Dynamic regulation of HIV-1 capsid interaction with the restriction factor TRIM5α identified by magic-angle spinning NMR and molecular dynamics simulations

There is no author summary for this article yet. Authors can add summaries to their articles on ScienceOpen to make them more accessible to a non-specialist audience.

Abstract

The host factor protein TRIM5α plays an important role in restricting the host range of HIV-1, interfering with the integrity of the HIV-1 capsid. TRIM5 triggers an antiviral innate immune response by functioning as a capsid pattern recognition receptor, although the precise mechanism by which the restriction is imposed is not completely understood. Here we used an integrated magic-angle spinning nuclear magnetic resonance and molecular dynamics simulations approach to characterize, at atomic resolution, the dynamics of the capsid’s hexameric and pentameric building blocks, and the interactions with TRIM5α in the assembled capsid. Our data indicate that assemblies in the presence of the pentameric subunits are more rigid on the microsecond to millisecond timescales than tubes containing only hexamers. This feature may be of key importance for controlling the capsid’s morphology and stability. In addition, we found that TRIM5α binding to capsid induces global rigidification and perturbs key intermolecular interfaces essential for higher-order capsid assembly, with structural and dynamic changes occurring throughout the entire CA polypeptide chain in the assembly, rather than being limited to a specific protein-protein interface. Taken together, our results suggest that TRIM5α uses several mechanisms to destabilize the capsid lattice, ultimately inducing its disassembly. Our findings add to a growing body of work indicating that dynamic allostery plays a pivotal role in capsid assembly and HIV-1 infectivity.

Related collections

Most cited references 29

Record: found
Abstract: found
Article: not found

Crystal structure of human cyclophilin A bound to the amino-terminal domain of HIV-1 capsid.

D K Worthylake, Wesley Sundquist, Jennifer R Gamble … (1996)

The HIV-1 capsid protein forms the conical core structure at the center of the mature virion. Capsid also binds the human peptidyl prolyl isomerase, cyclophilin A, thereby packaging the enzyme into the virion. Cyclophilin A subsequently performs an essential function in HIV-1 replication, possibly helping to disassemble the capsid core upon infection. We report the 2.36 A crystal structure of the N-terminal domain of HIV-1 capsid (residues 1-151) in complex with human cyclophilin A. A single exposed capsid loop (residues 85-93) binds in the enzyme's active site, and Pro-90 adopts an unprecedented trans conformation. The structure suggests how cyclophilin A can act as a sequence-specific binding protein and a nonspecific prolyl isomerase. In the crystal lattice, capsid molecules assemble into continuous planar strips. Side by side association of these strips may allow capsid to form the surface of the viral core. Cyclophilin A could then function by weakening the association between capsid strips, thereby promoting disassembly of the viral core.

0 comments Cited 161 times – based on 0 reviews      Review now

Bookmark

Record: found
Abstract: found
Article: not found

Assembly and analysis of conical models for the HIV-1 core.

Andrew J. Finch, Barbie Ganser-Pornillos, Sheng-gong Li … (1999)

The genome of the human immunodeficiency virus (HIV) is packaged within an unusual conical core particle located at the center of the infectious virion. The core is composed of a complex of the NC (nucleocapsid) protein and genomic RNA, surrounded by a shell of the CA (capsid) protein. A method was developed for assembling cones in vitro using pure recombinant HIV-1 CA-NC fusion proteins and RNA templates. These synthetic cores are capped at both ends and appear similar in size and morphology to authentic viral cores. It is proposed that both viral and synthetic cores are organized on conical hexagonal lattices, which by Euler's theorem requires quantization of their cone angles. Electron microscopic analyses revealed that the cone angles of synthetic cores were indeed quantized into the five allowed angles. The viral core and most synthetic cones exhibited cone angles of approximately 19 degrees (the narrowest of the allowed angles). These observations suggest that the core of HIV is organized on the principles of a fullerene cone, in analogy to structures recently observed for elemental carbon.

0 comments Cited 149 times – based on 0 reviews      Review now

Bookmark

Record: found
Abstract: found
Article: found

Is Open Access

Extreme Genetic Fragility of the HIV-1 Capsid

Suzannah J. Rihn, Sam J. Wilson, Nick Loman … (2013)

Introduction Genetic robustness is defined as the ability of a biological entity (e.g. a protein or organism) to maintain function in the face of mutations [1], [2]. More robust proteins or organisms tolerate higher mutation rates while less robust (more ‘brittle’ or ‘fragile’) proteins or organisms are intolerant of mutation and are more likely to lose function or be driven to extinction by high mutation rates. Viruses that replicate via RNA intermediates using non-proofreading polymerases exhibit high mutation rates, suggesting that robustness should be particularly advantageous to them [3]. Indeed, under some conditions, viral populations that exhibit high robustness at the expense of high fitness might be favored over those that have high fitness but low robustness [4]–[6]. In other words, robustness/fragility might be a more potent selective force than fitness under some circumstances. While the robustness or fragility of RNA viruses has been investigated in several studies [4]–[9], most reports characterize robustness by treating an entire viral genome as a single biological entity. However, viruses encode a variety of proteins that execute a range of functions to enable replication, and the genetic robustness of individual proteins is expected to vary within a given virus. Even within a single protein, individual domains may exhibit variation in robustness/fragility. Proteins that perform complex or multiple functions that are highly dependent on accurate structure (e.g. enzymes), should tend to be more genetically fragile, i.e less tolerant of mutation, and exhibit greater sequence conservation than those that do not (e.g. proteins that simply provide peptide binding sites to recruit other proteins). However, in their natural setting, i.e. a susceptible host, animal viruses often replicate in a hostile and changing environment, shaped partly by adaptive immune responses. Such immune responses can take the form of antibodies that target proteins displayed on the surface of the virion, or cytotoxic lymphocytes that can target epitopes in nearly any viral protein. Hence, at least some viral proteins, even those that should exhibit genetic fragility, are placed under strong evolutionary pressure to diversify in sequence, because they are targeted by adaptive immune responses. Under such conditions, otherwise fragile proteins might be expected to evolve higher robustness. However, the potential trade-off between robustness and fitness might impinge on each property, and it is unclear whether simple fitness, or the acquisition of robustness, would constitute the dominant selective force on a given viral protein in a natural setting. One such protein in which the competing needs to preserve function yet diversify sequence might conflict in a particularly acute manner, is human immunodeficiency virus type I (HIV-1) capsid protein (CA). As a critically important viral protein, it is placed under strong selective pressure to maintain its structure and perform several functions. Conversely, as a highly expressed, immunologically visible protein, CA also experiences a competing pressure to diversify its sequence. CA exists both as a domain within the Gag precursor polyprotein during particle assembly, and as an autonomous protein in the mature virion. Both Gag and CA are multifunctional proteins [10]–[13]. Specifically, during virion morphogenesis, Gag molecules assemble at the plasma membrane and drive the formation of roughly spherical immature virions, containing radially arrayed Gag molecules, that bud through the plasma membrane [14], [15]. However, during and after viral budding, the viral protease (PR) is activated and catalyses the cleavage of Gag at five positions, causing profound morphological transformations [13]. In particular, the liberated CA protein forms a conical capsid that encapsulates the nucleocapsid-genomic RNA complex [16], [17]. Like all retroviral CA proteins, a single HIV-1 CA molecule is composed of two domains: the N-terminal domain (NTD) comprised of 146 amino acids, and C-terminal domain (CTD) comprised of 85 amino acids. The NTD structure consists of an N-terminal β hairpin and 7 succeeding α helices, while the CTD has 4 α helices, and a C-terminal unstructured region of 11 residues [18]–[20]. The NTD and CTD are joined by an interdomain linker region (residues 146–150). Other noteworthy features of CA include the major homology region (MHR residues 153–172) of the CTD, which is a highly conserved 20 amino-acid region found in all retroviruses [21] and the likely binding site of ABCE1 [22], and a loop (NTD residues 85–93) which binds the cellular protein cylophilin A [23]. The mature HIV-1 capsid consists of ∼1100 CA monomers assembled into a hexameric lattice with 12 pentameric declinations [16], [24], that are distributed in such a way that the viral capsid takes the form of a fullerene cone. Within the lattice, the NTD is primarily responsible for intra-hexamer contacts, while the CTDs form dimers that link adjacent hexamers [25], [26]. However, interactions between the NTD and CTD also contribute to proper capsid formation [26]–[29]. In addition to its major structural role, CA is a key determinant of several other biological properties of HIV-1. For example, HIV-1 capsid is the key determinant that enables HIV-1 to infect non-dividing cells [30], [31]. Related work has identified genetic or physical interactions with host proteins karyopherin β transportin-3 (TNPO3), nucleoporin 153 (NUP 153), nucleoporin 358 (NUP358)/RanBP2 and cleavage and polyadenylation factor 6 (CPSF6) [32]–[36]. Moreover, interactions between HIV-1 capsid and cyclophilin A influence nuclear import and subsequent integration site selectivity, as well as replication efficiency, in a cell type dependent manner [37]–[40]. CA is a key target of intrinsic, innate and adaptive immune defenses. Specifically, CA is targeted by TRIM5α [41], and may be detected by undefined sensors in dendritic cells [42]. HIV-1 CA also adapts under immune selection in vivo [43]–[46]. In particular, host CD8+ cytotoxic T lymphocyte (CTL) responses to HIV-1 infection are critical determinants of viral control [47], [48], and there is an association between Gag- and capsid-specific CD8+ T-cell responses and in vivo viral burden[49]. The emergence of viral ‘escape’ mutations in CA can result in higher viremia [50], [51]. However, escape mutations often incur a significant fitness cost, which may be subsequently compensated for by secondary mutations in CA, or drive reversion when immune pressure is lifted [45], [51], [52]. Previous studies in which CA was mutated have aimed to elucidate the importance of particular domains, regions, and residues in CA functions. Those targeted mutagenesis studies largely relied upon insertion [53], deletion [54], or alanine or proline scanning [55], [56]. Here, we took a different approach to investigate the genetic robustness of HIV-1 CA. Specifically, we describe the generation of a large randomly mutagenized library of CA sequences to simulate the natural process of random mutation that occurs during HIV-1 replication. Strikingly, we find that CA is extremely intolerant of nonsynonymous substitutions, with ∼70% of random single nucleotide substitutions leading to a >50-fold reduction in replicative fitness. We also determined the biological basis for this extreme genetic fragility and found that requirements imposed by the need to accurately and efficiently assemble a mature virion are largely responsible. Indeed, a subset of mutants were temperature sensitive (ts), and the conditional non-viability of these mutants was always manifested during the formation of virions. Interestingly, fewer than half of the CA mutations that might be expected to occur in vivo, based on their high replicative fitness, were actually observed in natural populations. This finding suggests that CA sequence is constrained in vivo, not only by the need to maintain replicative fitness, but also by other unknown selective pressures. The mutational fragility of HIV-1 CA demonstrated herein is consistent with the relatively high overall degree of amino acid sequence conservation observed in natural populations, and potentially explains the apparently limited capacity of HIV-1 to evade immune responses directed against epitopes in CA. Results Experimental design and analysis To construct a library of random mutants of HIV-1 CA, a low fidelity PCR approach was used, as illustrated in Figure 1A. In this scheme, CA encoding sequences were amplified by error-prone PCR and cloned using a TOPO TA cloning Kit to generate a library with an estimated complexity of 1×104 clones. Plasmid DNA extracted from the pooled library was then digested, and CA sequences inserted into a replication competent proviral clone encoding EGFP in place of Nef (pNHGcapNM accession:JQ686832, referred to as WT or parental virus hereafter) using unique NotI and MluI restriction sites, introduced by silent mutagenesis into sequences flanking those encoding CA (Figure 1A). Thereafter, nearly 1000 individual colonies were picked and proviral plasmid DNA was extracted from individual mini-cultures. The mutagenized CA sequence was then sequenced for each clone, and after failed sequencing reactions or chromatograms indicating the presence of more than one template were removed, the resulting library consisted of 680 arrayed proviral plasmids. The distribution of all nucleotide changes (Figure 1B), missense mutations (Figure 1C), and nonsense and other mutations (Figure 1D) within the library was then determined. The PCR-mutagenesis conditions were optimized in pilot experiments so as to reduce the representation of non-mutated and heavily mutated CA sequences in the library. Consequently, the most frequent mutation in the library was a single nucleotide or amino acid substitution (Figure 1B, C, D). 10.1371/journal.ppat.1003461.g001 Figure 1 Generation and characterization of the CA mutant library. (A) Schematic representation of the replication competent HIV-1 proviral plasmid pNHGcapNM used in this study. NotI and MluI restriction sites flank a region of Gag that includes nearly all of CA, and a few amino acids in MA. This 717 bp sequence was subjected to Genemorph II mutagenesis and subcloned into pNHGcapNM. (B) Distribution of the number of nucleotide changes in each clone for the resulting library (this includes sequences that were represented more than once). (C) Distribution of the number of amino acid changes in each clone in the library (this includes sequences that were represented more than once, nonsense mutations, mutations in MA and frameshifts). (D) Pie chart describing all of the resulting single amino acid changes. SINGLE clones were those included in further studies of the library, and the remaining clones were discarded for the following reasons: STOP (mutation resulted in a stop codon), MA (contained a mutation in the small fragment of MA included in the amplicon), FS (frameshift), DUP (duplicate mutation of clone already in library). Viability of a large panel of HIV-1 CA mutants To measure the genetic robustness of CA, we analyzed the ability of the 680 mutants to replicate. Specifically, replicative fitness was measured by means of a spreading infection assay, in which MT-4 cells were challenged with a low volume of virus-containing supernatant derived from transfected 293T cells (equivalent to an MOI of ∼0.01 for the WT virus). Multiple rounds of the virus replication cycle were allowed to occur before halting the experiment when the WT virus had infected >50% of cells (72 to 80 hours later). Replicative fitness for each CA mutant was expressed as number of cells that were infected (EGFP-positive), as a percentage of the number of cells that were infected by the WT virus (Figure 2A, B). Additionally, to allow for the isolation of temperature sensitive mutants that could facilitate the subsequent characterization of the replication defects associated with CA mutations (see below), this viability screen was carried out at two temperatures, 35°C (Figure 2A) and 39.5°C (Figure 2B). 10.1371/journal.ppat.1003461.g002 Figure 2 Fitness effects of mutations in the CA mutant library. (A) Viability of all CA mutant clones at 35°C (this excludes nonsense mutations, duplicates, frameshifts, MA mutations), demonstrating how the number of amino acid mutations in CA affects fitness (Replication is plotted as the fraction of MT-4 cells that are GFP+ (infected), as a percentage of the fraction infected by WT virus following a spreading replication assay). (B) Viability of all CA mutant clones at 39.5°C (this excludes nonsense mutations, duplicates, frameshifts, MA mutations), measured as in (A). (C) Distribution of mutational fitness effects (DMFE) for all single amino acid CA substitution mutants at 35°C. (D) DMFE for all single amino acid CA substitution mutants at 39.5°C. Virtually all of the viral clones containing WT CA sequence, or only silent mutations, replicated at levels close to that of the starting construct. This indicated that the frequency with which defective viruses were generated by cloning artifacts or silent mutations was extremely low (Figure 2A, B). Moreover, this analysis revealed CA to be extremely fragile, or intolerant of randomly introduced amino acid substitutions (Figure 2A, B). If the cut-off for being viable is arbitrarily set at 2% of the parental virus replicative fitness, only 35% and 28% of the 135 clones that encoded single amino acid substitutions were viable at 35°C and 39.5°C respectively. These figures could overestimate the proportion of mutants that might be expected to be fit in vivo, as requirements for replication in an infected individual might be more stringent than in highly permissive MT-4 cells, and 2% of WT virus fitness in MT-4 cells is a rather generous cut-off figure for a designation of ‘viable’. Increases in the number of amino acid substitutions further decreased the frequency of viable mutants; only 6% and 3% of the 125 clones containing two random amino-acid substitutions were viable at 35°C and 39.5°C respectively. Only 1 of 73 clones containing 3 random amino acid substitutions was viable, and none of the 45 library clones with four or five random substitutions could replicate at all. Analysis of single amino acid CA mutants In total, the subset of mutants that had a unique single amino acid substitution within CA (135 mutants) covered 102 (44%) of the 231 CA residues (for 33 CA residues there was more than one unique mutation at each position). To refine our estimate of CA robustness, two more assays were conducted at a natural temperature of 37°C, using only the panel of CA mutants that had unique single amino acid substitutions. First, in the MT-4 spreading assay, 40 mutant viruses (30%) exhibited at least 2% of parental virus fitness, and are listed in Table 1. Conversely, Table 2 contains the longer list of 95 mutants that had less than 2% of WT fitness in the spreading infectivity assay, and are considered non-viable. Single-cycle infectivity was also measured by transfecting 293T cells with proviral plasmids and measuring infectivity using a larger dose of virus (equivalent to an MOI ∼1 for the parental clone) and treating the MT-4 cells with dextran sulfate 16 h post-infection to limit replication to a single-cycle. Replicative fitness in spreading versus single-cycle assays was well correlated (Table 1), although most mutant viruses fared better in the single-cycle assay than in the spreading assay. This phenomenon possibly reflects the effect of transfection in the single-cycle assay, whereby overexpression of the viral proteins may suppress defects, as opposed to the spreading infection assay where more natural levels of viral gene expression are attained. Alternatively, the multiple rounds of replication in the spreading assay might amplify the effect of a defect in the single-cycle assay. 10.1371/journal.ppat.1003461.t001 Table 1 Fitness measurements for viable HIV-1 CA mutants. HIV CA mutant Location in CA Single cycle infectivity (% of WT)a Spreading fitness (% of WT)b Frequency in subtype B isolates (%)c I2L β-strand 90 66 0.1 N5D β-strand 20 5 0.2 I6T β-strand 52 57 0 M10I β-strand 19 3 0.1 M10L β-strand 74 27 0 M10V β-strand 11 16 0.1 Q13H β-strand 27 6 4.2 A14S Loop 77 81 2.9 A14T Loop 98 87 1.2 I15V Loop 32 14 0.3 N21S Helix 1 56 47 0 S33C Loop 40 48 0 Q50H Helix 3 20 3 5.3 E75D Helix 4 41 22 5.8 V83M Helix 4 75 65 2.6 H87Q Cyclophilin b.l.* 120 92 13.7 H87R Cyclophilin b.l. 28 2.5 0.2 I91T Cyclophilin b.l. 85 77 0 I91V Cyclophilin b.l. 100 82 41.3 M96I Loop 120 92 11.6 E98D Loop 101 63 5.1 R100S Loop 54 43 0 S146C Loop 44 34 0 T148I Loop 66 37 0 S149C Loop 65 74 0 S149G Loop 71 30 0.2 I150V Loop 19 3 0.2 I153T Loop 21 5 0.4 Y164F Helix 8 64 31 0 R167Q Helix 8 11 6 0 A177S Loop 99 37 3.0 M185I Helix 9 24 3 0 E187V Helix 9 60 47 0 L190M Loop 38 33 0.1 T200S Helix 10 104 90 1.6 A204G Helix 10 77 89 8.7 G208E Loop 15 4 0 A209T Loop 108 94 0 A209V Loop 72 84 0.1 T216A Helix 11 72 38 3.0 a Infectivity measurement in which MT-4 cells were inoculated with supernatant from 293T cells (equivalent to MOI = 1 for WT virus) that had been transfected with single residue CA mutant proviral plasmids. Dextran sulfate was added approximately 16 h after infection to restrict replication to a single cycle. Values are reported as the percentage of the number of infected (GFP+) cells obtained with the WT virus. b Fitness measurement in which MT-4 cells were inoculated with supernatant from 293T cells (equivalent to MOI = 0.01 for WT virus) that had been transfected with single residue CA mutant proviral plasmids. Multiple cycles of replication were permitted over 72 h. Values are reported as the percentage of the number of infected (GFP+) cells obtained with the WT virus. c The frequency at which the indicated mutant residue occurs in 1000 HIV-1 subtype B sequences. * Abbreviation for cyclophilin binding loop. Mutants were considered viable if they maintained or 2% of WT fitness in the spreading replication assay. 10.1371/journal.ppat.1003461.t002 Table 2 Non-viable HIV-1 CA mutants. CA mutants in NTD Location in CA CA mutants in CTD Location in CA I2T β-strand I153M Loop H12R β-strand E159V Loop Q13R β-strand F161C Helix 8 S16P Loop D163G Helix 8 R18G Helix 1 V165A Helix 8 N21Y Helix 1 F168S Helix 8 K25I Helix 1 R173I Helix 8 V26E Helix 1 A174G Loop K30E Loop K182Q Helix 9 K30N Loop W184L Helix 9 F32L Loop W184R Helix 9 F32S Loop T186A Helix 9 S33I Loop L190S Loop P34S Helix 2 A194V Loop V36A Helix 2 N195S Loop M39T Helix 2 I201N Helix 10 M39V Helix 2 L211I Helix 11 F40L Helix 2 E212D Helix 11 F40Y Helix 2 M214L Helix 11 L43I Helix 2 M215V Helix 11 G46R Loop A217V Helix 11 P49L Helix 3 Q219P Loop L52F Helix 3 G220V Loop N57S Helix 3 V221A Loop V59A Loop K227I Loop V59M Loop K227N Loop G60W Loop K227R Loop Q63R Helix 4 A65V Helix 4 M66I Helix 4 M66V Helix 4 M68L Helix 4 M68T Helix 4 L69I Helix 4 I73N Helix 4 A77D Helix 4 W80R Helix 4 P90T Cyclophilin bl Q95L Loop Q95R Loop M96T Loop P99A Loop P99L Loop R100W Loop S102R Helix 5 A105G Helix 5 T110I Loop L111F Helix 6 E113V Helix 6 I115K Helix 6 I115L Helix 6 I115T Helix 6 M118L Helix 6 M118V Helix 6 H120P Loop N121I Loop I124N Loop V126A Helix 7 I129M Helix 7 I129T Helix 7 Y130C Helix 7 Y130H Helix 7 R132G Helix 7 G137E Helix 7 L138I Helix 7 M144V Helix 7 L151I Loop L151Q Loop Mutants were considered non-viable if they exhibited 50 Å2. The corresponding fitness measurement for each mutant is plotted. 10.1371/journal.ppat.1003461.t003 Table 3 Summary of CA mutant viability by CA region. Region Number of viable mutants Number of non-viable mutants % Viable mutants β-strand 7 3 70 Helix 1 1 4 20 Helix 2 0 7 0 Helix 3 1 3 25 Helix 4 2 10 17 Cyclophilin b.l. 4 1 80 Helix 5 0 2 0 Helix 6 0 7 0 Helix 7 0 9 0 Helix 8 2 5 29 Helix 9 2 4 33 Helix 10 2 1 67 Helix 11 1 5 17 MHR 3 6 33 Interdomain linker 5 0 100 Helices (all) 11 57 16 Loops (all) 22 35 39 NTD 27 68 28 CTD 13 27 33 When displayed in the context of a capsid hexamer structure (PDB: 3GV2, Figures 3B, C), amino acids corresponding to viable mutants (in green, Figure 3B) appeared to occur preferentially in surface exposed residues. Conversely, mutations in the interior of the CA structure almost always generated non-viable mutants (displayed in red, Figure 3C). This finding is particularly evident when the CA hexamer is viewed from the point of view of interior of the assembled capsid (Figure 3B,C, rightmost diagrams), and reinforces the impression of an inner CA structure, or ‘core,’ composed of helices that are particularly sensitive to mutation. Analysis of the solvent accessible surface area for individual mutated residues confirmed these findings. Specifically, mutated residues with solvent accessible surface areas of less than 50 Å2 (an appropriate cutoff for surface exposure based on previous mutagenesis analyses [57]) caused significantly greater fitness defects (p = 0.003). Viruses with such mutations and exhibited a mean fitness value of 2.8% of WT. Conversely, viruses harboring mutations of more surface exposed amino acids (solvent accessible areas of greater than 50 Å2) exhibited a mean fitness value of 18% of WT. Biological basis for genetic fragility in HIV-1 CA As outlined in the introduction, CA performs a number of functions in HIV-1 replication, including mediating the assembly of immature virions (in the context of the Gag precursor) and formation of a mature conical capsid. Moreover, a capsid of optimal stability is thought to be important during the uncoating step of the HIV-1 cycle in newly infected cells, and CA is also key for the import of the viral genome into the nucleus of newly infected cells. To determine which of these functions, if any, contributed the genetic fragility of HIV-1 CA, we performed a number of assays to elucidate the nature of the replication defects in constitutively or conditionally non-viable CA mutants. To facilitate the elucidation of the nature of the replication defects in CA mutants, we first focused on the subset of conditionally non-viable, ts mutants. At outlined in Figure 2, initial determinations of mutant fitness were carried out at 35°C and at 39.5°C. While most CA mutants were approximately equally fit or unfit at both temperatures, a few exhibited a ts phenotype that was quite large (Figure S1). Some of the ts mutants (circled in Figure S1 and listed in the materials and methods) included double mutants. We therefore generated CA mutants that encoded each single amino acid substitution in isolation. Ultimately, this yielded eleven single amino-acid CA mutants with substantial ts replication phenotypes. Because the initial screen measured replicative fitness over multiple rounds of replication, it could not determine what specific step in the viral life cycle was impaired in the constitutively or conditionally non-viable mutants. We therefore determined the single-cycle infectivity of the ts CA mutants using virions generated in 293T cells at 35, 37, or 39.5°C, which were then used to infect MT-4 cells at 35, 37, or 39.5°C (Figure 4A, B). Importantly, the WT control, pNHGcapNM, maintained a consistent viral titer regardless of production and infection temperatures (Figure 4). Most of the ts CA mutants had similar, or modestly reduced, infectivity as compared to the parental virus when virion production and infection was done at the permissive temperature (35°C). Strikingly however, single-cycle replication of all the ts mutants was inhibited at least 50 to 1000-fold when restrictive temperatures were applied during virion production (Figure 4A, B). Conversely, all of the ts mutants were completely unaffected when restrictive temperatures were applied only during infection and not during production (Figure 4A, B). This result indicated that the defects associated with conditionally non-viable ts CA mutants occurred exclusively and irreversibly, during or shortly after, particle production, and not during early steps of the viral life cycle (e.g. uncoating or nuclear import). 10.1371/journal.ppat.1003461.g004 Figure 4 Conditionally viable (ts) CA mutants exhibit defects only when assembled at the non-permissive temperature. (A) Measurement of infectious virion titer from 293T cells transfected with either the parental or the ts single residue mutant proviral plasmids, and placed at either 35°C (filled circles), 37°C (filled squares), or 39.5°C for virion production (filled triangles), as indicated by distinct plot lines. The titer of infectious virus in the resulting supernatants from each mutant was measured in MT-4 cells at 35, 37, or 39.5°C as indicated on the x-axis, with the addition of dextran sulfate at approximately 16 h post-infection to limit replication to a single cycle. Note that some single ts mutants were resolved from double or triple mutants (see Materials and Methods) so do not appear in Table 1 or 2. (B) As in (A), however, here the temperature of virus production is indicated on the x-axis, and inoculations were done in MT-4 cells at 37°C only. Analysis of Gag expression and processing by the WT virus and the ts CA mutants revealed that similar levels of cell-associated Gag precursor Pr55 and capsid p24 were present regardless of the production temperature (Figure 5A). However, unlike the WT control, each of the ts CA mutant viruses generated 57% to 88% less extracellular virion-associated p24 at the restrictive temperature (Figure 5A). Because it was possible that the mutations could have affected the CA protein stability or recognition by the p24 antibody, we also analyzed Gag expression, processing and particle generation by the ts mutants using an anti-p17 matrix (MA) antibody (Figure 5B). This analysis yielded similar results. Specifically, there were approximately uniform levels of p17 in cell lysates for each of the ts CA mutants regardless of temperature. However, reductions in extracellular particulate p17 protein of 2.4–25-fold indicated that particle production was decreased by 59–96% for the ts mutants at the nonpermissive temperature. 10.1371/journal.ppat.1003461.g005 Figure 5 All ts CA mutants exhibit protease-dependent reduction in extracellular particle yield at the non-permissive temperature. (A). Western blot analysis (using an anti-CA antibody) of cell lysates and virions generated by transfected 293T cells. Two lanes are shown for each mutant: on the left is the sample from cells that were incubated at 35°C following transfection (Lo) and on the right is the sample from cells that were incubated at 39.5°C following transfection (Hi). Numbers below each lane indicate fluorescence intensities (LiCOR) associated with the CA protein pelleted from virion-containing supernatant. (B) Western blot analysis using an anti-HIV-1 MA antibody (p17) of cell lysates and virions, for the same panel as shown in A. Numbers below each lane indicate fluorescence intensities (LiCOR) associated with the MA protein pelleted from virion-containing supernatant. (C) Western blot analysis for the same panel of ts CA mutants as in A, but expressed in the context of a protease-defective proviral plasmid. Numbers below each lane indicate fluorescence intensities (LiCOR) associated with the CA protein pelleted from virion-containing supernatant. Although the 2.3 to 25-fold reduction in particle production observed for the ts CA mutants is smaller than the corresponding 50 to 1000-fold reductions in infectious virion yield, it is important to note that the majority of this reduction in infectivity (specifically 57% to 96%) was due to the inability of these mutants to efficiently generate particles. However, because the block in the generation of extracellular particles was not absolute, and there was residual generation of poorly infectious particles, it appeared that particle formation by the ts mutants was attenuated to varying degrees at the non-permissive temperature rather being than completely defective. The inability of the ts mutants to efficiently generate virions at the restrictive temperature is entirely consistent with the finding that it was the temperature during virion production, not inoculation, that determined the phenotype of all of the ts mutants we identified, and that all these mutations conferred defects that are manifested during virion morphogenesis. Careful inspection of both anti-CA and anti-MA blots of the ts mutants at the nonpermissive temperature revealed a slight abnormality in Gag processing. The parental virus appeared to generate a single ∼41 kDa band that reacted with anti-CA and anti-MA at both permissive and restrictive temperatures (presumptively the p41 MA-CA-p2 processing intermediate, Figures 5A and 5B). Conversely, one or sometimes two additional protein species, of similar but not identical mobility to p41 were observed for each of the ts mutants, specifically at the restrictive temperature. Thus, for each of the ts CA mutants, attenuated particle formation was accompanied by perturbation of Gag processing, which may have contributed to the overall defects in particle yield and infectiousness. Notably, eliminating viral protease activity eliminated the temperature-induced reduction in particle yield for all eleven ts CA mutants (Figure 5C). Thus, capsid mutations that caused temperature-dependent attenuation of particle formation do not do so prior to protease activation. This being so, and because inoculation temperature did not affect infectivity, it appeared that all eleven conditionally non-viable CA mutants have defects that are manifested during, and not before or after, virion morphogenesis. Most constitutively non-viable CA mutants exhibit attenuated particle formation The above analysis of the eleven conditionally non-viable CA mutants suggested that requirements imposed during particle production, and not during any other phase of the viral life cycle, are responsible for the mutational fragility of HIV-1 CA. However, it was possible that the selection of conditionally rather than constitutively non-viable mutants could have biased this conclusion. Thus, we examined the ability of the larger set of 81 constitutively non-viable mutants to generate extracellular particles, using western blot assays (Figure 6A). 10.1371/journal.ppat.1003461.g006 Figure 6 Most non-infectious constitutively non-viable CA mutants exhibit attenuated particle formation. (A) Western blots, probed with an anti-CA antibody, of cell lysates and virions for all constitutively non-viable CA mutants (those exhibiting 40% of parental virus fitness had bimodal distribution of occurrence in vivo. Some occurred relatively frequently (i.e. in >3% of natural sequences Figure 9B), while others occurred extremely rarely ( 40% of WT replication) but that occurred in less than 0.3% of subtype B isolates. The leftmost image shows the hexamer viewed from the exterior of the intact conical capsid, the center image show the hexamer viewed from within the plane of the capsid lattice and the rightmost image shows the hexamer viewed from the interior of the intact conical capsid. CA mutations that exhibit high fitness in vitro but occur rarely in vivo do not exhibit replication defects in primary cells The existence of a subset of mutations that gave high fitness in MT-4 cells in vitro, yet were virtually absent in natural populations, suggested that some mutations may be selected against in vivo in a manner that was not revealed by our fitness assays. Previous studies have demonstrated that CA mutations can have cell-type dependent effects on HIV-1 infectiousness [38], [39], [59], [60], and it was therefore possible that these mutants might exhibit fitness defects that are manifested in natural target cells, but not in MT-4 cells. Thus, we performed replication assays in primary cell types using 8 apparently fit viruses containing mutations that occurred rarely in natural populations (Figure 9C) and 5 randomly chosen viruses containing mutations that occurred frequently (Figure 9B). In fact, there was no significant difference between the ability of the rarely occurring or frequently occurring CA-mutant viruses to infect PBMC, primary CD4+ T cells or macrophages in short term (quasi single-cycle) infection assays (Figure 10A, B, C). Furthermore, there was no difference in the capacity of the rarely and frequently occurring mutants to replicate in a spreading infection assay in PBMC (Figure 10C). These results suggest that the rarity of apparently fit CA mutant viruses in natural HIV-1 subtype B populations is not due to differences in their capacity to replicate in primary cells. Rather, this finding suggests the presence of some unknown selective pressure that allows some intrinsically fit CA variants, but not others, to persist in vivo. 10.1371/journal.ppat.1003461.g010 Figure 10 ‘Rare but fit’ mutants behave indistinguishably from frequently occurring mutants in primary cells. (A) Percentage of infected macrophages following infection with VSV-G pseudotyped WT virus (NHGcapNM, filled squares) or derivatives containing mutations that occurred either frequently (filled triangles) or rarely (filled circles) in natural sequences, as indicated. (B) Percentage of infected stimulated CD4+ T cells, as indicated. (C) Spreading replication, following infection of stimulated peripheral blood mononuclear cells with WT virus or derivatives containing mutations that occurred either frequently or rarely in natural sequences, as indicated by the number of infected (GFP+) cells at the indicated times, after infection at an MOI of 0.1. Selective pressures that occur in vivo select sequence changes in the most genetically fragile domains of HIV-1 CA To further examine the relationship between the impact of CA mutation on in vitro fitness and occurrence in natural HIV-1 subtype B populations, we next compared naturally occurring variability across the CA sequence (as measured by Shannon entropy value) with the propensity of CA domains to tolerate mutation. The expectation was that regions that are more robust (Figure 3A) should exhibit higher variability in natural populations. Like a previous analysis of the Shannon entropy of the N-terminal of subtype C HIV-1 [43], this analysis revealed regions of high entropy in CA (>0.4) while other regions were more conserved (Figure 11A). There was a degree of accord between the ability of CA domains to tolerate mutation and Shannon entropy values. For example, there were shared regions of comparatively high robustness and high entropy, such as the cyclophilin binding loop and regions encompassing residues ∼5 to ∼35, and ∼175 to ∼210. Additionally, there were regions where especially low robustness coincided with regions of low variability in vivo, such as in CA helices 2 and 3, and also in the MHR (Figure 11A). However, there were also significant discrepancies between the Shannon entropy and robustness profiles across CA. For example, in helices 5, 6, and 7 (from residues 101 to 145), Shannon entropy values were often quite high, but fitness measurements revealed this region to be highly genetically fragile. This finding strongly suggests that selective pressures in addition to fitness act on the HIV-1 CA sequence in vivo. 10.1371/journal.ppat.1003461.g011 Figure 11 Correlation between fitness measurements and natural variation across the subtype B HIV-1 CA sequence. (A) Plot of fitness HIV-1 for library CA-mutants, overlayed with the Shannon entropy values for every residue in capsid, derived from 1000 subtype B isolates. The locations of the CA residues are arranged on the X-axis from left (N-terminal residue) to right (C-terminal residue), and their corresponding fitness (as a % of WT in a spreading replication assay) and Shannon entropy values are plotted on the Y-axes. The location of so called ‘protective’ CTL epitopes is indicated by horizontal blue lines. (B) Distribution of mutational fitness effects (DMFE) (at 37°C) for all single amino acid CA substitution mutants that occurred in or outside the protective epitopes indicated in (A). One such selective pressure is likely imposed by the adaptive immune system, particularly cytotoxic T-lymphocytes (CTLs) that can recognize peptides derived from HIV-1 CA. Indeed, several studies have demonstrated that mutations in CA can be driven by selective pressures imposed by CTLs [43]–[46]. Moreover, some of these studies have indicated that immune responses to particular epitopes in CA are protective, in that their appearance correlates with slower disease progression. For example, the oft studied protective HLA-B*57-restricted epitopes KF11 (residues 30–40) and TW10 (residues 108–117), and the B*27 restricted epitope KK10 (residues 131–140), all occur in regions of apparently extreme genetic fragility (Figure 11A, Table 2). In fact, a comparison of the DMFE values for the 41 library mutations that occurred in so-called protective epitopes [specifically QW11 (residues 13–23), IW9 (residues 15–23), KF11 (residues 30–40), TL9 (residues 48–56), EW10 (residues 71–80), TW10 (residues 108–117), DA9 (residues 166–174), KK10 (residues 131–140)] with 94 library mutations that occurred outside these epitopes, suggested that protective epitopes occur in regions of CA that are more genetically fragile than remaining CA sequence (Figure 11B). Indeed, the mean fitness value for mutations occurring within these ‘protective’ epitopes was 7.9% of WT while the mean fitness value for mutations occurring outside these epitopes was 15.1% of WT. Summary of the fates of CA mutations The scope of the mutagenesis carried out in this study, in which almost half of the amino acids in CA were individually mutated, coupled with an analysis of the frequency with which these mutations are present in natural populations, should allow generalized predictions of the fate of randomly introduced single nucleotide (Figure 12A) or single amino acid (Figure 12B) substitutions into CA. In the case of random nucleotide changes (Figure 12A), 22% should give synonymous changes that, in the great majority of cases, should not affect fitness (Figure 2). If there was no negative selection pressure in vivo other than fitness, then 10% of nucleotide substitutions are expected to yield a nonsynonymous change that is sufficiently replication competent (40% of in vitro WT fitness) to be potentially capable of flourishing in natural populations. However, if, as suggested by the data in Figure 9, it is true that only a fraction of mutations that yield variants with >40% of WT in vitro fitness are capable of persisting in vivo, then only 4% of all nucleotide substitutions are expected to yield non-synonymous changes that would flourish in a natural setting. 10.1371/journal.ppat.1003461.g012 Figure 12 Summary of the effects of random mutations in HIV-1 CA. (A) Estimation of the frequency of various outcomes, based on in vitro and in vivo data, for random single nucleotide substitutions (A) and random single amino acid changes (B). The charts indicate changes that are synonymous (A only), nonsynonymous and nonsense (determined computationally using the CA sequence from pNHGcapNM). Mutants encoding stop codons in CA were assumed to be non-viable and synonymous changes were assumed to be viable. Because only mutants with >40% fitness in vitro are frequently observed in natural populations, this threshold was considered as sufficiently fit to thrive in vivo (the proportion of random mutants that exhibit less than this 40% fitness threshold was estimated using data from Tables 1 and 2, and is indicated by the red circumferential line). The proportion of mutants actually expected to thrive (nonsynonymous viable in vitro and in vivo) and expected not to thrive (nonsynonymous viable in vitro only) was estimated using the data from Figure 9. The fate of the remaining mutants was estimated using the data from Figure 6 (nonsynonymous, non-infectious particles or nonsynonymous, attenuated (>5-fold deficit) particle formation). If only nonsynonymous nucleoside substitutions are considered (Figure 12B), then 87% are expected to be non-viable, with most mutations (59%) inducing a >5-fold attenuation of particle formation and a smaller fraction (21%) resulting in the generation of particles (at least 20% of the level of WT) that are non-infectious. Only 13% of nonsynonymous mutations are predicted to be sufficiently replication competent (at least 40% of WT fitness) to be potentially capable of flourishing in vivo, based on measurements of intrinsic fitness. Given the finding that only a fraction of intrinsically fit mutants actually persist in vivo, these data suggest that only 5% of all CA amino acid substitutions are actually expected to thrive in a natural setting. Discussion The goal of this study was to generate a reasonably sized sample of random mutations in the CA protein that might arise naturally during HIV-1 replication, and examine their biological effects. In so doing, we could determine the genetic robustness of HIV-1 CA and correlate the effect of amino acid substitutions in vitro with their occurrence in natural viral populations. Moreover, such a large library constitutes a resource for investigating various functions and properties of the HIV-1 capsid. Our results uncover a rather extreme genetic fragility in the HIV-1 CA protein, with a large fraction (∼70%) of individual, random amino acid substitutions resulting in non-viable viruses ( 40% of WT fitness). Conversely, all mutants that exhibited a fitness of 200-fold fewer colonies when ligated without insert than when ligated with insert. Furthermore, representative restriction digests indicated that aberrant restriction patterns, perhaps due to recombination, occurred at a frequency of only 1–2% of clones. Proviral plasmid DNA was extracted from individual cultures derived from 1056 colonies and subjected to sequencing and further analysis. Proviral plasmid DNA from the single mutants listed in tables 1 and 2 was freshly isolated, re-sequenced and analyzed by NotI and MluI restriction digest. The pNL4-3 Pr- plasmid is derived from pNHGCapNM and pNL4-3 (NIH AIDS Research and Reference Reagent Program, Catalog No. 114), and has NotI and MluI sites flanking CA as well as a point mutation in the protease active site, as has been previously described [83]. CA sequences harboring the temperature sensitive mutations were transferred from the pNHGcapNM vector to pNL4-3 Pr- by digestion with NotI and MluI. Cell lines and transfection The adherent human cell lines, 293T and TZM-bl, were maintained in DMEM supplemented with 10% fetal calf serum (FCS) and gentamicin. Suspension MT-4 cells were maintained in RPMI with 10% FCS and gentamicin. For transfection experiments, 293T cells were plated at 2.5×104 cells/well in 96 well plates or 1.5×105 cells/well in 24 well plates. To measure the effect of CA mutations, and their temperature sensitivity, transfections were done the following day using polyethylenimine (Polysciences), and either 100 ng of the WT or mutant NHGCapNM plasmids described above (for 96 well plate experiments) or 500 ng (for 24 well plate experiments), or 500 ng of the pNL4-3 Pr- mutants (all in 24 well plate experiments). Plates for all transfection experiments were placed at 35, 37 or 39.5°C, as indicated, immediately upon addition of transfection mixture. Primary cells PBMCs, CD4+ T cells, and macrophages were isolated from buffy coats (from anonymous healthy blood donors and were purchased from the New York Blood Center) using a Ficoll gradient. Primary CD4+ T cells were extracted using a RosetteSep Human CD4+ T-cell enrichment cocktail. Macrophages were isolated by adhesion to plastic and treatment with 100 ng/ml of granulocyte/macrophage-colony stimulating factor (GM-CSF) for 96 hours prior to infection. All primary cells were maintained in RPMI supplemented with 10% FCS, penicillin/streptomycin, and L-glutamine. Activation of PBMCs and CD4+ T cells was achieved by addition of phytohemagglutinin (PHA) at 5 µg/ml for 72 hours prior to infection, with addition of 25 U/ml of interleukin-2 at the time of infection. Viral replication and infectivity assays 293T cells were transfected with the proviral plasmids and given fresh medium 16 hrs later. At ∼40 h post-transfection, cell supernatants were collected and filtered (0.22 µm). Single-cycle infectivity was measured using MT-4 cells that were seeded in 96 well plates at 3×104/well and inoculated with a volume of filtered supernatant that was equivalent to an MOI of ∼1 for the WT viral clone. Dextran sulfate at 100 µM was added 16 hrs later to limit replication to a single cycle, and cells were fixed in 4% PFA 48 hrs after infection. Alternatively, harvested supernatants were subjected to a low-speed spin and a freeze-thaw cycle before addition to MT-4 or TZM-bl cells in 96 well plates, in the presence or absence of aphidicolin (2 µg/ml) or at 35, 37, or 39.5°C, as indicated in figure legends. For spreading replication assays, MT-4 cells were inoculated with a volume of filtered supernatant that was equivalent to an MOI of ∼0.01 for the WT viral clone and fixed in 4% PFA at 72–80 hrs post-infection. FACS analysis for all infectivity and replication assays was carried out using a Guava EasyCyte instrument. The following mutations, some of which were part of either double or triple mutants, occurred in mutants that exhibited temperature sensitivity in spreading replication assay (Figure S1): S4C, L189M, I91F; Q6P, A78V; I15M, A78T; S16T, T48A; S33C, A92V; I91N, D163E; I91V, I124T; M96T; R100S, Q112L; R132G; R167Q; M214L; K227I. CA mutants encoding each individual substitution were generated, as necessary, prior to selection of the 11 ts mutants for analysis in Figures 4 and 5. Infections of primary cells required different conditions. For single-cycle infections of PBMCs and CD4+ T cells, 0.1–1×106 cells were inoculated using virus generated from 293T cells, as described above, for the 11 rarely occurring mutants described in Figure 9C, the WT virus, and 4 randomly selected frequently occurring mutants from Figure 9B (H87Q, I91V, E98D, T200S) at an MOI of ∼5, and were fixed in 4% PFA 36 hours later. For infection of macrophages, VSV-pseudotyped virus was generated in 293T cells for the same 16 mutants plus WT virus. Macrophages were infected at an MOI of ∼4 and fixed approximately 72 hours later. For spreading infection assays in PBMC, 1×106 cells were infected at an MOI of 0.1, and cells fixed at the time points indicated in Figure 10C. All MOIs used for primary cells were reference values derived from titrations on MT-4 cells. Western blotting Cell lysates and virions (pelleted by centrifugation through 20% sucrose) were resuspended in SDS sample buffer and separated by electrophoresis on NuPage 4–12% Bis-Tris Gels (Novex). Proteins were subsequently blotted onto nitrocellulose membranes, probed with a primary anti-HIV p24 capsid antibody (183-H12-5C) or anti-HIV p17 antibody (VU47 Rabbit anti-p17 [84]), and then probed with a goat anti-mouse/anti-rabbit IRDye® 800CW secondary antibody (LI-COR Biosciences). A LI-COR Odyssey scanner was used to detect and quantify fluorescent signals. A minimum of 3 separate Western blots was produced for each temperature sensitive mutant, including those in the protease negative background, and representative blots are shown. Structure analysis HIV-1 capsid hexamer structural analysis was done using MacPyMOL, with PDB reference 3GV2 [26]. Solvent accessibility surface area of residues was determined using UCSF Chimera. Thin-section electron microscopy To prepare samples for thin-section electron microscopy, 293T cells were seeded in 6-well plates at 0.8×106 cells per well in duplicate. Transfections were done the following day with the addition of polyethylenimine to 2 µg of the WT or mutant NHGcapNM plasmids (S33C, T48A, M96T, Q112L, R132G, L189M, G60W, K25I, L52F or N195S) plus 0.5 µg of modified human tetherin (delGI, T451 that is resistant to antagonism by HIV-1 Vpu [58]) to enhance visualization of virions at the plasma membrane. Immediately after addition of transfection mixture, plates were placed at either 37 or 39.5°C, as indicated. After 48 hrs, supernatant was removed and cells were fixed in a solution of 2% paraformaldehyde (PFA), 2.5% glutaraldehyde. One set of cells was then analyzed by FACS to compare transfection efficiencies (which ranged from 23% for WT to 32% for L189M). The other set were fixed with 2.5% glutaraldehyde and 1% osmium tetroxide, and stained with 2% aqueous uranyl acetate. Fixed and stained cells were harvested into PBS and pelleted through 1% SeaPlaque agarose (Flowgen) at 45°C. The agar was set at 4°C and the cell pellets were cut into ∼2 mm cubes, which were dehydrated through a graded alcohol series and infiltrated with TAAB 812 embedding resin. After polymerisation, thin sections (120 nm) were cut and examined in a JEOL 1200 EX II electron microscope. Numbers of virus particles associated with 150 randomly selected cells were counted for each sample. Cryo-electron microscopy To isolate virions for cryo-electron microscopy, 10 cm plates of approximately 4×106 293T cells were transfected with 7 µg of the indicated plasmids (NHGcapNM, R18G, K30N, G60W or M215V) using polyethylenimine. After 48 hours, supernatant was collected and filtered and virions were pelleted by centrifugation at 14000 rpm through 20% sucrose. Virions were fixed in 10 µl of 2% PFA, 2.5% glutaraldehyde solution. Fixed aliquots of 3 ul of each sample were loaded onto freshly glow-discharged c-flat holey carbon grids (CF-22-4C, Protochips, Inc.) held at 4°C and 100% humidity in a Vitrobot vitrification robot (FEI). Grids were blotted for 4 s prior to being frozen by plunging into a bath of liquid nitrogen-cooled liquid ethane. Vitrified specimens were imaged at low temperature in a JEOL 2200 FS cryo-microscope equipped with Gatan 626 cryo-stages. Low dose (10 e/Å2), energy-filtered images (slit width, 20 eV) were recorded on a Gatan ultrascan 16-megapixel charge-coupled-device camera at a magnification of 50,000×. Analysis of natural CA variants A set of 1,000 HIV-1 subtype B sequences isolates was obtained from the Los Alamos HIV sequence database (www.hiv.lanl.gov/). All sequences were sampled from distinct infections between from 1980 and 2009. To minimize risks of sampling biases, multiple sequences from known transmission clusters were excluded. Sequences with frameshift mutations or stop codons that were likely to represent nonfunctional viruses or poor quality sequencing were excluded. Sequences were aligned using MUSCLE [85], and PERL scripts were used to examine genetic variation in the resulting sequence alignment. An information-theoretic measure of diversity (Shannon's entropy) [86] was applied to quantify the amount of amino acid variation at each position in capsid. Supporting Information Figure S1 Temperature sensitive mutants in entire CA mutant library. The Y-axis indicates the percentage of infected (GFP+) MT-4 cells from mutant viruses following a spreading replication assay done at 35°C, while the x-axis shows the percentage of infected cells following a spreading replication assay done at 39.5°C. Mutants that generated less than 0.1% infected cells are not shown. (TIF) Click here for additional data file.