Shared quasispecies architecture in experimental and natural RNA virus populations Authors: Samuel Martínez-Alcalá, Iker Atienza-Diez, Pilar Somovilla, Brenda Martínez-González, Celia Perales, Luis F Seoane, Ester Lázaro, Susanna Manrubia Published: 2026 ABSTRACT AbstractRNA viruses form genetically diverse populations structured as mutant spectra, or quasispecies, whose internal organization influences their evolutionary and adaptive dynamics. While genetic diversity has been extensively characterized, the structural organization of viral populations in sequence space remains less explored. Here, we compare genotype network architectures in two RNA viruses with markedly different evolutionary contexts: bacteriophage Qβ evolving in controlled laboratory conditions and severe acute respiratory syndrome coronavirus 2 evolving within infected human hosts. Using deep sequencing data, we reconstruct the genotype network of mutationally coupled variants within viral populations and analyse their topological properties. Despite large differences in genome size, mutation rate, and ecological setting, both viruses exhibit a common organization: a highly abundant central haplotype surrounded by layers of variants of diminishing abundance as Hamming distance to the central haplotype increases. All reconstructed networks share qualitative and quantitative topological features, displaying a hierarchical structure. The robust organization of both populations under multiple conditions suggests that RNA viruses may share a common genotype network architecture governed by fundamental properties of sequence space and the generic mechanisms of replication and mutation. Genotype networks provide a unifying framework to describe viral population structure beyond conventional diversity measures and, by revealing how local constraints shape mutational search, offers insights into the predictability of viral evolution.Keywords: viral quasispecies, genotype networks, deep sequencing, hierarchical networks, network topology Introduction RNA viruses replicate with high error rates, giving rise to genetically heterogeneous populations composed of closely related genomes. These populations are commonly described as viral quasispecies (Andino and Domingo, 2015), a concept originally formulated in theoretical studies of self-replicating molecules (Eigen, 1971) and later extended to RNA virus evolution (Domingo et al., 1978). In this framework, viral populations are viewed as dynamic distributions of mutants shaped by the interplay between mutation and selection. Such mutant spectra constitute the raw material for rapid adaptation to environmental challenges, including new hosts, immune pressures, and antiviral therapies (Mattenberger et al., 2021; Somovilla et al., 2022; Haddox et al., 2026). Early experimental and theoretical work on RNA viruses consistently showed that viral populations are not dominated by a single genotype. Rather, they are composed of a structured distribution of mutants around a most abundant sequence. In bacteriophage Qβ, classic studies demonstrated that mutation continuously generates single-step variants, such that most genomes differ from the consensus at only one or a few positions, and that the frequency of genotypes declines with mutational distance from the master sequence (Batschelet et al., 1976; Domingo et al., 1978). Similar conclusions were reached for vesicular stomatitis virus, where populations were described as mutant ‘clouds’ tightly clustered around a central sequence, with progressively fewer high-distance variants due to the action of purifying selection (Holland et al., 1982). This is a rough but ubiquitous pattern subsequently observed in other RNA viruses, such as hepatitis C (Martell et al., 1992), human immunodeficiency virus (Henn et al., 2012), or severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) (Delgado et al., 2024). Together, these observations support a general view in which viral mutant spectra are strongly concentrated around a central genotype, with abundance decreasing as mutational distance increases due to the combined effects of mutation and selection (Lauring and Andino, 2010). Empirical work on RNA viruses has often focused on quantifying genetic diversity within populations (Sanjuán et al., 2021) and in exploring how diversity relates to viral fitness, pathogenesis, or adaptation (Lauring and Andino, 2010; Bordería et al., 2011). High-throughput sequencing has enabled increasingly detailed descriptions of mutant spectra across a wide range of viruses (Posada-Cespedes et al., 2017; Pérez-Losada et al., 2020), including human pathogens such as hepatitis C virus (Quer et al., 2017) and SARS-CoV-2 (Lythgoe et al., 2021). Analyses of SARS-CoV-2 mutant spectra in infected patients have shown that intra-host genetic complexity correlates with clinical outcomes (Martínez-González et al., 2022) and may represent an epidemiologically relevant trait (Martínez-González et al., 2022, 2025). Notably, populations with similar levels of diversity can still exhibit distinct patterns in the organization of their mutant clouds. The internal architecture provides complementary information on the likelihood that minority variants persist in the population, on the range of mutations actively explored by the quasispecies as a whole, and also on the evolutionary trajectories available to the virus under changing selective pressures. Despite previous studies on the ‘cloud’ of mutants surrounding the most abundant sequence, the detailed structure of the quasispecies, namely, how individual haplotypes are mutationally related and how their abundances are distributed across the underlying genotype space, is a relevant but still poorly explored feature. This internal organization can be studied through the reconstruction of genotype networks, defined as graphs of mutationally connected variants in which haplotypes differing by a single position are linked. Genotype networks derived from consensus haplotypes have uncovered the extent of convergent evolution and positive selection in influenza A (Aguilar-Rodríguez et al., 2018), as well as correlations between viral diversity and local network structure (Williams et al., 2022). However, studies addressing the characterization of genotype networks for intra-host viral diversity are scarce so far. A recent work on the RNA bacteriophage Qβ has analysed the structure of a set of populations evolved through serial passages at different temperatures that impose distinct selective pressures on the virus. Genotype networks reconstructed from deep-sequencing data revealed the existence of hierarchical networks centred around the most abundant haplotype, with variants arranged in concentric layers corresponding to increasing Hamming distance from this central sequence (Seoane et al. 2026b). This structure determines the extent to which populations explore sequence space. Remarkably, this hierarchical organization persists across evolutionary passages and environmental conditions, suggesting that it may arise from simple and general processes. Whether such a structure is generalizable to different RNA viruses, subject to different selective pressures or evolving in natural infections, remains an open question. In this contribution, we perform a systematic comparison between bacteriophage Qβ and SARS-CoV-2, two positive-sense RNA viruses with markedly different biological and evolutionary contexts. The SARS-CoV-2 genome (≈ 30 kb) is about an order of magnitude larger than that of Qβ and encodes a proofreading exonuclease (Moeller et al., 2022) that reduces the effective mutation rate relative to many RNA viruses (Symons et al., 2025). Indeed, analyses of biological clones show that viable SARS-CoV-2 genomes accumulate on average several-fold fewer mutations than other RNA viruses such as foot-and-mouth disease virus (Ávila et al., 2025). In addition, SARS-CoV-2 populations evolve within infected hosts under immune pressure and are transmitted through epidemiological chains characterized by severe population bottlenecks (Lythgoe et al., 2021) followed by periods of intra-host replication and diversification. Using deep sequencing data, we reconstruct and compare the genotype network structure of Qβ populations evolving in vitro with that of SARS-CoV-2 populations sampled from infected patients. Our results reveal striking similarities between the two systems. Despite the profound differences in their evolutionary histories and replication environments, their genotype networks display the same hierarchical architecture, characterized by a highly connected central haplotype surrounded by layers of increasingly rare mutants. Minor quantitative differences between the two viruses are only weakly associated with viral identity and are more strongly influenced by sequencing coverage, which determines how deeply sequence space is sampled. Within a virus, differences are also independent of the specific sample and of the genomic regions analysed. These findings suggest that the structural organization of viral quasispecies may be governed primarily by universal properties of RNA virus replication and sequence space rather than by virus-specific ecological conditions. Materials and methods Qβ phage data Qβ phage has a short RNA genome (4217 nt) that encodes four proteins (see Fig. 1a). It infects the bacterium Escherichia coli using the conjugative F pilus as a receptor. Its optimal replication temperature is C, although it can readily adapt to replicate between C and C (Arribas et al., 2014; Somovilla et al., 2019; Arribas and Lázaro, 2021; Somovilla et al., 2022). Schematic of viral genomes and genotype network reconstruction. (a) Genome architecture of the Qβ phage and (b) SARS-CoV-2. The positions and relative sizes of the two amplicons used for genotype network reconstruction are indicated. (c) Summary of data origin and filtering pipeline, with main procedures numbered. (d) Illustration of genotype network reconstruction and key topological quantities. For this study, we consider representative populations from various experiments in which the phage was evolved at different constant temperatures (C, C, C, and C) and analyse two amplicons in different genomic regions (see Fig. 1a and Table 1). Experiments proceeded for up to 60 passages, and population samples from different passages were deep sequenced following previously described procedures (Seoane et al. 2026b). Populations analysed for amplicon 1 correspond to the same evolutionary lineages studied previously (Seoane et al. 2026b), but sampled at passage 30 instead of passage 60. Thus, the populations evolved at C, C, and C and analysed in the main text all derive from the ancestral population P2, a lysis plaque that had undergone two laboratory passages at C, and represent earlier stages of the same evolutionary trajectories described previously (Seoane et al. 2026b). Table 1Characteristics of viral population samples used in this study. A, total number of sequences and D, total number of haplotypes after data curation; , number of sequences and , number of haplotypes (nodes) in the LCC; , number of links in the LCC; Root haplotype: indicates whether it matches the WT or lists mutations relative to WT (see also section 2.3); and , maximum Hamming distance from the root within the LCC. SC2 denotes SARS-CoV-2. In this study, amplicon 2 for SARS-CoV-2 corresponds to amplicon 6 in a previous publication (Martínez-González et al., 2022). The last three columns report fitted values of exponents characterizing topological features of the genotype networks: α for the degree distribution derived from fits to the CCDF, β for assortativity and δ for sequence space covering (see Section 4). Figures for amplicon 1 are presented in the main text; results for amplicon 2 are provided as Supplementary Information.Virus/AmpliconSample A D \cal L Root α β δ Qβ/130C543 45846 037526 87529 46366 203WT72.28(1)0.57(1)0.137(1)Qβ/137C264 63313 316261 066975520 595WT52.54(2)0.76(1)0.132(1)Qβ/143C513 81751 443496 74934 41878 477G1312A72.21(1)0.60(1)0.137(2)Qβ/230C709 64145 670696 78832 85472 625WT62.21(1)0.66(1)0.142(0)Qβ/240C346 88617 423341 37711 91723 939WT62.47(2)0.75(1)0.138(2)Qβ/243C286 48717 171281 43012 11724 703WT62.56(2)0.75(1)0.141(2)SC2/1Mild134 66821 706127 75314 79429 215WT41.76(1)0.91(1)0.129(2)SC2/1Moderate147 11220 074139 92712 89025 558WT41.87(2)0.92(1)0.126(2)SC2/1Severe146 84718 099139 94911 20122 279WT42.02(3)0.93(1)0.124(2)SC2/2Mild140 67224 843131 57715 74931 091A23403G41.77(1)0.91(1)0.130(2)SC2/2Moderate172 44528 106161 32816 99033 795A23403G41.78(1)0.91(1)0.129(2)SC2/2Severe138 09820 815129 49712 21524 291A23403G41.98(3)0.92(1)0.126(2) Characteristics of viral population samples used in this study. A, total number of sequences and D, total number of haplotypes after data curation; , number of sequences and , number of haplotypes (nodes) in the LCC; , number of links in the LCC; Root haplotype: indicates whether it matches the WT or lists mutations relative to WT (see also section 2.3); and , maximum Hamming distance from the root within the LCC. SC2 denotes SARS-CoV-2. In this study, amplicon 2 for SARS-CoV-2 corresponds to amplicon 6 in a previous publication (Martínez-González et al., 2022). The last three columns report fitted values of exponents characterizing topological features of the genotype networks: α for the degree distribution derived from fits to the CCDF, β for assortativity and δ for sequence space covering (see Section 4). Figures for amplicon 1 are presented in the main text; results for amplicon 2 are provided as Supplementary Information. To evaluate whether the observed genotype-network properties were robust across genomic regions, we also analysed a second genomic fragment (amplicon 2; see Fig. 1a and Table 1). Because sequencing data for this amplicon were not available for the same populations described above, a different set of Qβ populations was used. The amplicon-2 populations evolved at C and C correspond to passage-60 samples from the same P2-derived evolutionary lineages described previously (Seoane et al. 2026b). In contrast, the population evolved at C corresponds to passage 30 of an independent evolutionary lineage derived from another ancestral population, termed P25, which originated from a lysis plaque that had undergone 25 serial passages at C before the temperature-adaptation experiment (Somovilla et al., 2022). All populations were propagated under constant-temperature conditions up to the passage analysed here. The evolution temperatures were chosen based on previous laboratory experience (Arribas et al., 2014; Somovilla et al., 2019; Arribas and Lázaro, 2021; Somovilla et al., 2022), taking into account that extreme temperatures impose strong selective pressures on both the virus and the host. At C, host metabolic processes are slowed down, whereas at C the heat shock response is fully activated in E. coli. In Qβ, temperature has a strong impact on replication dynamics by affecting RNA stability, protein folding, and the efficiency of viral assembly. RNA extraction, reverse transcription polymerase chain reaction (RT-PCR) amplification, and deep-sequencing library preparation for the Qβ populations followed protocols described in detail elsewhere (Somovilla et al., 2019, 2022); these procedures and the data-filtering pipeline have also been summarized previously (Seoane et al. 2026b). Universal Molecular Indices for per-molecule error correction were not used in this workflow. The basal sequencing error rate and the frequency of PCR-induced recombination associated with this methodology were characterized independently using hepatitis C virus (Soria et al., 2018), prior to its application to the Qβ and SARS-CoV-2 datasets analysed here. The reproducibility of the amplification and sequencing workflow was previously assessed through independent PCR and sequencing replicates, showing a high recovery of reconstructed haplotypes above a conservative abundance threshold (Somovilla et al., 2022). Quality controls supporting the filtering procedure used to distinguish genuine low-frequency haplotypes from sequencing artefacts have also been reported (Seoane et al. 2026b, Supplementary Information). SARS-CoV-2 data SARS-CoV-2 has a non-segmented, large positive-sense single-stranded RNA genome of ∼29.9 kb, comprising 14 open reading frames and encoding 27 proteins (16 non-structural, four structural, and seven accessory proteins), see Fig. 1b. For this study, we analyse samples from patients admitted to Fundación Jiménez Díaz Hospital (Madrid, Spain) during the first COVID-19 outbreak in Spain, within a short time window early in the pandemic and during the transition between clades 19 and 20. All patients were confirmed to be positive for SARS-CoV-2 (Martínez-González et al., 2022). Patients were classified based on COVID-19-associated clinical parameters, including the need for hospitalization, requirement for mechanical ventilation, admission to the intensive care unit, and death attributed to COVID-19. According to these criteria, cases were categorized as mild (no hospital or ICU admission), moderate (hospitalization without ICU admission), and severe (ICU admission, all resulting in death). In the present analysis, we focus on a subset of patients, including two representative cases from each clinical category (mild, moderate, and severe). Viral RNA was extracted from patient samples and reverse-transcribed, followed by PCR amplification of the target regions and library preparation for Illumina sequencing, using sample- and amplicon-specific barcoding for pooling and demultiplexing. Full details of the RT-PCR conditions, amplification cycling, and library preparation protocol have been published previously (Martínez-González et al., 2022a, 2022b) and are summarized in a recent methodological review (Perales et al., 2026). Technical reproducibility of the amplification and sequencing workflow, assessed through independent replicate amplifications of patient samples and of the reference isolate USA-WA1/2020, has also been reported (Delgado et al., 2024). Universal Molecular Indices for per-molecule error correction were not used in this workflow; potential effects of this limitation on low-frequency variants are discussed in Section 5. Data curation From the available datasets for the two viruses, a subset of samples was selected for detailed comparative analysis. Specifically, representative samples were chosen for each clinical category in SARSCoV-2 (mild, moderate, and severe), and for each temperature condition in Qβ. Two different amplicons were analysed: in the main text we show and discuss results for amplicon 1 in both viruses; results for amplicon 2 are provided as Supplementary Information. In Qβ, the first deep-sequenced fragment spans nucleotides 1060 to 1330 (amplicon 1); positions 1060 to 1320 code for a fragment of the A2 maturation-lysis protein, 1321 to 1323 correspond to a termination codon, and 1324 to 1330 are non-coding positions. The second fragment (amplicon 2) spans nucleotides 2145 to 2473, coding for fragments of the A1 coat protein (2145–2330), a non-coding region (2331–2351), and the replicase (2352–2473). In the case of SARS-CoV-2, the two selected amplicons were deep sequenced in samples obtained from infected patients. The first fragment spans nucleotides 14 534 to 14 920 and codes for part of the non-structural protein nsp12, corresponding to the viral RNA-dependent RNA polymerase (RdRp). The second fragment spans nucleotides 23 259 to 23 645 and codes for a region of the structural Spike protein. Figure 1a,b summarizes the viral genomes and illustrates the positions of the analysed amplicons. FASTQ files derived from SARS-CoV-2 samples within the patient cohort are publicly available in the European Nucleotide Archive (ENA) repository under project ID PRJEB48766. This study received approval from the Ethics Committee and the Institutional Review Board of Fundación Jiménez Díaz (FJD) Hospital (no. PIC-087-20-FJD). Raw data curation in both cases consisted of removing low-quality sequences, sequences not aligned to the reference genome, and sequences containing insertions or deletions. This process resulted in a large set of high-quality sequences of equal length, which were subsequently collapsed into D unique sequences (haplotypes) while retaining their abundances, following previously described protocols (Villanueva et al., 2022; Seoane et al. 2026b). Figure 1c schematizes the data curation process. After curation, each dataset consisted of a set of haplotypes (one per virus, experimental condition or patient, and passage or sample) and their abundances. In the genotype network representation, each haplotype corresponds to a node, and two nodes are connected if they differ by a single nucleotide (Fig. 1d). In this study, we consider only haplotypes belonging to the largest connected component (LCC) of the genotype network, defined as the largest set of haplotypes connected through single point mutations (see the next section for network reconstruction). This component represents >95% of all sequences and includes only mutationally coupled variants. However, a substantial fraction of haplotypes, mostly represented by one sequence and only rarely by two, falls outside the LCC. This can be seen by comparing columns D and in Table 1; the fraction of haplotypes outside the LCC ranges from 0.27 to 0.41. Haplotypes outside the LCC are more likely to originate from sequencing errors, whereas singletons and doubletons retained within the LCC are mutationally connected to more abundant neighbouring haplotypes and are therefore more likely to correspond to sequences actually present in the quasispecies population, a collection of closely related genomes. Previously published quality controls support the reliability of this filtering procedure (Seoane et al. 2026b, Supplementary Information). In our representation of genotype networks, node size is proportional to the logarithm of abundance, łn s(i), where s(i) is the number of occurrences of haplotype i in the curated dataset. For convenience, we define the root sequence as the most abundant haplotype in each population. Note that the root haplotype does not necessarily coincide with the reference or consensus genome, labelled WT for short. In the case of SARS-CoV-2, the reference genome is Wuhan, with accession number NCBI NC045512.2 (Martínez-González et al., 2022a, 2022b), and for Qβ the reference genome is GenBank AB971354.1 (Somovilla et al., 2022; Seoane et al. 2026b). Table 1 summarizes the collection of samples analysed, including the total number of sequences A and haplotypes D after curation, the number of sequences and haplotypes in the LCC, and the number of links in each sample. WT indicates that the root haplotype coincides with the wild type (see Materials and Methods); otherwise, mutations in the root relative to the WT sequence are specified. HD denotes the maximum Hamming distance from the root among sequences in the LCC. Because of the filtering procedure, there is at least one continuous parsimonious mutational pathway (single-nucleotide steps) connecting the root haplotype to any other haplotype in the LCC, even those at distance . The last three columns report least-squares fits to selected topological features of the reconstructed genotype networks (see Section 4). Algorithm for network reconstruction Network reconstruction was performed using a custom implementation of a pivot-based metric-similarity algorithm. The approach exploits the triangle inequality for Hamming distances. Consider a reference sequence, , and two additional sequences, and . Let d(0,i), d(0,j), and d(i,j) denote the corresponding Hamming distances, i.e. the numbers of point mutations separating each pair of sequences. For and to be connected in the genotype network, they must satisfy d(i,j)=1. Because Hamming distances obey the triangle inequality, distances between sequences can be bounded using their distances to a reference sequence. In particular, if d(i,j)=1, then |d(0,i)-d(0,j)| ≤ 1. This condition allows us to avoid many direct pairwise comparisons. However, explicit comparisons are still required for candidate pairs because |d(0,i)-d(0,j)| ≤ 1 is a necessary but not sufficient condition for d(i,j)=1. The same relationship holds for any reference sequence; therefore, additional pivots can be used to impose further constraints and reduce the number of direct comparisons. The logic is analogous to triangulating a position on a map from several anchors. Specifically, we proceeded as follows. We first selected a reference sequence, , usually the most abundant haplotype, because this also provides the full distance distribution from the central node used in later analyses. We then computed the Hamming distance from every other sequence to . Next, we selected the haplotype farthest from , denoted , as a second reference and computed the Hamming distance from every sequence to . For two nodes to be connected, they must satisfy both |d(0,i)-d(0,j)| ≤ 1 and . We skipped direct comparisons for pairs of sequences that failed at least one of these constraints, and explicitly computed distances only for the remaining candidate pairs. This algorithm, and variations of it, have been independently rediscovered in different fields, including genotype-network reconstruction (Ruiz, 1986; Micó et al., 1994; Catanese et al., 2018). We implemented our own version, specifically tailored to directly reconstruct genotype networks from collections of. fasta files. We made our code available in a public repository (Seoane 2026a), including instructions for ease of use. Network visualization We systematically compare the structure of genotype networks for Qβ and SARS-CoV-2 across the populations listed in Table 1. Specifically, we examine population structure across different environments and assess potential differences arising from sequencing coverage. For SARS-CoV-2, we analyse one sample per level of disease severity (mild, moderate, and severe), while for Qβ we consider populations at passage 30 at each temperature condition. All analyses refer to amplicon 1; results for analyses of amplicon 2, and a comparison of the structure of both amplicons is provided in the Supplementary Information. Details of the six populations and their corresponding genotype networks are summarized in Table 1. Figure 2 presents a hierarchical representation of the reconstructed genotype networks. In these representations, genotype networks are constructed by connecting sequences that differ by a single nucleotide, such that the network encodes local mutational accessibility between variants. This framework provides a direct visualization of how viral populations are organized in sequence space beyond simple measures of diversity. In all cases, each node corresponds to a unique nucleotide sequence (haplotype) of the genomic fragment selected for each virus. Hierarchical representation of genotype networks. Three representative populations of amplicon 1 for Qβ phage and SARS-CoV-2 have been chosen to illustrate the hierarchical structure of the LCC genotype network. The most abundant sequence (Root) is located in the centre of the network, and concentric circles with variants at increasing Hamming distance are represented. The colour and line code will be maintained all through this work to facilitate feature comparison. The most abundant haplotype (Root) is located at the centre of the network and is surrounded by concentric layers of variants at increasing Hamming distance. Thus, proximity to the root reflects the number of mutational steps separating each haplotype from the dominant sequence. Node colour and node size both represent the natural logarithm of the abundance. These representations reveal broadly similar structures across all populations. However, the extent of variation around the root differs between the two viruses. SARS-CoV-2 populations contain variants up to three or four mutations away from the root, whereas Qβ populations reach distances of up to seven mutations (; Table 1). This difference may reflect disparities in sequencing coverage between the datasets (column A in Table 1), a possibility that we examine in a later section. In general, higher coverage (A) implies that the LCC of the population is larger, with more haplotypes () and number of links (\cal L). The average degree per haplotype in the LCC, , is approximately constant and independent of the coverage, with a small difference between the two viruses. Averaged over all populations, Qβ networks exhibited , whereas SARS-CoV-2 networks yielded . A Welch two-sample t-test indicated that this difference is statistically significant (t = 3.60, P =.015), whereas no systematic differences were observed between amplicons within either virus. In the next section, we analyse the topology of the reconstructed networks for each virus to determine whether the observed qualitative similarities are also supported by quantitative measures. Network topology We focus now on network topology to compare the main features of both viruses, with particular attention to their similarities and differences. Following previous analyses of genotype networks as complex networks (Aguirre et al., 2011; Seoane et al. 2026b), we consider three relevant features: the degree distribution p(k), assortativity, and the number of triangles per node. We also measure the following quantities as a function of the Hamming distance d to the root: the total number of distinct haplotypes, the fraction q(d) of variants present at each distance, an average dN/dS ratio customized for genotype networks, and the degree of exploration of sequence space (sequence space covering, C(d)). We begin by defining these quantities. Definitions Degree distribution. The degree of a node i in the LCC is defined as the number of links it has (see Fig. 1d). This corresponds to the number of haplotypes in the dataset within the 1-mutant neighbourhood of the focal haplotype i. The degree distribution p(k) represents the probability that a node has degree k in a given sample. A common alternative representation of this probability is the complementary cumulative distribution which yields the fraction of haplotypes with degree K greater than k. The cumulative distribution provides clearer visualization as it averages out noise. This also enables a more robust estimation of distribution parameters. Degree assortativity. Assortativity quantifies how similar two nodes in a network are with respect to a given property, in this case node degree. The average degree of the haplotypes in the 1-mutant neighbourhood nn(i) of haplotype i is The assortativity coefficient r is the Pearson correlation coefficient of degree between pairs of linked nodes (Newman, 2002). Positive values r> 0 indicate that nodes of similar degree tend to be mutually linked (assortative networks). Negative values r < 0 reveal a preference for links between nodes of different degree, i.e. nodes of lower degree preferentially link to hubs, and vice versa (diassortative networks). Networks in which links are randomly assigned between nodes have values of independent of degree k, and r ≃ 0. Triangles. Triangles in genotype networks are formed when at least three of the possible substitutions at a given sequence position are present in the population sample (see an example in Fig. 1d). Around a given haplotype i, one triangle is formed for each position with two mutants present, while three triangles are formed if all three possible variants at that nucleotide are observed. The total number of triangles △ (i) in the 1-mutant neighbourhood of haplotype i is the sum of all such contributions along the sequence. △ (i) is an informative measure of local clustering in genotype networks, as it quantifies deviations from neutral expectations for each haplotype and admits an interpretation in terms of sequence neutrality (Seoane et al. 2026b). Specifically, one obtains if mutations are randomly distributed along the sequence, meaning that all sites have the same degree of neutrality and are equally prone to accept mutations. Significant deviations towards the maximum possible number of triangles, , indicate variation in site neutrality along the sequence, while deviations towards the minimum possible number of triangles represent situations in which additional mutants at an already mutated position are disfavoured. The analytical derivation of , , and has been published previously (Seoane et al. 2026b). Empirical haplotype diversity. The total number of distinct haplotypes in the LCC, , of each sample is given in Table 1. This set can be partitioned into subsets according to their Hamming distance d (number of point mutations) from the root sequence. Each subset has size , such that . Haplotype diversity explored at distance d. The maximum number of haplotypes at distance d is given by for a four-letter alphabet (Krug and Jain, 2005; Seoane et al. 2026b), where L is the length of the amplicon in each sample. Note that each haplotype has at most 3L neighbours. The ratio measures the fraction of the total possible diversity explored in each sample at distance d. dN/dS ratio per sequence and averaged for each d. For the coding region of each haplotype, the fraction of non-synonymous versus synonymous mutants in its 1-mutant neighbourhood, w ≡ dN/dS, is calculated as follows: given node i, we define S(i) as the total number of possible synonymous haplotypes at distance 1, and as the number of observed synonymous haplotypes at distance 1, regardless of abundance. Similarly, N(i) and denote, respectively, the total possible and observed numbers of non-synonymous haplotypes in the 1-mutant neighbourhood of i (note that, by definition of the 1-mutant neighbourhood, ). Then is expected to be close to 0 if synonymous haplotypes are more frequently observed than haplotypes with non-synonymous mutations around haplotype i. The average dN/dS ratio at each distance d is calculated as the mean over all haplotypes at that distance, This quantity presents a subtle bias towards small values, particularly in low-degree nodes. When few neighbours exists, it might happen that no synonymous variants are present, resulting in a 0 in the denominator of Equation 5 and infinitely large w(i). We must leave these cases out. Below, we show how results from number of triangles (which is unbiased) and dN/dS are consistent for low-degree haplotypes, suggesting the bias is not critical. Sequence space covering. To quantify how extensively each population explores sequence space, we define sequence space covering C(d) as the average sampled abundance per possible haplotype at distance d where A(d) stands for sequence abundance at distance d. General topological properties Both the degree distribution p(k) and its complementary cumulative distribution function (CCDF) exhibit fat-tailed behaviour, as evidenced by their approximately linear appearance in log-log representations as a function of the degree k (Fig. 3a). Over a sample-dependent range of k, these distributions can be well approximated by a power law of the form , or equivalently , where α is the exponent characterizing the degree distribution. This scaling regime spans a broader interval of k values for Qβ populations, and the corresponding fits are slightly more robust, although SARS-CoV-2 CCDFs also display clear fat-tailed behaviour. Estimated values of α, together with their fitting uncertainties, are reported in Table 1. We observe a significant difference between the two systems: SARS-CoV-2 exhibits lower exponent values, typically in the range 1.7–2.0, whereas Qβ shows consistently higher values, between 2.2 and 2.6. To quantify this difference, we compared the distributions of estimated exponent values using a Welch two-sample t-test, obtaining a highly significant separation between the two viruses (t ≈ 5.7, ), with Qβ exhibiting systematically larger α values than SARS-CoV-2. Topological structure of the LCC of genotype networks. Various topological quantities measured for different viral populations are qualitatively similar for Qβ and SARS-CoV-2. (a) CCDF, (b) assortativity and (c,d) triangles. In (c,d), thin black lines indicate the theoretical bounds on the expected number of triangles as a function of node degree (upper and lower lines) and the thick black line represents expected values if mutations are randomly assigned along the haplotype sequence; dark grey and lighter areas regions correspond to one and two standard deviations from average. The explicit derivation of these quantities has been published previously (Seoane et al. 2026b). As in previous representations, Qβ populations are shown in different shades of blue with circles, upward triangles, or crosses as markers; and SARS-CoV-2 in shades of orange with squares, downward triangles, or plus symbols as markers; as specified in the legend. These differences in α reflect quantitative changes in the reconstructed genotype networks. The lower exponent observed in SARS-CoV-2 reveals that nodes of intermediate degree are more similar in the exploration of their neighbourhood as compared with Qβ haplotypes. However, the fast decaying cut-off of SARS-CoV-2 degree distributions reveals a rapid depletion of high-degree nodes that is consistent with differences in sequencing coverage (see Section 5); different coverage, however, does not explain differences in α, which seem to be characteristic of each virus. While SARS-CoV-2 datasets comprise of the order of sequences, this figure is approximately five times larger for Qβ. Assortativity, quantified through the average nearest-neighbour degree as a function of k, shows an inverse dependence over a broad range of k values in all analysed networks, consistent with a scaling behaviour (see Fig. 3b), with β> 0 indicating a decreasing function of degree. This behaviour is further captured by the assortativity coefficient , defined as the Pearson correlation between the degrees of connected node pairs (Newman, 2002). Both measures, β> 0 and r < 0, indicate that the reconstructed genotype networks are strongly disassortative. In other words, low-degree mutants preferentially link to highly connected haplotypes. However, the distribution in Fig. 3b presents an up-kick for large degree, indicating that hubs also connect to other high-degree nodes. Our genotype networks are strongly disassortative except for this characteristic. These structural features, together with the fat-tailed degree distributions, are consistent with the hierarchical organization observed in Fig. 2. Beyond this global property, we observe a systematic relationship between assortativity and sequence coverage A. Specifically, the fitted values of β show a strong negative correlation with A across all datasets (Pearson correlation r ≈ -0.88, P ≈ .001), indicating that more deeply sampled populations tend to exhibit smaller β values, i.e. weaker degree decay in the nearest-neighbour function. This trend is consistent with previous interpretations (Seoane et al. 2026b): as quasispecies diversity increases, the number of intermediate- and high-degree nodes grows, increasing the probability of connections among them and thus reducing the effective disassortativity measured through β. These results therefore suggest that the change in disassortativity could be a sampling effect. The number of triangles per haplotype in Qβ and SARS-CoV-2 populations shows a pattern consistent with previous observations in other Qβ populations (Seoane et al. 2026b): for large values of the degree k, mutations tend to be randomly distributed along the haplotype’s sequence, and (see Fig. 3c,d). In contrast, at low degree values, significant deviations from site neutrality are observed, with an excess of low-degree haplotypes exhibiting values of △ (i) more than two standard deviations above the expected random value. No significant deviations towards a deficit of triangles are observed. This excess of triangles at low degree suggests the presence of local constraints in sequence space, where low-connectivity haplotypes form neighbourhoods more clustered than expected under a random mutational model, i.e. there are certain positions that admit more mutations than others. Hierarchical organization around the root haplotype Topological measures indicate that all our viral populations are organized in a broadly similar manner. The degree distribution and the disassortative nature of the reconstructed networks, together with representations such as those in Fig. 2, support a hierarchical organization of extant variation and suggest that additional features of the evolutionary process may be revealed by quantities that depend on the distance d to the most abundant haplotype. Figure 4 represents five quantities of interest measured as a function of the Hamming distance d (number of mutations) to the most abundant haplotype (dubbed the root) in each of the six samples. All panels illustrate the exhaustive search performed at distances d=1 and d=2, where almost all possible mutants are sampled at each time step. Since a natural population is two to three orders of magnitude larger than the size of the deep-sequenced sample, it is to be expected that most mutants until distance d=3 are actually present at each passage. Through successive replication cycles and passages, a large fraction of mutants at distance d=4 are likely generated. The exploration of mutants at larger distances from the root decays rapidly as a result of the rapid increase in the number of different haplotypes (Equation (4)), together with the decrease in average abundance. Overall, this rapid decay highlights the predominantly local nature of intra-host exploration in genotype space, where evolutionary dynamics efficiently and exhaustively sample the immediate mutational neighbourhood but face progressively stronger constraints in accessing more distant regions of the sequence landscape (Manrubia et al., 2026). Structure of intra-host genotype networks as a function of distance to the root. Different measures are comparatively explored for the two viruses as a function of Hamming distance to the root sequence (distance level 0): (a) absolute number of nodes (haplotypes); (b) fraction q(d) of haplotypes explored; (c) average dN/dS ratio; (d) average abundance C(d) per possible haplotype and (e) absolute sequence abundance A(d) at each distance d. The dN/dS ratio (Fig. 4c) reflects the balance between synonymous (S) and non-synonymous (N) mutations as a function of distance from the root. This quantity does not account for the abundance of each variant type, but it shows that both synonymous and non-synonymous mutations are exhaustively sampled at d = 1, where the high abundance of the root ensures near-complete coverage of its mutational neighbourhood, largely independent of the fitness effects of first-step mutants. Although the average dN/dS provides a useful summary of the mutational balance, it should be interpreted with caution. Individual haplotypes display a broad distribution of dN/dS values at every distance from the root (Supplementary Fig. 5), indicating that the mean does not represent a characteristic value of the population. Because dN/dS is itself a ratio, its arithmetic mean is intrinsically asymmetric: values above one are not compensated symmetrically by values below one, leading to a systematic upward bias in the average. At larger distances, the dN/dS ratio decreases, and in regions of low abundance the effects of purifying selection become apparent as a relative enrichment of synonymous mutations. This pattern indicates a progressive shift in the balance between mutation and selection across the network. Close to the root, where mutation supply saturates the local neighbourhood, both types of mutations are frequently generated and observed. At larger distances, however, purifying selection increasingly constrains the persistence of non-synonymous variants, leading to a relative depletion of amino-acid-altering mutations. In this regime, selection becomes detectable because mutation supply is no longer sufficient to uniformly populate the available sequence space. These results are consistent with the relationship between degree and triangle abundance: at high degree (corresponding to low distance d from the root), no systematic deviation is observed in either the distribution of triangles or the dN/dS ratio, and both quantities are consistent with random assignment of mutations along the haplotype. At low degree, the dN/dS ratio decreases, indicating a relative enrichment of synonymous mutations, while the number of triangles simultaneously shows significant positive deviations from the random expectation. This establishes a correspondence between a purely topological measure (triangle abundance) and a population-genetic quantity (the ratio of non-synonymous to synonymous mutations). Note that the number of triangles is not affected by the bias discussed above that affects Equation (5). Finally, the sequence space coverage C(d) (Fig. 4d) decays approximately exponentially for all samples, following , where A(0) is the abundance at the root and δ is a characteristic decay length. This exponential decay results from a non-trivial combination of the decay in total sequence abundance with distance d (Fig. 4e) and the slower-than-exponential decrease in haplotype diversity . The values of δ are of the same order of magnitude across populations (Table 1), although a Welch two-sample t-test indicates a statistically detectable difference between viruses (t ≈ 8.9, ), with Qβ exhibiting slightly larger values. Importantly, δ shows no significant dependence on sequence coverage, A, within either viral group. This suggests that this decay scale reflects intrinsic features of local exploration in genotype space rather than sampling depth. Effect of sequencing coverage on inferred quasispecies structure To assess whether the observed differences between Qβ and SARS-CoV-2 reflect intrinsic properties of the underlying quasispecies or are instead driven by sequencing depth, we perform a controlled subsampling analysis. The working hypothesis is that several topological quantities inferred from genotype networks are sensitive to coverage, and that differences between viral systems may be partially reproduced by reducing sampling depth in a high-coverage dataset. Empirically, several observables show a systematic dependence on sequence coverage. For instance, sequence space coverage C(d) exhibits a similar functional form across viruses (Fig. 4d), although the maximum explored distance is reduced in SARS-CoV-2. Analogous trends are observed for the total number of haplotypes at each distance (Fig. 4a) and for the fraction of diversity explored (Fig. 4b). Consistently, the total diversity D is lower in SARS-CoV-2 samples (Table 1), and these quantities correlate with the total number of sequences A. To test whether these patterns can be explained by sampling effects alone, we use a high-coverage Qβ population (30C, amplicon 1) as a reference system and generate artificial datasets by random subsampling of sequences. This procedure preserves the empirical abundance distribution while systematically varying the effective coverage , thereby emulating reduced sequencing depth under controlled conditions. Figure 5a shows the complementary cumulative degree distributions (CCDFs) obtained under the same subsampling scheme. The results demonstrate a systematic modification of the degree distribution at low coverage, where the apparent power-law regime is progressively shortened, developing a cut-off consistent with a reduction of the degree in high-degree nodes. For values of comparable with those of SARS-CoV-2 datasets, the resulting CCDF closely reproduces the empirical curves in Fig. 3a, indicating that differences in the inferred tail behaviour can arise naturally from incomplete sampling. Effect of sequencing coverage on inferred genotype network structure. (a) CCDF for varying sequence coverage. (b) Number of nodes (haplotypes) as a function of Hamming distance from the root sequence (distance level 0) for different levels of sequence coverage. Data are obtained by random subsampling of the Qβ (amplicon 1) population at 30C. If sample size is smaller than , each curve corresponds to averages over 100 independent realizations of the subsampling and network reconstruction procedure. Figure 5b shows the number of haplotypes detected at each distance d for different subsample sizes . For each , sequences are drawn at random according to their observed abundances, and the corresponding LCC is reconstructed as described above. Each curve corresponds to an average over 100 independent realizations of the subsampling and reconstruction procedure. As coverage decreases, the inferred exploration of sequence space becomes increasingly restricted, reproducing the qualitative behaviour observed in SARS-CoV-2 datasets of lower coverage (Fig. 4a). Overall, these results support the interpretation that several apparent differences between viral populations are, at least in part, attributable to coverage effects rather than fundamental differences in quasispecies organization. In particular, subsampling reproduces key trends in diversity, connectivity, and degree distributions, highlighting the importance of accounting for sequencing depth when comparing genotype network structure across datasets. We note that the effects of amplification bias and early-cycle PCR bottlenecks, while not explicitly modelled here, are expected to act similarly to reduced effective coverage on the reconstructed genotype network, and are thus partly captured by the subsampling analysis above. Discussion This study compares the structure of genotype networks in two RNA viruses evolving under fundamentally different conditions: bacteriophage Qβ in controlled laboratory environments and SARS-CoV-2 within infected human hosts. The hierarchical structure of viral quasispecies, previously described for Qβ (Seoane et al. 2026b), is here shown to extend to SARS-CoV-2 populations evolving in vivo. Despite large differences in genome size, mutation rate, and ecological context, both systems display a remarkably similar organization of their mutant spectra. Importantly, the consistency of all main topological measures across independent amplicons (see Figures 1 to 4 in Supplementary Information) supports the robustness of the inferred network properties and indicates that the reported differences between SARS-CoV-2 and Qβ are not driven by local genomic context. Measures broadly applied to characterize complex networks (Newman, 2010), such as the degree distribution and the number of triangles (a measure of clustering), are quantitatively comparable across all populations analysed for both viruses. This consistency further supports the existence of a shared network architecture underlying the mutant spectra of Qβ and SARS-CoV-2 Populations show some variation in their assortativity, although this quantity is expected to depend on the fraction of the underlying neutral network explored (hence on population diversity) (Seoane et al. 2026b), and wide variation has been reported in genotype networks of different origins (Aguirre et al., 2011; Aguilar-Rodríguez et al., 2018; Williams et al., 2022). The organization of these viral populations is characterized by a hierarchical genotype network centred around a dominant haplotype and surrounded by layers of variants at increasing Hamming distance. The dense occupancy observed in the immediate neighbourhood of the dominant haplotype is consistent with near-exhaustive local exploration of sequence space. Given the high mutation supply in RNA viruses, most single-mutant variants are expected to be generated repeatedly (Lauring and Andino, 2010; Domingo et al., 2012; Sender et al., 2021), implying that the presence of a variant does not necessarily indicate neutrality or selective advantage, but may instead reflect the stochastic generation of mutations in highly abundant genetic backgrounds. Apparent differences between the two viruses can be largely attributed to sampling effects, as supported by subsampling experiments in Qβ: lowering coverage yields sequence ensembles that reproduce the topological and hierarchical features observed in SARS-CoV-2 datasets. The emergence of comparable network architectures suggests that the global organization of quasispecies is neither fundamentally determined by the absolute number of mutations per genome, nor by environmental differences, but by more general constraints governing the exploration of sequence space. Beyond minor quantitative differences, the internal organization of genotype networks appears highly conserved and can be understood as the outcome of three general processes. First, mutation continuously generates a dense cloud of single-mutant neighbours around the dominant haplotype through a process akin to diffusion. Second, the number of possible genotypes increases almost exponentially with Hamming distance, leading to progressively sparser sampling of sequence space. This fact notwithstanding, the local sampling is so exhaustive that we estimate that all mutants up to distance 3 are likely present in the complete quasispecies. Third, purifying selection increasingly limits the persistence of variants carrying multiple mutations. Together, these processes naturally produce a hierarchical structure and a highly heterogeneous exploration of sequence space, a mechanism that may represent a generic feature of RNA virus populations and, more generally, of ensembles of replicators (Manrubia et al., 2026). This distance-dependent regime reconciles two key observations. In the immediate neighbourhood of the dominant haplotype, mutational space is densely and nearly exhaustively explored, and even deleterious mutations can be repeatedly generated and transiently observed. At larger mutational distances, exploration becomes progressively sparser and selection increasingly determines which variants persist. This shift in the balance between mutation and selection is consistent with the observed decay in dN/dS with distance, as well as with changes in local network structure (noted by ). These results support a view of viral evolution as a continuous, out-of-equilibrium process in which populations expand locally in sequence space through mutation-driven diffusion, while selection shapes the distribution of variant abundances. In this framework, exploration does not rely primarily on neutral pathways but emerges from the high connectivity of genotype space and the constant generation of variants in the immediate neighbourhood of highly abundant sequences. The hierarchical organization of genotype networks and their persistence, even in samples where mutations have recently been fixed, also has important implications for viral transmission and adaptation. Transmission bottlenecks are expected to sample only a small subset of the mutant spectrum, typically variants close to the dominant haplotype. However, once replication resumes in the new host, the high mutation rate characteristic of RNA viruses should rapidly regenerate a cloud of mutants around the transmitted genome. Our results strongly point at hierarchical genotype networks acting as structural attractors, which are repeatedly reconstructed during viral evolution, with selective pressures determining which haplotype occupies the central position while the overall geometry of the mutant cloud remains largely invariant. More generally, genotype networks capture aspects of viral population structure that are not evident from diversity measures alone. While previous studies have reported differences in mutant spectrum complexity across clinical categories (Martínez-González et al., 2022), such differences do not necessarily translate into qualitative changes in global network organization. Rather, diversity and network structure provide complementary information: the former reflects the distribution of variant frequencies, whereas the latter encodes their mutational connectivity and is comparatively robust to sampling effects. Taken together, the existence of a shared genotype network architecture across viruses with distinct life histories points to quasispecies organization being governed by general principles. This convergence is particularly striking given the profound differences in the evolutionary processes shaping each system. In the case of SARS-CoV-2, viral populations evolve within individual hosts under heterogeneous and fluctuating selective pressures, including innate and adaptive immune responses, tissue compartmentalization, and genetic backgrounds shaped by prior transmission events. By contrast, Qβ evolves in a highly controlled in vitro setting, where adaptation proceeds through serial passages imposing recurrent population bottlenecks and where the dominant selective pressure is externally defined. Despite these contrasting demographic, ecological, and selective regimes, both viruses give rise to genotype networks with the same hierarchical architecture. This observation suggests that while selective pressures determine which haplotype becomes dominant in a given environment, they do not substantially alter the overall organization of the mutational neighbourhood that emerges around that haplotype. Testing this hypothesis along different SARS-CoV-2 waves, across additional RNA viruses, and extending it to DNA viruses with lower mutation rates, will help determine the extent to which hierarchical genotype networks represent a universal organizing principle of viral evolution. Supplementary Material Contributor Information Samuel Martínez-Alcalá, Department of Biodiversity and Evolutionary Biology, Museo Nacional de Ciencias Naturales (CSIC), c/ José Gutiérrez Abascal 2, E-28006 Madrid, Spain; Grupo Interdisciplinar de Sistemas Complejos (GISC), E-28911 Madrid, Spain. Iker Atienza-Diez, Department of Biodiversity and Evolutionary Biology, Museo Nacional de Ciencias Naturales (CSIC), c/ José Gutiérrez Abascal 2, E-28006 Madrid, Spain; Grupo Interdisciplinar de Sistemas Complejos (GISC), E-28911 Madrid, Spain. Pilar Somovilla, Centro de Astrobiología (CAB), CSIC-INTA, Carretera de Ajalvir Km 4, E-28850 Torrejón de Ardoz, Madrid, Spain. Brenda Martínez-González, Microbes in Health and Welfare Program, Centro de Biología Molecular Severo Ochoa (CSIC-UAM), C/ Nicolás Cabrera, 1, Campus de Cantoblanco, E-28049 Madrid, Spain; Department of Clinical Microbiology, Instituto de Investigación Sanitaria-Fundación Jiménez Díaz University Hospital (UAM), Av. de los Reyes Católicos, 2, E-28040 Madrid, Spain. Celia Perales, Microbes in Health and Welfare Program, Centro de Biología Molecular Severo Ochoa (CSIC-UAM), C/ Nicolás Cabrera, 1, Campus de Cantoblanco, E-28049 Madrid, Spain; Department of Clinical Microbiology, Instituto de Investigación Sanitaria-Fundación Jiménez Díaz University Hospital (UAM), Av. de los Reyes Católicos, 2, E-28040 Madrid, Spain; Centro de Investigación Biomédica en Red de Enfermedades Infecciosas (CIBERINFEC), C/ de Melchor Fernández Almagro, 3, E- 28029 Madrid, Spain. Luis F Seoane, BIT Lab, Complexity of Life program, Institut de Biologia Evolutiva (CSIC-UPF), Pg. Marítim de la Barceloneta, 37, 08003 Barcelona, Spain; Theoretical Sciences Visiting Program, Okinawa Institute of Science and Technology Graduate University, Onna 904-0495, Japan. Ester Lázaro, Centro de Astrobiología (CAB), CSIC-INTA, Carretera de Ajalvir Km 4, E-28850 Torrejón de Ardoz, Madrid, Spain. Susanna Manrubia, Department of Biodiversity and Evolutionary Biology, Museo Nacional de Ciencias Naturales (CSIC), c/ José Gutiérrez Abascal 2, E-28006 Madrid, Spain; Grupo Interdisciplinar de Sistemas Complejos (GISC), E-28911 Madrid, Spain. Conflicts of interest None declared. Funding This research was supported through grants PID2023-153225NA-I00 (L.F.S.), PID2023-146622OB-I00 (B.M.-G. and C.P.), PID2023-147963NB-C21 (S.M.-A., I.A., and S.M.), and PID2023-147963NB-C22 (P.S. and E.L.) funded by MICIU/AEI/10.13039/501100011033 and by European Regional Development Fund, European Union. L.F.S. received funding from the Occident Foundation (Grant No. FJSCNB-2022-12-B) and developed part of this research while visiting the Okinawa Institute of Science and Technology (OIST) through the Theoretical Sciences Visiting Program (TSVP). B.M.-G. and C.P. were supported through the ‘Centro de Excelencia Severo Ochoa’ award to the Centro de Biología Molecular Severo Ochoa (CBM) (Grant No. CEX2021-001154-S); project TEC-2024/BIO-66 (SALAINDEC-CM from Comunidad de Madrid/FEDER), and the European Commission-Next Generation EU (regulation EU 2020/2024) through the Consejo Superior de Investigaciones Científicas (CSIC’s) Global Health Platform [Plataforma Temática Interdisciplinar (PTI) Salud Global]. Institutional grant from the Fundación Ramón Areces Foundation to the CBM is also acknowledged. Data availability The FASTQ files derived from the SARS-CoV-2 patient samples are publicly available in the European Nucleotide Archive under project accession PRJEB48766. The Qβ sequencing datasets analysed in this study were generated and described in previous studies (Somovilla et al., 2022; Seoane et al. 2026b) and are available upon request. The custom PivoNet code used to reconstruct genotype networks from FASTA files is publicly available through the GitLab repository cited in Seoane (2026a). References Associated Data Supplementary Materials Data Availability Statement The FASTQ files derived from the SARS-CoV-2 patient samples are publicly available in the European Nucleotide Archive under project accession PRJEB48766. The Qβ sequencing datasets analysed in this study were generated and described in previous studies (Somovilla et al., 2022; Seoane et al. 2026b) and are available upon request. The custom PivoNet code used to reconstruct genotype networks from FASTA files is publicly available through the GitLab repository cited in Seoane (2026a).