Drivers of adaptive evolution during chronic SARS-CoV-2 infections – Nature.com

Diverse evolutionary patterns in chronic infections

We begin by defining criteria for a chronic infection. In clinical settings, a chronic infection is often defined as one with both prolonged shedding of viral RNA and evidence of infectious virus, either through virus isolation in tissue culture or via detection of subgenomic RNA. However, when surveying various studies reporting chronic infection, we noted a lack of standardization, with different studies defining chronic infections somewhat inconsistently. Hence, we expanded our focus to include patients displaying high-viral-load (VL) shedding for 20 or more days while mining the literature for all such cases that were accompanied by longitudinal whole-genome sequencing of the virus (Methods). The criterion of 20days was based on a meta-analysis of the duration of viral shedding (defined as a positive nasopharyngeal polymerase chain reaction (PCR) test) across thousands of patients diagnosed until June 2020, which revealed that mean duration of upper respiratory tract shedding was around 17days, with a 95% confidence interval ranging from 15.5days to 18days16. Of note, shedding of replication-competent virus lasted markedly less than 20days. Moreover, estimates of viral shedding are different in some of the more recently detected SARS-CoV-2 variants, such as Delta and Omicron17,18, yet, as described below, our analysis focused on variants that were found in earlier stages of the pandemic.

Our search yielded a total of 21 case reports, all of which reported patients who were diagnosed during 2020 or early 2021, and all of which reported patients who were infected with viruses belonging to lineages that pre-dated the Alpha variant (Supplementary Table 2). In addition, six patients adhering to the above criteria were identified in TASMC, and all available samples were sequenced (Methods). Five TASMC patients suffered from hematologic cancers. The sixth patient suffered from an autoimmune disorder and was treated with a high dose of steroids. The six TASMC patients were all diagnosed in late 2020 or early 2021, with four patients infected with a virus from pre-Alpha lineages and two patients infected with a virus from the Alpha lineage (Supplementary Table 2).

Of the 27 chronically infected patients (mean age (s.d.) 55 (21.3) years; 17/27 male), we inferred that all were immunocompromised due to one or more of the following: hematologic cancer (that inherently tends to lead to immunosuppression), direct anti-B cell treatment, high-dosage steroid treatment or very low CD4+ T cell counts (due to AIDS). We observed very different evolutionary outcomes across the range of patients examined, from considerable evolution and antibody evasion observed in some patients to relatively static evolution in others (Table 1 and Supplementary Tables 1 and 2).

We searched for patterns of evolution across all 27 patients with chronic infection and compared this pattern to the pattern observed under (1) mostly neutral evolution, in the first approximately 9months of viral circulation19,20 (data were obtained from a sample of ~3,500 sequences generated by NextStrain https://nextstrain.org/21 (Methods)) and under (2) presumed positive selection, which occurred in the lineages leading to the five currently defined VOCs (Alpha, Beta, Gamma, Delta and Omicron) (data on lineage-defining mutations (LDMs) of VOCs were obtained from https://covariants.org (Fig. 1a and Supplementary Table 4)). In each scenario, we searched for binsthat is, consecutive regions of 500 basesenriched for mutations (P<0.05, binomial test, after correction for multiple testing; Methods).

a, Comparison of substitutions observed in chronic infections to VOC LDMs and to substitutions dominated by genetic drift during globally dispersed acute infections. Shown are the number of substitutions observed along the SARS-CoV-2 genome, in bins of 500 nucleotides. The upper panel displays substitutions observed at any timepoint of the 27 chronic infections. The middle panel displays LDMs of the five currently recognized VOCs. The lower panel displays substitutions observed globally during the first 9months of the pandemic, mostly before the emergence of VOCs. Asterisks mark bins enriched for more substitutions using a one-tailed binominal test, after correction for multiple testing (P<0.05; Methods and Supplementary Table 8). The genomic positions are based on the Wuhan-Hu-1 reference genome (GenBank ID NC_045512), and the banner on the top shows a breakdown of ORF1a/b into individual proteins and domains of the S protein (see main text). b, A network of co-occurring substitutions across patients with chronic SARS-CoV-2 infection. Each colored circle represents a locus, and a black asterisk and dot represent a significant enrichment under a one-tailed Fishers exact test with P<0.05 and P<0.1, respectively, after correction for multiple testing. Blue asterisks represent enrichment of co-occurring substitutions in globally observed sequences using a one-tailed X2 test, with P<0.05 and P<0.1, respectively, after correction for multiple testing (Methods).

During the first 9months of virus circulation, we noted that 61% of substitutions were non-synonymous, which is generally what we could expect under lack of both positive and purifying selection and in line with reports suggesting incomplete purifying selection during the early stages of SARS-CoV-2 spread22. During this time, we observed a relatively uniform distribution of substitutions across most of the genome, with some enrichment in ORF3a, ORF7a, ORF8 and N. This enrichment was previously reported and may be due to more relaxed purifying selection in these regions or higher mutation rates19; adaptive evolution at these regions also cannot be ruled out.

In general, the patterns obtained in chronic infections and in the LDMs of VOCs were very similar. The average proportion of non-synonymous substitutions in chronic infections and LDMs of VOCs was 78% and 82%, respectively, which was much higher than that observed during the first stage of the pandemic and generally suggestive of positive selection. On the other hand, we see less similarity between mutations in chronic infections and mutations that fix after a VOC has emerged (Supplementary Fig. 1), with a much lower proportion of non-synonymous substitutions in the latter (on average, 61%). A likely explanation for this observation is that after a VOC spreads in the population, selection is more limited due to the very tight transmission bottleneck9,10,11,12.

The most striking similarity between chronic infections and VOC LDMs was observed along the S protein and, in particular, at the regions that correspond to the N-terminal domain (NTD) (genomic nucleotides 21,59822,472) and the receptor-binding domain (RBD) (genomic nucleotides 22,51723,183). Several mutations at the RBD have been shown to enhance affinity to the ACE2 receptor and allow for better replication23,24, whereas other mutations, both at RBD and NTD, are known to enhance antibody evasion25,26,27. The most commonly observed substitutions in chronic infections were in the S protein: E484K/Q and various deletions in the region spanning the NTD supersite, particularly amino acids 140145, all shown previously to confer antibody evasion28. Chronic infections shared the enrichment of ORF3a/ORF7a/ORF8 mutations with the neutral set but lacked an enrichment across most of the N protein. Overall, it seems that mutations in chronic infections are predictive of LDMs of VOCs, as was noted previously2.

When focusing on the differences between VOCs and viruses in chronic infections, several intriguing differences emerged. First, four VOCs bear a three-amino-acid deletion in the nsp6 protein (ORF1a:3,6753,677), which is an event not observed in our set of chronic infections. Next, in VOCs, there is an enrichment in the region of the S encompassing the S1/S2 boundary (positions 23,50024,000 in Fig. 1a). This enrichment is primarily driven by S:P681H/R, a highly recurrent globally occurring mutation29, surprisingly never observed in our chronic infection set. A recent study analyzed recurrent mutations, with recurrence indicative of positive selection, and tested which of the recurrent mutations led to clade expansionthat is, were associated with onwards transmission30. Some recurrent mutations led to more dense clades, suggesting that they were especially successful in driving transmission, whereas others did not lead to considerable onwards transmission, suggesting that they were less successful. Notably, we observed that successful recurrent mutations were almost never present in our chronic set, whereas less successful recurrent mutations (S:E484K/Q and S:144) were the most abundant (Table 2). Overall, these results suggest that there may be a tradeoff between antibody evasion and transmissibility. This tradeoff, if it exists, might not play a role in chronic infections but would affect the ability of a variant created in a chronic infection to be transmitted onwards. Thus, only under specific conditions, a transmissible variant would emerge in chronic infections. Four of five VOCs independently acquired a mutation at or near the S1/S2 boundary (S:P681H/R or H655Y), suggesting that this may be a factor driving transmissibility. We note that Beta is an exception with no such mutations, yet this variant also displayed limited global transmission.

We went on to examine co-occurring substitutions, defined as pairs of substitutions that appeared in two or more patients. We used Fishers exact test to assess whether pairs of substitutions occurred together more often than expected from their individual frequencies (Methods) as a measure of possible epistasis. Intriguingly, four pairs of substitutions across four different proteins emerged as significantly enriched and formed a network of interactions: T30I in envelope, H125Y in the membrane glycoprotein, S13I in the S protein and T3058I in ORF1a (Fig. 1b). This finding was intriguing on multiple fronts. First, envelope and membrane glycoprotein have generally remained very conserved throughout the entire pandemic, and, specifically, the two replacements found are at highly conserved sites (Supplementary Table 1). However, despite their rarity, we found that some of the pairs of mutations also tend to significantly co-occur in globally dispersed sequences (blue asterisks in Fig. 1b). The replacements in S and ORF1a, on the other hand, have been observed only a small number of times in the global phylogeny. Notably, all of the first three proteins form a part in the virion structure itself; however, the functional meaning of this remains unclear. Other pairs of mutations found to co-occur were the three most common S antibody evasion mutations, yet these co-occurrences were not statistically significant. Larger cohorts of patients and further data will be required to determine the implications of these findings.

We noted very wide variation in the background and treatments given to different patients, both for their background condition and for Coronavirus Disease 2019 (COVID-19). When examining medical background, the patients could be roughly classified into one of the following categories: hematologic cancers, HIV/AIDS, organ transplantation and autoimmune disorders (Table 1). The latter two categories were often treated with steroids. Some, but not all, of the patients with hematological cancer and others were treated with antibodies targeting B cells, presumably causing profound B cell depletion. In line with this, most of the patients with confirmed B cell depletion showed negative serology for SARS-CoV-2 at one or more timepoints (Supplementary Table 1). Some patients were treated with ABT against SARS-CoV-2, whereas others were not; and, in some ABT-treated patients, antibody evasion mutations were detected, whereas, in others, they were not. Finally, we found that, whereas in some ABT-treated patients, antibody evasion mutations were detected, sometimes these mutations fixed before the treatment. The course of VL across time, coupled with ABT, is illustrated for some patients in Fig. 2b. Thus, for example, patient P5 and the patient described by Choi et al.8 are shown to fix antibody evasion mutations just before ABT.

a, Results of a random forest classifier used to explain an outcome of antibody evasion. The effect of each feature on model outcome is shown: mean SHAP absolute values (left) and individual SHAP values for each feature, ordered based on contribution (right). The color range corresponds to the values of each feature, from red (high value) to blue (low value). b, Illustration of individuals who experienced viral rebound and mutations associated with antibody evasion. Ct values are used here as an inversed proxy for VL and are presented according to the day of infection (denoted as number of days after the first positive PCR test), with the dashed red horizontal line and shaded area representing a negative or borderline result, respectively. Blue dots represent samples that were sequenced. Only amino acid replacements in the S protein are shown, with predicted antibody evasion mutations shown in bold (Supplementary Table 1). Positive samples from BAL, ETA or sputum are indicated in brown. Antibody-based anti-COVID-19 treatments are represented by dashed vertical lines on the day of administration. ALL, acute lymphoblastic leukemia; APS, antiphospholipid syndrome; CLL, chronic lymphocytic leukemia; ETA, endotracheal aspirates; P, patient.

We noted that many patients (four of the six patients sequenced herein and several others in the total set of 27 patients) displayed an intriguing cycling pattern of VL (reflected by cycle threshold (Ct) values), with very high Ct values reaching negative or borderline-negative results at one or more stages of the infection, followed by rebound of the virus (Fig. 2b). In the four above-mentioned patients, this rebound was accompanied by clinical evidence of disease, which is highly suggestive of active viral replication. Several different hypotheses could explain this pattern. First, the virus may have cleared and been followed by re-infection with another variant. Because this pattern can be ruled out using sequencing, such cases were excluded from our analysis (Methods). Second, the virus may cycle between different niches, such as upper and lower airways. Its re-emergence in the upper airways (nasopharynx) may be due to selective forces or genetic drift. When considering selective forces, viral rebound may occur due to the near clearance of the virus, driven either by ABT or by the endogenous immune system, and followed by the emergence of a more fit variant with antibody evasion properties.

We fit a random forest classifier to assess the effect of different clinical and demographic features on an outcome of antibody evasion (Methods and Supplementary Tables 2 and 3). We treated each sequencing timepoint as a sample and used age, sex, B cell depletion, steroid treatment, days-since-infection, ABT and viral rebound as explaining variables. We then trained a classifier while considering the structure of the data, composed of samples belonging to the same patient (Methods). After training, we generated SHapley Additive exPlanations (SHAP) values31,32 that quantified the effect of each feature on the classifiers outcome. We found that the feature with the strongest association with antibody evasion was viral rebound, followed by days-since-infection and age (Fig. 2a). Other features had a relatively minor effect, and similar results were obtained with other classifiers (Supplementary Figs. 2 and 3). Regarding the effect of age, we note that young individuals are a minority in this dataset and rarely present an antibody evasion mutation, and, thus, the small sample size may be responsible for the small effect observed with this feature. All in all, these results suggest that ABT is not necessary for driving antibody evasion, in line with the fact that evasion is sometimes observed before (for example, E484K in P5; Fig. 2b) or in the absence of ABT (for example, ref. 33). If so, what may be driving immune escape in some patients is actually the weakened immune system of the patient, although ABT and its waning may also play a role in some patients. To summarize, viral rebound may serve as an indicator for the emergence of a mutant with properties of antibody evasion (Fig. 2b), and monitoring for viral rebound in patients with chronic disease is critical.

Next, we went on to examine patterns of variation over time across the different patients. In many of the case reports, the authors noted the emergence and disappearance (and sometimes re-emergence) of particular substitutions (Fig. 3). For example, in patient B reported by Perez-Lago et al.34, the mutation S:A1078V is present at a low frequency on day 81, rises to fixation on day 100 and then drops and disappears from day 107 onwards (Fig. 3). When re-analyzing the data, we noted that this pattern of dynamic polymorphisms across time was observed in most patients (Supplementary Table 2). From an evolutionary point of view, it is quite unlikely for one or more substitutions to disappear from a given population, and, because we observe this at very different loci across all patients, we consider that it is not likely that all of this pattern is due to recurrent sequencing problems or due to biases of the viral polymerase. We and others have previously noted sequencing errors that occur predominantly when VL is low, when errors that occur during reverse transcription or early PCR cycles are carried over to higher frequencies10,11,35. However, this phenomenon most often leads to errors in intra-host variants segregating at relatively low frequency and is less common at the consensus sequence level, which is defined here as mutations present at a frequency of 80% or higher. We, thus, conclude that the existence of dynamic polymorphisms likely reflects subpopulations of the virus that co-exist in a patients body, as further discussed below.

Each series of boxed lines represents a patient, and each line represents a sequenced timepoint with time-since-infection on the right. The different open reading frames are color-coded. For each patient, only mutations relative to the first timepoint sequenced that appeared at a frequency ranging from 20% to 100% are shown. Most samples were nasopharyngeal, except those marked by asterisks, which were obtained from endotracheal aspirates.

Visit link:

Drivers of adaptive evolution during chronic SARS-CoV-2 infections - Nature.com