Deep phylogenetic-based clustering analysis uncovers new and shared mutations in SARS-CoV-2 variants as a result of directional and convergent evolution

Nearly two decades after the last epidemic caused by a severe acute respiratory syndrome coronavirus (SARS-CoV), newly emerged SARS-CoV-2 quickly spread in 2020 and precipitated an ongoing global public health crisis. Both the continuous accumulation of point mutations, owed to the naturally imposed genomic plasticity of SARS-CoV-2 evolutionary processes, as well as viral spread over time, allow this RNA virus to gain new genetic identities, spawn novel variants and enhance its potential for immune evasion. Here, through an in-depth phylogenetic clustering analysis of upwards of 200,000 whole-genome sequences, we reveal the presence of previously unreported and hitherto unidentified mutations and recombination breakpoints in Variants of Concern (VOC) and Variants of Interest (VOI) from Brazil, India (Beta, Eta and Kappa) and the USA (Beta, Eta and Lambda). Additionally, we identify sites with shared mutations under directional evolution in the SARS-CoV-2 Spike-encoding protein of VOC and VOI, tracing a heretofore-undescribed correlation with viral spread in South America, India and the USA. Our evidence-based analysis provides well-supported evidence of similar pathways of evolution for such mutations in all SARS-CoV-2 variants and sub-lineages. This raises two pivotal points: (i) the co-circulation of variants and sub-lineages in close evolutionary environments, which sheds light onto their trajectories into convergent and directional evolution, and (ii) a linear perspective into the prospective vaccine efficacy against different SARS-CoV-2 strains.


Introduction
In the last two decades, human health has been threatened by the emergence of three important zoonotic and pathogenic betacoronaviruses, namely the severe acute respiratory syndrome coronavirus (SARS-CoV) [1], the Middle East respiratory syndrome coronavirus (MERS-CoV) [ (COVID- 19) pandemic, SARS-CoV-2 [3]. Likely originated from bats, pandemic SARS-CoV-2, like other endemic human alpha-(NL63 and 229E) and beta-(OC43 and HKU1) CoVs known for causing upper respiratory tract infections, overcame the interspecies barrier as a result of spillover and/or recombination events, and gained a pervasive ability to rapidly infect and spread around the globe [4][5][6].
The COVID-19 pandemic precipitated an intense genomic surveillance via data depositories and sequencing platforms and led to an unprecedented accumulation of public genomic data concerning a human pathogenic virus [5,7]. The sheer amount of available sequencing data has the potential to facilitate higher-precision micro-evolutionary analyses mapping escape and point mutations in presumed positively selected sites and residues putatively associated to an increased virus fitness and pathogenesis and allows inferences concerning the dynamics of SARS-CoV-2 spread [8,9].
Although the analysis of micro-evolutionary mechanisms is of paramount importance and may provide powerful information to promote the prediction of vaccination perspectives and the tracing of SARS-CoV-2 epidemiological chains, there is as yet a lack of data-based investigations examining the presence of eventual shared mutations and their evolutionary characteristics in classified SARS-CoV-2 Variants of Concern (VOC) and Variants of Interest (VOI) [10,11].
Given the importance of monitoring mutations to track the emergence of novel variants, here we investigate the influence of directional selection and the dynamics of SARS-CoV-2 genomic plasticity in VOC and VOI by clustering partition high-scale phylogenetic and directional evolution (DEPS) approaches. Additionally, we show the presence of several mutations common for both VOI/VOC and convergently emerged sub-lineages, and provide a perspective of possible effects on the vaccination efficacy and the ongoing COVID-19 pandemic.

Sequence data and filtering strategy
High-coverage and complete HCoV-229E and HCoV-NL63 (alpha-CoVs), HCoV-OC43, HCoV-HKU1, MERS-CoV, SARS-CoV and SARS-CoV-2 VOC and VOI (beta-CoVs) genome sequences (� 29,000 bp), sampled from humans, were retrieved from the Global Initiative on Sharing Avian Influenza Data-EpiCoV (GISAID-EpiCoV) and GenBank databases at three different times, each date representing a snapshot of the COVID-19 spread at that time: February 12 th (MERS-CoV, SARS-CoV and SARS-CoV-2), July 12 th (HCoV-229E, HCoV-NL63, HCoV-OC43, HCoV-HKU1 and SARS-CoV-2) and updated in August 26 th 2021 (SARS-CoV-2), totalling 238,990 sequences. With regards to SARS-CoV-2, we particularly focused on strains of countries from South America, China, India, and the United States of America (USA). At the time of analysis, India, the USA, and Brazil had reported the largest numbers of cumulative confirmed COVID-19 cases and deaths. This approach was used to compare putative mutual sites and residue changes under directional evolution over time [12].

Clustering and sub-clustering analysis
A methodological approach to extract large-scale phylogenetic partitions was applied to identify transmission cluster chains on the large Maximum Likelihood (ML) phylogenetic trees of the SARS-CoV-2 variants circulating in South America, China, India and the USA on the basis of a depth-first search algorithm which unifies evaluation of node reliability, tree topology and patristic distance [14]. In this case, different datasets from each particular SARS-CoV-2 scenario were used in order to extract the clustering and sub-clustering data. Therefore, each ML tree was implemented in FastTree v.2.1.7 by using the standard implementation General Time Reversible (GTR) plus CAT with 20 gamma distribution parameters and a mix of Nearest-Neighbor Interchanges (NNI) and Sub-Tree-Prune-Regraft (SPR) [15]. Thereafter, in view of identifying SARS-CoV-2 cluster transmission events, for datasets comprising more than 100 sequences, we first selected sequences (one per cluster) from nodes/sub-trees with � 2 distinct individuals and � 90% reliability of statistical support (Shimodaira-Hasegawa test), where initially the patristic distance was adjusted to find a representative number of clusters (n = 100, which represent 100 sequences) from each large reconstructed ML tree. In addition to this strategy, a second approach included sub-clustering analysis as an indirect way to infer and investigate the possibility of co-circulating sub-lineages. For this, we selected sequences (two per cluster) with � 95% node reliability of statistical support from a threshold of 0.05, thus corresponding to the 5 th percentile when considering the whole-tree patristic distance distribution.

Recombination and directional evolution analyses
Before proceeding to directional evolution analysis, sequence datasets from the output provided by the clustering analyses were submitted to the Genetic Algorithm for Recombination Detection (GARD), a likelihood-based tool to pinpoint recombination breakpoints [16]. To check the outcome of the strategy described above, an additional test was conducted using the Pairwise Homoplasy Index (PHI; default settings) [17]. Evidence-based analysis through phylogenetic maximum-likelihood was then performed implementing the Datamonkey webserver and the program Hyphy v.2.5 to track directional selection in amino acid sequences (DEPS) [18]. The DEPS method identifies both the residue and sites evolving toward it with great accuracy and detects frequency-dependent selection-scenarios as well as selective sweeps and convergent evolution that can confound most existing tests [8]. Further, the DEPS method has shown better performance than (traditional) substitution rate-based analyses (dN/dS) in detecting transient and frequency-dependent selection and directionally evolving sites and residues. For the most part, a Beta-Gamma site-to-site rate variation was used to conduct the analysis. The best-fit protein substitution model was chosen according to the corrected Akaike Information Criterion (cAIC). Only target sites and residues with Empirical Bayes Factors for evidence in favour of a directional selection model equal to or greater than 100 were considered for further exploration. Certain datasets, randomly chosen, were run multiple times (more than eight) to confirm obtained results.
Statistical analyses were performed using one-way analysis of variance (ANOVA) and nonparametric methods followed by post hoc Kruskal-Wallis and Friedman (both with Dunn's Multiple Comparison), and Bartlett's tests (Tukey's, Newman-Keuls and Bonferroni's multiple comparisons). Additionally, Mann Whitney and Wilcoxon matched-pairs signed-rank (T test) and Pearson/Spearman (Correlation), all one-tailed methods with 99% confidence interval (CI), were run. P-values equal to or less than 0.005 (p � 0.005, SARS-CoV-2 from South America: DEPS [sites and/or residues] vs infections) and 0.05 (p � 0.05, SARS-CoV-2 from Brazil, China, India and the USA: DEPS [residues] vs circulating variants and infections) were considered as statistically significant. Data analyses were carried out using GraphPad Prism v. 5

Results and discussion
Recombination is known to be a crucial evolutionary process for many RNA viruses [19][20][21]; the process is frequently observed in the Coronaviridae family where recombination is likely facilitated by discontinuous transcription involving jumps of the replication-transcription complex during minus strand RNA synthesis. However, the consequences of recombination events occurring in the context of the current SARS-CoV-2 VOC/VOI evolutionary landscape are still speculative [22,23]. Here, we address this knowledge gap, revealing the presence of recombination and shared mutations in the SARS-CoV-2 Spike-encoding protein, demonstrating them to be under directional and convergent evolution amongst SARS-CoV-2 VOC/VOI and sub-lineages, and tracing an interconnection with viral spread. First, endemic and epidemic human coronaviruses (HCoVs) were compared to identify similar evolutionary patterns that could help clarify the evolution of SARS-CoV-2. An initial recombination breakpoint analysis showed that four of six HCoVs analyzed presented such signals ( Fig 1A). Endemic viruses OC43, NL63 and HKU1 also showed a similar pattern of residue accumulation and directional evolution, despite these viruses being subject to differing selective pressures [24].
A subsequent comparison of SARS-CoV-2 to the other two pathogenic HCoVs (SARS-CoV and MERS-CoV), highlighted differences in the number of directionally-evolving sites and residues ( Fig 1B). These patterns, putatively reflecting the initial evolutionary paths of the individual viruses, may suggest that SARS-CoV was initially under lower positive evolutionary pressure than MERS-CoV and SARS-CoV-2. Nevertheless, deletions and mutations acquired by SARS--CoV have been shown to have had an impact on adaptation to human-to-human transmission, modifying both the capacity for viral proliferation and profiles of pathogenesis [25][26][27].
The evolution of SARS-CoV-2 was initially marked by genetic drift in a typical process of neutral evolution [28,29]; the virus reached a large number of new and susceptible hosts and, although some mutations appeared along the genome, there was no significant shift [30]. However, as SARS-CoV-2 rapidly spread worldwide [31,32], fitness changes resulting from mutations in the viral genome as well as the emergence of new variants were increasingly reported [33][34][35].
The first epidemic wave of SARS-CoV-2 severely affected most countries in South America as a probable result of multiple viral introductions [36]; rapid increases of case numbers were especially reported in Brazil, the largest and most populous country in Latin America [37,38]. The uncontrolled viral spread created a favorable scenario for the emergence of new variants [39][40][41][42]. To identify the impact of directional-positive selection sites at the rate of infections under these particular conditions, we traced the evolutionary scenario of SARS-CoV-2 in South America (via analysis of a significant and representative amount of genome sequences).
Remarkably, our data showed that an increase of DEPS was correlated with viral spread dynamics, with Brazil exhibiting a lower proportion of COVID-19 cases (per 100,000 inhabitants) when compared to French Guiana and the same amount of SARS-CoV-2 clusters inferred in Chile (n = 97) (Fig 1C), probably due to a higher diversity of circulating viruses. Our results also highlighted a series of mutations; while certain mutations have previously been described, but have hitherto remained unidentified in SARS-CoV-2 VOC and VOI, multiple further mutations are identified for the first time in this study, to our knowledge (Table 1).

PLOS ONE
Deep phylogenetic-based clustering analysis uncovers new and shared mutations in SARS-CoV-2 variants Table 1 Spike mutations such as E484K, N501Y, L452R, S13I and W152C, seem to be fundamentally important in the process of adaptation of SARS-CoV-2 to human hosts, this by enhancing the affinity to the human ACE2 receptor and mediating immune system evasion [43][44][45] (Table 1), allowing for inferences concerning a SARS-CoV-2 reproductive number increase. An increase in virus circulation augments the chance of viral coinfection, which in turn (and as a prerequisite for recombination), can heighten the risk of emergence of new variants [46,47].

Empirical Bayes Factors
The Delta variant, first identified in late 2020 in India as B.1.617.2 [48], harbors a constellation of non-synonymous mutations in the Spike protein [49] and had become the leading VOC worldwide by the end of July 2021, this VOC accounted for 90% of all sequenced samples [50,51]. Brazil, India and the USA, the countries most severely affected by the pandemic, are now once again threatened by this highly contagious variant. Analysis of the molecular evolution of SARS-CoV-2 taking into account the influence of local demography in these specific scenarios has the potential to generate important insights into the spread and infection dynamics of this pathogen.
Using SARS-CoV-2 sequences from China (the most populated country in the world and also a country where control measures against COVID-19 infections have been deployed very effectively) as reference, we analyzed all datasets from Brazil, India, and the USA via a large-scale phylogenetic partitions analysis. Increases in SARS-CoV-2 infections were observed to be proportional to locally circulating variants and were not (in the scenarios analyzed), correlated with any particular demography ( Fig 1D); this indirectly reinforces the importance of measures implemented to avoid viral propagation. Analysis of phylogenetic partition clusters along the length of the circa 30 kb CoV genome evidenced several directionally-evolving sites under convergent evolution. Thus, a possible

PLOS ONE
Deep phylogenetic-based clustering analysis uncovers new and shared mutations in SARS-CoV-2 variants association between the number of infections from locally circulating SARS-CoV-2 variants carrying distinct residue profiles as well as sites in the Spike-encoding protein under DEPS may be established ( Fig 1D). Interestingly, this supports a hypothesis of convergent evolution due to repeated and multiple site-specific substitutions in distinct SARS-CoV-2 VOC and VOI (see Table 1 and Fig 2). Additionally, we also inferred the possible appearance of SARS-CoV-2 sub-lineages and traced the influence of an environment favoring directional evolution acting on SARS-CoV-2 variants. We showed different patterns among sites in the VOC and VOI, with a particular emphasis on the Kappa VOI currently circulating in the USA. We further demonstrated recombination among SARS-CoV-2 VOC and VOI from India (Beta, Eta and Kappa) and the USA (Beta, Eta and Lambda) ( Fig 3A).
As one of the first countries in the world to develop efficient immunizations and implement a vaccination policy, the USA vaccinated more than 30% of its population by April 2021 [52,53]. By September 2021, 60% of the booster-immunized population possessed neutralizing antibodies against several viral variants [54]. Similar outcomes were observed following widespread vaccination with various SARS-CoV-2 vaccines (different technologies leveraged for vaccine production) in many other regions, including South America and India [55,56]. We hypothesize that the site-specific mutations found under convergent evolution at strategic positions in the SARS-CoV-2 VOC/VOI Spike protein targeted by vaccine-induced neutralizing antibodies, as shown by Andreata-Santos et al. (2022) [57], may also help to explain these findings (Fig 4). Nonetheless, viral circulation in the face of incomplete immunization has been described as one of the probable causes of the  Table 1 and Fig 1D Venn

PLOS ONE
Deep phylogenetic-based clustering analysis uncovers new and shared mutations in SARS-CoV-2 variants emergence of new variants [42] (Sabino et al., 2021). Accordingly, our own analysis identified SARS-CoV-2 VOC and VOI sub-clusters (Fig 3B), thus indicating co-circulation of variants and sub-lineages carrying new mutations under convergent evolution. Surprisingly, the same evolutionary pattern was also observed for other endemic and epidemic CoVs studied (see Data availability).
This study demonstrates the influence of positive directional evolution on SARS-CoV-2 circulating in South America and in those countries most severely affected by the COVID-19 pandemic. Furthermore, our methodology allowed the identification of distinct transmission sub-clusters and recombination breakpoints in many SARS-CoV-2 variants, to our knowledge, not previously shown so far. We were able to indirectly infer transmission of a viral epidemiological chain and the generation of new variants. Through sequence data collected in February, July and August 2021, we also further identified and classified several convergently emerged shared mutations in different SARS-CoV-2 VOC and VOI. Lastly, we hypothesize that the cocirculation of SARS-CoV-2 variants and their possible sub-lineages takes place within a very close evolutionary environment, which can be translated to a setting of strong convergent and directional evolution. The latter may have affected the number of different lineages within each country, with highly infected countries being impacted the most (as can be seen from comparison against China). This agrees with the general virological perspective that the larger the number of viruses circulating, as well as the number of lineages within a region, the more likely it becomes that sub-lineages will originate and spread. Our results confirm the importance of critical assessment, monitoring, and control of SARS-CoV-2 lineages and sub-lineages throughout the pandemics, even more so in countries with large populations where many opportunities arise for positive selection, due to both recurrent substitutions and recombination events.