Memory Gaps in America: Mutational and Immunoinformatic Analysis of Evolving SARS-CoV-2 Variants of Concern and Interest

VOC and VOI mutation assessment

Summaries of each variant, its classification, defining spike mutations, earliest sampling date, current variant sequence count, and number of variant sublineages are reported in Table I. All four VOCs were identified within a 3-mo span in 2020 (July 15–October 1), with VOIs emerging before and after this period (Fig. 1, Table I). The event count for the VOCs ranges from 27,000 to 1,000,000, and the event count for the VOIs ranged from 4,000 to 18,000; the variant with the highest sequence count to date is alpha (B.1.1.7) at 1,020,185 reported events at GISAID (Table I).

Genetic plasticity, as evidenced by the number of sublineages, ranged between 4 and 34 for the VOCs and 0 and 1 for the VOIs; the variant with the most genetic plasticity, as evidenced by number of sublineages, is delta (B.1.617.2), with 34 sublineages currently reported in Pango (Table I). Accordingly, VOCs exhibit increased propagation and subsequent genetic variation in relation to the VOIs.

Variability was additionally assessed using a representative variant for each VOC and VOI (see Materials and Methods). An unusually high number of nucleotide mutations were observed in the delta variant, at 76, compared with the predicted average of 6–11 nucleotide diseases (14). In comparison, the remaining VOCs/VOIs exhibited 21–45 mutations and the control group averaged 17 ± 6.8 mutations (Table II). To reconcile this, we additionally assessed NSP14, a proofreading enzyme that imparts high fidelity replication of the virus (15). NSP14 mutations were cataloged for both the delta variant as well as 13 of the 44 control variants (Table II). Specifically, the delta variant possessed the NSP14 A394V mutation (Table II), which occurs in ∼30% of total GISAID-archived variants. However, direct comparison of mutation frequency across distinct delta variant sequences did not support the conclusion that the A394V mutation conferred a reduction in enzyme function (data not shown). An additional consideration was NSP12, the RNA-dependent RNA polymerase that catalyzes viral replication and transcription (16). All VOCs and VOIs possessed the P323L mutation occurring in the non-catalytic domain of NSP12. Importantly, the GISAID database reports that the P323L mutation has 3,474,929 occurrences among all SARS-CoV-2 variants, indicating that it is a favorable mutation, not just in VOCs and VOIs.

An accepted indicator applied to mutational and evolutionary analyses is the ratio of nonsynonymous to synonymous nucleotide mutations (NS/S). Critically, nonsynonymous mutations lead to amino acid substitutions that impart biological change and evolutionary consequences. For SARS-COV-2, the genome-wide NS/S ratio is 1.88 (14), a value that is orders of magnitude higher than the predicted values observed for other (+) sense mRNA viruses (17) and suggests significant accrual of adaptive mutations. Although all VOCs/VOIs exhibited more amino acid mutations, relative to the control variants (10.5 ± 3.3, mean ± SD), the delta variant and mu variant sequence possessed disproportionately more amino acid substitutions (compare 33 and 34 versus 16–21 mutations) (Table II). Relating these changes back to NS/S ratios, the beta, delta, epsilon, kappa, and lambda variant sequences observed NS/S ratios ranging from 3.2 to 9.5, indicating that they have undergone accelerated evolution (Table II). The reason for these unusually high ratio values cannot be rationalized based on mutation in NSP12 or NSP14, and thus it is likely that extrinsic factors related to chronic infection in the setting of immunodeficiency may have led to this pronounced evolution rate (18). In comparison, the alpha, gamma, and mu variants possessed NS/S ratios that approximated the predicted ratio, between 1.7 and 2.7, respectfully, indicating that they are evolving along a more predicted pattern, as likewise observed in the random control sequences (1.8 ± 0.8, mean ± SD) (Table II).

Mutations influencing innate immune function

Mutations in SARS-CoV-2, as a whole, exhibit a strong bias for deamination of cytosine to uracil (C→U), comprising ∼35–50% of the total nucleotide mutations at the genomic level (14, 19). Of note, C→U mutations are postulated to be catalyzed by host RNA editing enzymes and appear to confer resistance to zinc finger antiviral protein (ZAP), an IFN-induced gene product that promotes degradation of viral RNA (20). Specifically, earlier studies assessing HIV have shown that C→U mutations reduce the occurrence of viral CpG motifs in ssRNA that in turn reduce the ability for recognition and subsequent elimination of viral RNA via a ZAP-dependent mechanism (21). All VOCs exhibited C→U mutation frequencies that were significantly lower than the predicted rate, ranging from 21 to 25%, whereas the VOI and control sequence values were within range, at 36–40% and 47.2 ± 10.85% (mean ± SD) (Fig. 2, Table II,). This dichotomy implies that distinct evolutionary pressures appear to be driving the mutation pattern in the highly successful VOCs. Indeed, excess C→U mutations likely confer amino acid substitutions that negatively affect protein function, thus reducing the overall fitness of the virus when occurring at exaggerated frequencies.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2.

Percent C→U mutations.

Values represent the percent of nucleotide mutations, relative to the reference strain, that represented C→U substitutions.

Mutations influencing Ab detection

Amino acid changes that influence Ab binding were assessed through three distinct approaches. First, high-frequency immunodominant epitopes described in the United States (see Materials and Methods) derived from reference strain spike, matrix, and nucleocapsid proteins were compared against the corresponding mutant sequences. Only gamma and mu possessed mutations in these previously defined immunodominant epitopes (Table III), specifically at spike H655Y, near the furin cleavage site associated with increased transmissibility, which modestly reduces the theoretical Ab-binding affinity to 68% (Table III). This mutation is observed in ∼3% of archived sequences and therefore does not appear to confer an overt selective advantage for these lineages. Correspondingly, the lambda variant EPI_ISL_1111130 possessed the nucleocapsid T366I mutation, although this is not lineage-defining, nor does it represent a high-frequency event.

Second, key Ab-binding sites in spike protein previously cataloged at GISAID on the basis of their ability to impart antigenic shift were systematically assessed. All VOCs and VOIs possessed two or more amino acid substitutions at these defined sites (Table III). Importantly, the delta and mu variants possessed the most substitutions, with a total of four and five changes in Ab-binding sites, respectively, implying that both lineages pose significant challenges to humoral immunity, particularly to vaccine-induced immunity, which is singularly dependent on the spike protein. In comparison, only 1 of 44 (<3%) of the control sequences had mutations at these sites. Thus, mutations that confer evasion from Abs are selectively enriched in VOC and VOI lineages.

We additionally assessed lineage-specific mutations in the spike protein occurring at positions 417, 452, 484, 501, and 681, which likely represent adaptive traits based on their independent recurrence in distinct VOCs and VOIs (Table I). K417N and K417T mutations present in the beta, delta, and gamma variants appear to enhance affinity for the ACE2 receptor as well as reduce the in vitro neutralizing activity of vaccine and infection-acquired Abs (Ref. 22 and K. Wu, A.P. Werner, J.I. Moliva, M. Koch, A. Choi, G.B.E. Stewart-Jones, H. Bennett, S. Boyoglu-Barnum, W. Shi, B.S. Graham, manuscript posted on bioRxiv, DOI: 10.1101/2021.01.25.427948). The mutations E484K and E484Q occurring in the receptor-binding domain of the delta, epsilon, kappa, and mu variants have also been shown to reduce the binding of Abs derived from postvaccine and convalescent serum by 3- to 5-fold (6, 23). The N501Y mutation in alpha, beta, gamma, and mu variants is associated with increased viral replication in the nasal cavity as well as increased affinity for the ACE2 receptor (24). Furthermore, the N501Y mutation also appears to reduce reactivity with convalescent sera in vitro (25). The P681R mutation, located within the furin cleavage site, is thought to enhance delta variant infectivity through increasing S1/S2 cleavage (B. Lubinski, T. Tang, S. Daniel, J.A. Jaimes, and G.R. Whittaker, manuscript posted on bioRxiv, DOI: 10.1101/2021.01.25.427948). The L452R mutation, which presents in the delta, epsilon, and kappa variants, has been shown to increase viral infectivity and impairment of neutralizing Ab binding (26). Accordingly, mutations in spike protein amino acid residues 417, 452, 484, and 501 are associated with humoral immune evasion.

Finally, we assessed therapeutically significant amino acid substitutions that influence mAb binding to spike protein, specifically bamlanivimab, etesevimab, sotrovimab, and the combined casirivimab/imdevimab mixture, which are influenced by changes to spike protein at positions 417, 452, and 484 (2, 27, 28). To date, there are no reported mutants with in vitro susceptibility to sotrovimab or the combined casirivimab/imdevimab mixture. The beta and gamma VOCs, with the combined K417T, E484K, and N501Y mutations, show reduced binding to bamlanivimab and etesevimab. The L452R mutation present in delta and epsilon modestly reduces the neutralizing ability of these therapeutics as well. The N501Y mutation alone does not appear to influence any of the mAb-based therapies. The clinical significance of these in vitro findings remains unclear but they imply reduced reactivity in vivo. The propagation of these mutations in response to mAb-based interventions is unlikely given the low application rate of these drugs.

Mutations influencing HLA binding and T cell detection

In comparison, to assess T cell–mediated recognition, we identified the most immunodominant epitopes of the spike, matrix, and nucleocapsid proteins from the reference strain and ran the sequences against the top three high alleles for HLA-A, HLA-B, HLA-C, HLA-DR, and HLA-DQ alleles seen in the U.S. population (see Materials and Methods). The mutant sequences were then compared against the reference strain epitopes to assess fold change in binding affinity. No lineage-defining mutations were observed for HLA class I binding immunodominant epitopes derived from spike, nucleocapsid, or matrix protein (Table IV). This finding implies that in the U.S. population, recognition of VOCs and VOIs by CD8⁺ T cells remains largely unaffected. Correspondingly, each VOC and VOI possessed at least one mutation in nucleocapsid protein affecting HLA-DQA/DQB binding, whereas immunodominant epitopes of spike and matrix remained unchanged (Table IV). In some instances, double mutations occurred in the same immunodominant epitope, as seen for alpha, gamma, and lambda (Table IV). This double mutation event was only observed to occur once in the randomly selected control strains. Critically, these mutations clustered around the same region and were shared across different lineages: nucleocapsid R203K and G204R double mutations were present in alpha, gamma, and lambda lineages, resulting in 37% original affinity. The nucleocapsid R203M single mutation was observed in epsilon and kappa lineages, resulting in 50% binding affinity, and T205I was observed in beta, epsilon, and mu lineages reducing relative affinity relative to 84% that of the reference strain. The R203K/G204R mutations have previously been associated with increased infectivity and transmissibility (29). Although the T205I mutation has been shown to reduce detection capability by SARS Ag fluorescent immunoassay tests (30), the outcome of this mutation of HLA binding or T cell recognition has not yet been established.