Detection and prevalence of SARS-CoV-2 co-infections during the Omicron variant circulation in France

Evaluation of 4 different primer sets to detect SARS-CoV-2 co-infections using WGS

To simulate co-infections, Delta (B.1.617.2) and Omicron (BA.1) viral isolates were mixed using different Delta:Omicron ratios: 0:100, 10:90, 20:80, 30:70, 40:60, 50:50, 60:40, 70:30, 80:20, 90:10, and 100:0. Resulting mixes had fixed viral loads (median = 4.2 log10 cp/ml; IQR = 0.4, Supplementary Data 1). Four sets of primers (Artic V4 and V4.1, Midnight V1 and V2) were used in duplicate on extracts to test the impact of PCR amplification prior to sequencing on co-infection characterization. All mixes were sequenced to 1 M paired-end reads, leading to SARS-CoV-2 genome covered > 98% with a median coverage of 2276X (IQR = 315X) (Supplementary Data 1).

The evaluation of the primer sets was performed using a previously published method based on a curated list of mutations specific to Delta and to Omicron derived from co-variants^5,6,7 (Supplementary Data 1 and Fig. 1). More than 90% of the Delta-specific mutations were found in all mixes with the 4 primer sets. In contrast, the detection rate of Omicron-specific mutations ranged from 26% in the 90:10 mix using Midnight primers (V1 and V2) to >77% for mixes with the expected frequency of Omicron above 30% (Fig. 1A and Supplementary Data 1). Lower detection rate of Omicron-specific mutations was associated with different primer bias preferentially amplifying Delta over Omicron (Suppl Results, Supplementary Data 2). Medians of covered allele frequency for the specific mutations were used to estimate viral frequency. Relations between measured and expected frequency were not linear (Fig. 1B). Over-estimation of Delta was observed for all mixes with all primer sets, and especially in mixes with expected frequency of Delta under 30% and sequenced with Midnight primer sets. Measured frequencies of Delta for the 10:90 mix were between 30–33% with Midnight V1 and V2, and 21–25% with Artic V4 and V4.1.

Eleven mixes of Delta: Omicron isolates with different proportions were extracted and sequenced in duplicate. A Detection rate of Delta- and Omicron-specific mutations. These detection rates are defined as the number of Delta- or Omicron-specific variants found in each sample (as minor or major allele) out of the total number of Delta- or Omicron-specific variants based on covariants.org. Horizontal line (Delta-specific mutation ratio) at 0.9 and vertical line (Omicron-specific mutation ratio) at 0.25 discriminate co-infections from pure isolates. B Frequency of the Delta variant was measured using the median of covered allele frequencies of Delta-specific mutations and compared with the expected relative frequency in each mix. Pure isolates (denoted as 100:0 for Delta and 0:100 for Omicron) are depicted with dots, and Delta: Omicron mixes with triangles. Colors represent the expected frequency of each mix using a red (Delta: 100:0) to blue (Omicron: 0:100) gradient. C Representation of variant allele frequency of Delta- and Omicron-specific mutations for the mix 20:80. Delta-specific mutations are depicted in red, Omicron-specific mutations in blue and shared mutations are in grey. Horizontal lines at 50% show which mutations are called in the consensus sequence based on the majority rule. Chimeric sequences with both Delta and Omicron mutations were detected for the four primer sets.

Importantly, consensus sequence calling based on majority rule resulted in artefactual chimeric Delta-Omicron sequences for several mixes and with different patterns depending on the primers used for amplification (Fig. S4). Sequences bearing both Delta- and Omicron-specific polymorphisms were found independently of the bioinformatic pipeline used to call the consensus sequence (Fig. S5). Omicron sequences bearing Delta-specific mutations were found in mixes with Delta expected frequency of 10–30%. The highest number of Delta-specific mutations was observed in the 20:80 mixes sequenced with Midnight primers, and in the 30:70 mixes sequenced with Artic primers (Fig. S5), which were the mixes with 50% measured frequency of Delta (Fig. 1B). Such sequences were observed with the four primer sets in all duplicates only for the 20:80 mix (Fig. 1C). With Artic primer sets, chimeric sequences were characterized by Omicron sequences bearing the S:L452R and M:I82T mutations, in relation with amplicon 76 and 89 bias (Suppl Results), respectively, and additional Delta-specific mutations with increasing Delta concentration. With Midnight primers, chimeric sequences were characterized by Omicron sequences with 3’ end of the genome belonging to Delta (starting from nt 27,874, in relation with amplicon 28 bias).

Altogether, the Artic V4.1 primers were the least biased for Delta/Omicron co-infection detection and relative frequency estimation, but all primer sets could lead to artefactual chimeric sequences, highlighting the importance of proper co-infection detection.

Novel bioinformatic algorithm to detect SARS-CoV-2 co-infections using WGS

Independent to this specific set of mutations, an agnostic approach was developed to detect co-infections regardless of the lineage present in the sample, as long as these lineages are genetically distinct (Fig. 2). This approach is based on the identification of a potential secondary lineage, after excluding variants shared with the main lineage. A secondary lineage is potentially identified when six or more specific mutations of this lineage are present as minor variants; this threshold was determined on composite criteria based on the mixes and samples as detailed in supplementary methods. Two ratios are calculated: the main lineage mutation ratio and the secondary lineage mutation ratio quantifying the fraction of present mutations among covered lineage-defining and lineage-specific mutations, respectively. Based on this co-infection detection script, co-infection was successfully identified in all mixes, except for the pure Delta (100:0) and Omicron (0:100) isolates, for which only Delta (lineage B.1.617.2) and Omicron (lineage BA.1) were identified as the main lineages, respectively. B.1.617.2 and BA.1 were identified as the main and secondary lineages, respectively, in all mixes with expected frequency of Delta above 40%, independent of the primer sets (Fig. 2 and Supplementary Data 1). Main lineage mutation ratios were above 0.9 for all mixes (Fig. 2A). Secondary lineage mutation ratios were between 0.216 and 0.941 (Fig. 2B). The lowest ratios were found for the mix 90:10 with only 0.23, 0.25, 0.39, and 0.55 of BA.1-specific mutations found using Midnight V2, Midnight V1, Artic V4, and Artic V4.1, respectively. These low ratios were associated with primer bias preferentially amplifying Delta over Omicron (Suppl Results, Supplementary Data 2).

A Best main lineage match ratio for each Delta: Omicron mix, considering all mutations expected for the given lineage. B Best secondary lineage match ratio for each mix, considering mutations expected for the given lineage, excluding those in common with the main lineage. Pure isolates (denoted as 100:0 for Delta and 0:100 for Omicron) are depicted with dots, and Delta: Omicron mixes with triangles. Colors represent the expected frequency of each mix using a red (Delta: 100:0) to blue (Omicron: 0:100) gradient.

To test whether our pipeline may detect co-infection in mixes with a lower percentage of Delta at risk of leading to chimeric sequences, we performed additional mixes with Delta: Omicron ratios of 1:99 and 5:95 that were sequenced in 10 replicates (Fig. S6 and Supplementary Data 3). All chimeric sequences were detected by seqmet as co-infected, except one chimeric 1:99 mix sequenced with Midnight V1 primers classified as non co-infected (Suppl Results, Fig. S6).

Altogether, the results of the unbiased co-infection detection scripts were consistent with the curated list approach with a better detection of low abundant BA.1 with Artic primers.

Prevalence of SARS-CoV-2 co-infections during the fifth wave in France

Between December 6, 2021 (week 49-2021) and February 27, 2022 (week 08-2022), 23,242 samples were sequenced using Artic V4 or V4.1 primer sets as they became available. In total, WGS (coverage >90%) were obtained for 21,387 samples collected from Flash surveys (n = 16,220) and from HCL and peripheral hospitals (n = 5,167).

Among the 21,387 samples, 61 samples (0.29%) had a secondary lineage identified with a positive secondary lineage mutation ratio (Fig. 3A). Notably, changing the specific mutation cutoff between 3 and 17 had no significant impact on this detection rate (Suppl Results, Fig. S7). To rule out potential contamination during initial sequencing, all these 61 samples were re-extracted and sequenced in duplicate. In total, eight samples had no secondary lineage identified in duplicate while sequenced with high efficiency (coverage >90%), suggesting potential contamination during the first sequencing process (Suppl Results, Figs. S8 and S9). In contrast, 53 samples had a positive secondary lineage mutation ratio in duplicate: 28 samples were identified as a Delta/Omicron (BA.1) co-infection; 1 sample was identified as a Delta/Omicron (BA.2) co-infection; 24 samples were identified as a co-infection between two different Omicron lineages (BA.1 and BA. 2) (Fig. 3A).

A Violin plot of main and secondary virus mutation ratios for the 21,387 samples, including 61 samples sequenced in duplicate. The lineage ratio is considered to be 0 in cases where less than 6 mutations were found or covered with at least 100x for any lineage. Positive secondary virus mutation ratios in independent duplicates with concordant virus identification were considered as potential co-infections and depicted with dots (n = 53). Colors indicate lineages: co-infections between Delta/BA.1 are in green, Delta/BA.2 in dark blue, and BA.1/BA.2 in salmon. B Correlation between the relative abundance of the minor lineage in n = 53 natural co-infections sequenced in duplicate. Relative abundance is measured by the median of allele frequencies specific to the minor lineage. Distributions of relative abundance in first and second replicates are depicted as marginal boxplots. Boxplot represents median and IQR, with upper and lower whiskers extending to the largest and smallest value, respectively. Dotted lines are drawn at 5%, which is the threshold used to call a mutation in the variant calling step of seqmet. Colors indicate lineage as in panel A. Shapes indicate whether the minor lineage is identified as the secondary or main lineage in duplicate or a potential discordance between duplicates. C Number of chimeric sequences detected among the 53 natural co-infections sequenced in duplicate, depending on the relative abundance of the minor lineage. Chimeric sequences were defined by sequences bearing specific mutations of two lineages, with a gradient of color depending on the number of specific mutations from the less abundant lineage: 0 mutation (light grey), 1–2 specific mutations (medium grey), >2 specific mutations (dark grey).

For these 53 natural co-infections, relative abundance of minor lineages were correlated in duplicate (Pearson correlation coefficient = 0.92, p-value <2.2e-16) with a median relative abundance of 20% (iqr = 26.25) (Fig. 3B). Minor lineages were identified as the secondary lineages in both duplicates for 45 samples but were identified either as the main or secondary lineage in duplicates for 8 samples with a higher relative abundance of minor lineage (median relative abundance = 38,75%, Kruskall–Wallis p-value = 9.479e-06) (Fig. 3B).

The number of mutations specific to the minor lineage present in the consensus sequence was used as a proxy to identify chimeric sequences (Fig. 3C). Chimeric sequences were identified in 69/106 (65%) natural co-infected duplicates. All samples with a minor lineage above 38% were chimeric with a median of 4 minor lineage-specific mutations present in the consensus sequence (Fig. 3C).

All co-infections were confirmed by visual examination of vcf plots using the lists of specific mutations from our lineage variant database (Figs. S10–S12). Uniform frequencies of lineage-specific mutations along the genome were observed in each sample, except for three samples: “021228537801 (R1)” and “021229656701 (R3)” among the Delta/BA.1 co-infections and “722000801801 (R2)” among the BA.1/BA.2 co-infections (Suppl Results, Fig S13). In these samples, the distribution of lineage-specific allele frequencies suggests the presence of potential recombinants with different relative abundance within co-infected samples (Suppl Results, Figs. S14 and S15). The suspicion of a recombination event among co-infected samples was therefore established for 3 out of 53 samples (5.6%).

Delta/Omicron (BA.1) co-infections were detected between weeks 50-2021 and 04-2022 (Fig. 4). Considering only the period of Delta and Omicron (BA.1) co-circulation at relative frequencies above 1% (Weeks 50-2021 to 05-2022), Delta/Omicron (BA.1) prevalence was estimated at 0.18% (28/15,253; 95 CI: 0.12–0.26% assuming a binomial distribution). The highest prevalence of Delta/Omicron (BA.1) co-infections was observed in weeks 51 and 52-2021 with 0.31% (6/1,921) and 0.25% (7/2,847), respectively. These 2 weeks were characterized by the highest co-circulation of Delta and Omicron (BA.1) (week 51: 69.2% Delta and 30.1% BA.1; week 52: 34.5% Delta and 65.1% BA.1). BA.1/BA.2 co-infections were detected between weeks 04 and 08-2022 with prevalence reaching 0.78% of the sequences during the highest co-circulation of BA.1 and BA.2 (week 08-2022: 57.3% BA.1 and 42% BA.2). Considering only the period of BA.1 and BA.2 co-circulation at relative frequencies above 1% (Weeks 03 to 08-2022), BA.1/BA.2 prevalence was estimated at 0.26% (24/9,120; 95 CI: 0.17–0.39%).

Colors indicate lineages: Delta (B.1.617.2 and AY.* lineages) in red, Omicron (BA.1) in blue, Omicron (BA.2) in purple, Omicron (BA.3) in light blue, other lineages (including B.1.640.1) in grey. Co-infections between Delta/BA.1 are in green, Delta/BA.2 in dark blue, and BA.1/BA.2 in salmon.

Clinical presentation of Delta/Omicron and BA.1/BA.2 co-infections

To assess the impact of co-infections on clinical presentations, demographic features, including age and sex were reported for 13,187 out-patients, 6242 hospitalized patients, and 803 healthcare workers (Table 1). In the three groups, no significant difference was noted between BA.1 and BA.2 infections regarding proportion of men (p > 0.05 Fisher tests) or regarding median age for out-patients and healthcare workers (p > 0.05, Mann–Whitney test). Therefore, BA.1 and BA.2 cases were grouped into Omicron cases for further analysis. Delta cases were significantly older than Omicron cases for out-patients (p = 0.003, Mann–Whitney test) and for hospitalized patients (p < 0.001, Mann–Whitney test). Delta cases were also significantly more predominant in men for hospitalized patients (p < 0.001, Fisher test). No difference regarding age or sex was found between Delta and Omicron cases for healthcare workers (p > 0.05, Mann–Whitney test or the Fisher tests). Among the three groups, no significant difference in age or sex was found for Delta/Omicron or BA.1/BA.2 co-infections compared to any other type of infection (p > 0.05, Mann–Whitney test or the Fisher tests), except for BA.1/BA.2 co-infections significantly reported in younger out-patients than Delta out-patients (p = 0.03, Mann-Withney test) and in younger hospitalized patients than Delta/Omicron hospitalized patients (p = 0.004, Mann-Withney test). ICU admission for hospitalized patients was reported in 5.56%, 15.38%, 1.95%, and 0% of Delta, Delta/Omicron, Omicron, and BA.1/BA.2 cases, respectively. Binomial logistic regression was tested and after correction by age and sex, ICU admission was significantly associated with Delta (p < 0.001, Wald Chi-square test) or Delta/Omicron (p < 0.01, Wald Chi-square test) infection. Among the 53 co-infected patients, a total of 21 patients (39.6%) were not vaccinated, corresponding to 9/16 (56.3%) hospitalized patients and 12/36 (33.3%) out-patients with no significant difference between the two groups (p > 0.05). Vaccinated patients (32/53, 60.4%), had received mainly two doses (18/32, 56.3%) or three doses (13/32, 40.6%) of the vaccine. The median (IQR) delay between last dose and the date of RT-PCR test was 124 (57–166) days.