x) and were quantified once more using Qubit dsDNA high-sensitivity kit (Invitrogen, USA). A non-templated nucleotide was attached to the 3′ ends of each sample, followed by adaptor ligation. Samples were again purified using an AMPure bead cleanup (1x) and eluted in 25 μL of resuspension buffer. Lastly, samples were amplified using 8 PCR cycles, cleaned with a 1:1 bead clean-up, and eluted in 30 μL of RSB. The average sample fragment length and purity was determined using the Agilent High Sensitivity DNA kit and the Agilent 2100 Bioanalyzer (Agilent, Santa Clara, CA). After passing quality control measures, samples were pooled equimolarly to a final concentration of 4 nM, and 5 μl of each 4 nM pool was denatured in 5 μl of 0.2 N NaOH for 5 min. Sequencing pools were denatured to a final concentration of 10 pM with a PhiX-derived control library accounting for 1% of total DNA and was loaded onto a 500-cycle v2 flow cell. Average quality metrics were recorded, reads were demultiplexed, and FASTQ files were generated on Illumina’s BaseSpace platform.

Processing of the raw sequence data, mapping, and variant calling

Raw FASTQ files were analyzed using a workflow called “SARSquencer”. Briefly, reads are paired and merged using BBMerge (https://jgi.doe.gov/data-and-tools/bbtools/bb-tools-user-guide/bbmerge-guide/) and mapped to the reference (MW219695.1) using BBMap (https://jgi.doe.gov/data-and-tools/bbtools/bb-tools-user-guide/bbmap-guide/). Mapped reads were imported into Geneious (https://www.geneious.com/) for visual inspection. Read coverage for index cat samples is plotted in S6 Fig and for contact samples in S7 Fig. Variants were called using callvariants.sh (contained within BBMap) and annotated using SnpEff (https://pcingola.github.io/SnpEff/). The complete “SARSquencer” pipeline is available in the GitHub accompanying this manuscript in ‘code/SARSquencer’ as well as in a separate GitHub repository– https://github.com/gagekmoreno/SARS_CoV-2_Zequencer. BBMap’s output VCF files were cleaned using custom Python scripts, which can be found in the GitHub accompanying this manuscript (https://github.com/katarinabraun/SARSCoV2_transmission_in_domestic_cats) [60]. Variants were called at ≥0.01% in reads that were ≥100 bp in length and supported by a minimum of 10 reads. Only variants at ≥3% frequency in both technical replicates were used for downstream analysis. Variant concordance across technical replicates is plotted in S8 Fig for index cats and S9 Fig for contact cats. In addition, all variants occurring in ARTIC v3 primer-binding sites were discarded before proceeding with downstream analysis.

Quantification of SARS-CoV-2 vRNA

Plaque forming unit analysis was performed on all nasal swabs as published in Halfmann et al. 2019 [33]. Viral load analysis was performed on all of the nasal swab samples described above after they arrived in our laboratory. RNA was isolated using the Viral Total Nucleic Acid kit for the Maxwell RSC instrument (Promega, Madison, WI) following the manufacturer’s instructions. Viral load quantification was performed using a sensitive qRT-PCR assay developed by the CDC to detect SARS-CoV-2 (specifically the N1 assay) and commercially available from IDT (Coralville, IA). The assay was run on a LightCycler 96 or LC480 instrument (Roche, Indianapolis, IN) using the Taqman Fast Virus 1-stepMaster Mix enzyme (Thermo Fisher, Waltham, MA). The limit of detection of this assay is estimated to be 200 genome equivalents/ml saliva or swab fluid. To determine the viral load, samples were interpolated onto a standard curve consisting of serial 10-fold dilutions of in vitro transcribed SARS-CoV-2 N gene RNA.

Pairwise nucleotide diversity calculations

Nucleotide diversity was calculated using π summary statistics (S2 Table). π quantifies the average number of pairwise differences per nucleotide site among a set of sequences and was calculated per gene using SNPGenie (https://github.com/chasewnelson/SNPgenie) [62]. SNPGenie adapts the Nei and Gojobori method of estimating nucleotide diversity (π), and its synonymous (πS) and nonsynonymous (πN) partitions from next-generation sequencing data [63]. When πN = πS, this indicates neutral evolution or genetic drift, with neither strong purifying nor positive selection playing a large role in the evolution of the viral population. πN < πS indicates purifying selection is acting to remove deleterious mutations, and πN > πS shows positive or diversifying selection acting on nonsynonymous variation [64]. We tested the null hypothesis that πN = πS within each gene using an unpaired t-test (S1 Table). The code to replicate these results can be found in the ‘diversity_estimates.ipynb’ Jupyter Notebook in the ‘code’ directory of the GitHub repository [65].

SNP Frequency Spectrum calculations

To generate SNP Frequency Spectrums (SFS), we binned all variants detected across timepoints within each index cat into six bins– 3–10%, 10–20%, 20–30%, 30–40%, 40–50%, 50–60%. We plotted the counts of variants falling into each frequency bin using Matplotlib 3.3.2 (https://matplotlib.org). We used code written by Dr. Louise Moncla to generate the distribution of SNPs for a given population assuming no selection or change in population size, which is expected to follow a 1/x distribution [37]. The code to replicate this can be found in the GitHub accompanying this manuscript, specifically in the ‘code/SFS.ipynb’ Jupyter Notebook. This model predicts 42.8% of variants will fall within the 3–10% frequency range, 24.6% will fall within the 10–20% frequency range, 14.4% of variants will fall within the 20–30% frequency range, 10.2% of variants will fall within the 30–40% frequency range, and 7.9% of variants will fall within the 40–50% frequency range. We used a Mann-Whitney U test to test the null hypothesis that the distribution of variant frequencies for each index cat was equal to the neutral distribution. The code to replicate these results can be found in the `SFS.ipynb`Jupyter Notebook in the `code`directory of the GitHub repository [65].

Focal Nextstrain build of S H655Y sequences

The focal H655Y build (S5 Fig) was prepared as described in Hodcroft et al. (2020), with different mutations targeted for the S:655 mutation [66]. Briefly: sequences with a mutation at nucleotide position 23525 (corresponding to a change at the 655 position in the spike glycoprotein) were selected from all available sequences on GISAID as of 29th December 2020. These sequences were included as the ’focal’ set for a Nextstrain phylogenetic analysis, to which ’context’ sequences were added, with the most genetically similar sequences given priority.

Code and data availability

Code to replicate analyses and re-create most figures is available at https://github.com/katarinabraun/SARSCoV2_transmission_in_domestic_cats [65]. Fig 1 was created by hand in Adobe Illustrator and S6 and S7 Figs were created using samtools command line tools, were visualized in JMP Pro 15 (https://www.jmp.com/en_in/software/new-release/new-in-jmp-and-jmp-pro.html), and were then edited for readability in Adobe Illustrator. Code to process sequencing data is available at https://github.com/gagekmoreno/SARS_CoV-2_Zequencer and dependencies are available through Docker [67]. Results were visualized using Matplotlib 3.3.2(https://matplotlib.org), Seaborn v0.10.0 (https://github.com/mwaskom/seaborn), and Baltic v0.1.0 (https://github.com/evogytis/baltic).