Clustering of HIV-1 sequences among men who have sex with men in Beijing, China

Simon Frost, M.A. D.Phil.

Dept. of Veterinary Medicine, and Institute of Public Health

University of Cambridge

Co-authors and funders

Introduction

• While HIV is endemic in many regions, in some populations, HIV is just emerging
• The HIV epidemic is emerging among men who have sex with men in Beijing, China
• What might we learn about this emerging epidemic from combining phylogenetic and behavioural data?

Individual-level data and clustering

• How can we combine individual-level data with sequence data?
  • Highlighted as a challenge by Frost et al., Epidemics (2015)
• Most studies have reduced sequence data to whether an individual is clustered or not
  • Two individuals cluster if their viral sequences are within a threshold distance
  • May also be other criteria, e.g. bootstrap/posterior probability support

Why cluster?

• Computational reasons: easier to fit clusters of related sequences than all of them
  • e.g. UK HIV Drug Resistance Database c. 100,000 sequences
  • (but can lead to biased estimates of phylodynamic parameters; see Bethany Dearlove's talk)
• Populations are structured; clusters may reveal subepidemics with higher transmission rates
• Can use standard statistical approaches

Why not cluster?

• Loss of information
• Problems with clustering thresholds
  • Evolutionary rate. The rate may change with transmission rate, and vary by subtype.
  • Time since infection. What happens with a mixture of individuals at different stages of disease?
  • Sampling. Sampling rate can change over time and space
  • Intermediate number of infections may vary by stage of infection
  • Choice of threshold. Too small - not enough clusters. Too large - everyone is clustered.

Methods for Prevention Packages Program study

• NIH funded program for reducing HIV transmission
• Vanderbilt/China CDC study in Beijing, China
  • Recruited from March 2013 to March 2014
• Subset of 356 individuals with HIV pol data, who self identify as MSM, completed a behavioural questionnaire

Modeling the transmission of HIV in Beijing

Projections of HIV in Beijing

Subtyping process

• We used a variety of subtyping tools:
  • Rega (V2, V3)
  • SCUEAL
  • COMET
  • jpHMM
  • STAR
  • RIP
  • BLAST
• Exploratory phylogenetic analysis

Subtypes

Subtypes and internet use

Clustering and minimum distance

• We calculated pairwise distances between all sequences in the sample
  • Sergei Kosakovsky Pond's TN93 program
• For each sequence, selected the minimum of these distances
• Large differences between subtypes

Differences between subtypes

Patterns of clustering (<1%)

Distribution of distances

Clustering with 'background' sequences

• Downloaded partial pol sequences from Genbank, at least 1000 bp long
  • 108,135 'background' sequences
    • 424 were sampled from MSM in Beijing
• Calculated the (TN93) distance from the sample sequences to the background sequences

'Background' clustering

Modeling clustering

• Problem:
  • On the one hand, we have a clear excess of very close (<0.1%) clustering
  • On the other, the distances vary greatly by subtype
    • Sampling times are too close together to reliably estimate evolutionary rate
• Solution:
  • Take minimum distance as an outcome
  • Model this as a zero adjusted Gamma (ZAGA) model

Zero adjusted Gamma

Clustering and receptive intercourse

Clustering and active syphilis

Distance and years resident in Beijing

Who are people clustering with?

• In addition to calculating the minimum distance, we can examine the characteristics of the individual who is linked to each
• We calculated the number of ambiguous nucleotides ('mixtures') that did not affect the amino acid sequence for each individual
  • Crude measure of within-host diversity, which increases during infection

Links per alter

Lessons from modeling

• We combined the mathematical model of Lou et al. (2014) with the phylodynamics framework of Volz (2012)
• Simulated trees with 356 taxa, with the same sampling times as the observed data
• Calculated the minimum cophenetic distance for each taxon

Minimum distance in simulation

Next steps

• Retrospective analysis of HIV sequence data collected by China CDC
• Better behavioral data collection?
  • No obvious patterns between subtypes
• Model development
  • Immigration of infecteds
  • More structure in the population

Conclusions

• Exponential epidemic of HIV-1 among MSM in Beijing
  • Particularly CRF07_BC, perhaps linked to finding partners on the internet
  • Mixture of potentially direct transmission and looser association with risk groups
  • Not seen in 'well mixed' models of transmission
  • Demographic data reinforce the importance of movement of people
• Such 'late' emergence is not unique
  • Manila, Philippines

Thanks!

 sdwfrost@gmail.com  @sdwfrost  http://github.com/sdwfrost