Summary of statistics of signature sites of b12 sensitivity.

1HXB2 position refers to the amino acid position of interest in the HXB2 reference strain (www.hiv.lanl.gov: Locator tool).
2Amino acid refers to the particular amino acid or combination of amino acids that was statistically related to b12 resistance (underlined) or sensitivity (not underlined and italics). An exclamation point means “not”; thus in the first line, when T is an ancestral state, Y mutates to “not Y” (!Y) with a statistically higher frequency in b12 resistant strains than sensitive strains.
3Statistic is the statistic that was used to identify the signature, by either the phylogenetically corrected contingency approach (Fisher exact test) employed as described in [54]; the conditional mutual information approach (CMI); or a comparison of all variable region loop lengths (length) and number of glycosylation sites (sequons with amino acid pattern Nx[ST]) with the b12 neutralization values using a Spearman rank correlation test.
4The p-values, the q-values (false discovery rates), and the odds ratios are provided. The Fisher's exact test q-values were calculated for discrete tests as implemented in [54]. For the CMI analyses, p-values were acquired by shuffling phenotypes and counting the relative frequency at which random CMIs exceeded the original CMI. The q-values were calculated using the method of [124], after stripping off the highest p values (essentially a few hundred of p-value = 1). Only associations with a q-value<0.2 are shown.
5Rows and columns of the 2×2 contingency table. As an example of how to read these, in position 173, r1c1 refers to row 1 column 1 and is the number of times among b12 sensitive viruses that Y→!Y mutates to another amino acid (change). r1c2 refers to row 1 column 2, and it is the number of times among sensitive viruses that the ancestral state was Y and it stayed Y (stable) in the Env sequence.
6Strength is a measure that expresses how predictive a given signature amino acid is of the b12 sensitive/resistant phenotype, essentially an augmented odds ratio, where each count was augmented by 1 pseudo-count to avoid issues with zeros and infinities, and strength = (r1c1+1)(r2c2+1)/(r1c2+1)(r2c1+1).
7Several explorations of the Env alignment were used, and this is described in the “test”. In our first screen, every amino acid found in every column was tested (1aa). Then combinations of 2 or more amino acids in every column were tested (>1aa). Then positions known to be key for the b12 binding site (Sup. Table S3) were specifically tested for all combinations of amino acids over all pairs of positions in the binding site (b12). Although pairs of positions were tested, single positions essentially accounted for the signal in that analysis. Only these single site associations are shown.
8Some lines are shown in bold. In these lines, the change in the amino acid is associated with a reverse in the majority of cases found among sensitive or resistant viruses; thus the change in these sites is particularly predictive of NAb phenotype.
9All PNLG sites in Env were tested for phylogenetically corrected association with b12 sensitivity using the contingency table approach. None reached significance with a q-value of <0.2; the glycosylation site at position 149 was the only one to reach even borderline significance and is included here for completeness.
10For the initial analysis of loop length and number of PNLGs in each loop, we did not used a phylogenetically corrected method. Rather we used a non-parametric Spearman's correlation test comparing loop length with the geometric mean 50% neutralization titer for the 25 Envs. It is reasonable to forego the phylogenetic correction in these cases because the loop lengths vary by insertion and deletion and often change dramatically within infected individuals. These parameters are less likely to be biased by phylogeny at the population level.