• Tidak ada hasil yang ditemukan

3.3 Results

3.3.4 Most probable network for Gag-PR

The most probable Gag-PR network comprised 35 variables including the LPV/r treatment node (Figure 3.3B). Together with the six PRMs described in Figure 1A, 10 variables from Gag was included in the final dataset. Interestingly, instead of altering the interactions observed in PR (Figure 3.3A), the Gag mutations added to the existing network and mostly correlated with other Gag mutations. In terms of resistance pathways, a direct association was observed between LPV and 54V via P63 (through an inverse association with 63L). While 54V maintained its association with 46I+82A, Gag mutation 431V correlated with this combination. Moreover, a strong direct association linked 10F with 431V. Unlike its dependency on Q58 and L24 in the PR network, L76V was also found associated with 46I, occurring after the 46I+54V+82A combination.

Noteworthy was Gag polymorphic variable 69K which formed a parent arc with the LPV treatment node.

56

Figure 3.3 Most probable network showing associations between drug resistance mutations, polymorphisms, wild-type residues and Lopinavir (LPV/r) treatment in (A) protease and (B) Gag-protease. Arc thickness represents bootstrap support. Note: Arc direction does not represent the order of accumulation of mutations or causality but may indicate a multivariable effect in the network.

57 3.4 Discussion

Using a combination of phylogenetic and Bayesian statistical inferences, we identified several positively selected and coevolving Gag-PR sites under drug selection pressure and used this information to construct Bayesian networks to identify pathways to LPV/r resistance.

In the context of evolution, reconstructing ancestral phylogeny is crucial since phylogenetic trees can provide information on the evolutionary history of genes (Soltis and Soltis, 2003). Maximum- likelihood and Bayesian methods use probabilistic approaches to estimate tree topologies.

Usually, statistical nodal support is calculated by random alignment resampling with replacement in a technique known as bootstrapping. Although acceptable, interpreting bootstrap values is challenging (Hassan et al., 2017). Instead of inferring the single most probable tree observed in ML, Bayesian methods produce a set of credible trees (0.95 CI17) that reflects the posterior probability topology given the data (Paraskevis et al., 2004; Hassan et al., 2017). In comparison to bootstrapping, Bayesian posterior probability has been suggested to be less biased in phylogenetic accuracy (Hassan et al., 2017). Therefore, in this study, both ML and BI methods were used to construct the phylogenetic trees. To obtain an effective sample size (ESS) >200 in BI, the Markov chain Monte Carlo (MCMC) length was set to 10,000,000. With only a 6.33%

difference between the ML and BI topologies, the trees were considered acceptable for the positive selection and CAPS analyses as discussed below.

Positive selection is the acquisition of advantageous mutations. These signature mutational patterns contribute towards functional protein shifts (Qian et al., 2017) and has been widely studied using codon substitution models (Romero et al., 2016). In this study, a comparison of different site models revealed that models allowing for positive selection (M2a and M8) fit the data better than their counterparts (M1a and M7). Although most sites were under a strong purifying selection, several of the positively selected sites were mostly located in Gag. Similarly, CAPS revealed that most coevolving sites were found in Gag than PR. Taken together, this suggests that Gag can tolerate mutational changes better than PR. Interestingly, this theory was corroborated in a study conducted by Darapaneni et al. (2015) where 61% of the viral PR was conserved, particularly within the dimer interface, active site and flap regions. This suggests that AA changes are restricted to specified regions in PR to maintain the structural integrity of the enzyme during cleavage. Contrastingly, Gag displays high natural variability (Li et al., 2013), particularly at the CSs (Torrecilla et al., 2014). Prabu-Jeyabalan et al. (2002) demonstrated that

17 Confidence Interval

58

substrate recognition by PR is determined by the asymmetric structure of the CSs and not necessarily the AA sequence. This theory, coined the “substrate envelope” hypothesis was based on studies involving inactive PR variants complexed with peptide substrates (Prabu-Jeyabalan et al., 2002; Tie et al., 2005). Although few, codons located at the CSs including E365 (CA|p2), S373 (p2|NC) A431 (NC|p1) and R452 (p1|p6) were identified in this study. Additionally, Özen et al. (2014) showed that the “conserved dynamic behaviour” of PR’s active site, together with complimentary enzyme-substrate mutations increases substrate recognition. Considering the location of the PRMs identified in these sequences (Table 3.1), this suggests that PR and CS mutations work together to alter the substrate envelope of PR, increasing Gag proteolysis and providing a competitive advantage over the PIs.

To understand the resistance dynamics of specific mutations at various coevolving sites, we explored pathways to resistance in PR and observed the multivariable effect of what happens when Gag mutations are then introduced. In this study, the combination of M46I+I54V+V82A selected by PR is unsurprising since this is the three most common substitutions seen in LPV/r failure (Barber et al., 2012; Van Zyl et al., 2013; Grossman et al., 2014). As previously mentioned, in tandem with this combination, the introduction of L76V can also lead to DRV/r cross-resistance (Tang and Shafer, 2012). Although several studies have demonstrated the impact of genetic diversity on pathways to drug resistance (Sylla et al., 2008; Martinez-Cajas et al., 2009; Barber et al., 2012), the stepwise accumulation and multivariable interaction of mutations in pathways to resistance is largely unknown. In this study, two probable pathways in PR leading to LPV/r failure were described, one beginning with I54V and the other with V82A. Interestingly, Zhang et al.

(2010) described a conditional structural independence between 46+54+82, where 46 and 54 are mutually independent given 82. Therefore, the authors posited that the sequential order of mutations would be either 46-82-54, 82-46-54 or 82-54-46. However, a separate study conducted by Zhang et al. (1997) described an order of 54-46-82 and 46-54-82. Although both studies reiterate these dependencies, our study suggests that the deciding factor as to whether I54V or V82A is selected first is dependent on L10F, where if I54V is selected first then L10F is also selected. However, if V82A is selected first followed by I54V then either the L10F mutation or the WT L10 could be favoured. Although more work would be required to confirm this theory, L10F could be functionally important. Previously, Martinez-Picado et al. (1999) showed that viruses harbouring the L10R mutation together with M48I+L63P+V82T+I84V replicated at the same rate as the WT, thereby suggesting its role in restoring viral fitness. However, the role of L10F when the M46I+I54V+V82A combination is selected together with L76V as seen in this study, is still unclear. Interestingly, Wong-Sam et al. (2018) showed that L76V reduces

59

hydrophobic contacts with I45/47 in PR’s flaps and D30/T74 in PR’s core, corresponding with protein instability. However, the authors postulated that reduced contacts caused by Val76 can allow for greater flap mobility. Furthermore, Louis et al. (2011) showed that the co-selection of M46I counteracts this instability by enhancing proteolytic cleavage in subtype B. This suggests that L76V occurs to modify flap dynamics to maintain cleavage with the altered Gag CSs whilst maintaining resistance to LPV/r.

Interestingly, the pathways to resistance changes significantly when other PIs are investigated.

For example, Deforche et al. (2006) showed that though L10F, M46I and I54V were present in the network, only major PRMs D30N, N88S and L90M connected with the NFV node as parent.

However, using a temporal nodes Bayesian network, Hernandez-Leal et al. (2000) showed that when IDV was present, a direct pathway existed between M46I, I54V and V82A through L90M as parent. Thus, the selection of specific mutations and intricate changes in the pathways to resistance are attributed to the drug itself. However, these networks may still change if Gag mutations are also introduced.

In the Gag-PR network a combination of L10F+M46I+I54V+V82A and A431V was observed.

The A431V Gag mutation can cause resistance to all PIs except DRV (Fun et al., 2012) even in the absence of PRMs (Nijhuis et al., 2007). Moreover, its level of resistance is comparable to that of a single major PRM such as V82A (Nijhuis et al., 2009) and it can improve Gag cleavage regardless of the protease variant (Kolli et al., 2009). Although A431V has been associated with V82A (Prabu-Jeyabalan et al., 2004), a direct relationship was observed with I54V in this study.

This could suggest that A431V and its link to V82A is dependent on the presence of I54V and M46I. This is consistent with the findings observed by Singh (2015) wherein A431V was mostly found in viruses with PRMs (M46I+I54V+V82A) than those without. This suggests that the if PRMs are selected, A431V serves to act in a structural compensatory role at a later stage.

Of note was non-CS mutation Q69K (chapter two; Figure 2.5). In this study, Q69K indirectly associated with A431V via G62K and V77 in PR. Singh (2015) reported that Q69K increased viral replication by 13% in the presence of A431V and M46I+I54V+V82A suggesting that it occurs to further compensate for changes associated with these mutations. Additionally, although CAPS identified the coevolving pair of 77 and 431, an indirect inverse association between these two codons was noted in the network. Interestingly, Gupta et al. (2015) demonstrated that a PR double mutant (V77I+L33F) resulted in a more stable structure with altered flap dynamics in the presence of Nelfinavir (NFV) in comparison to the WT. This could suggest that while A431V

60

works to restore cleavage, V77I works to stabilize the mutant PR by changing the open/closed conformation of the flaps to compensate for the modified CS. However, more work is required to elucidate the relationship between A431V+V77I in the presence of LPV/r.