Selection Bias at the HIV-1 Transmission Bottleneck

An in depth walkthrough of the data published in Science July 2014, which showed evidence that heterosexual HIV-1 transmission favors viruses that generally replicate better. The modeling predict (and data confirm) that an important implication is that anything that reduces the biological risk of infection will increase selection for "stronger" viruses, leading to more severe disease when breakthrough infection does occur. See http://research.microsoft.com/~carlson/papers/selectionBias14.html for more details and links to the paper.

biologystatisticsHIV
1.0x

Selection Bias at the HIV-1 Transmission Bottleneck

Created 3 years ago

Duration 0:50:08
lesson view count 83
An in depth walkthrough of the data published in Science July 2014, which showed evidence that heterosexual HIV-1 transmission favors viruses that generally replicate better. The modeling predict (and data confirm) that an important implication is that anything that reduces the biological risk of infection will increase selection for "stronger" viruses, leading to more severe disease when breakthrough infection does occur. See http://research.microsoft.com/~carlson/papers/selectionBias14.html for more details and links to the paper.
Select the file type you wish to download
Slide Content
  1. Jonathan Carlson, PhD

    Slide 1 - Jonathan Carlson, PhD

    • Microsoft Research
    • http://research.Microsoft.com/~carlson
    • carlson@Microsoft.com
    • Science 345:1254031, July 11, 2014
  2. The transmission bottleneck

    Slide 2 - The transmission bottleneck

    • HIV+
    • HIV-
    • >99% no infection
    • 80-90% of infection by 1 virus
    • Binomial
  3. The transmission bottleneck

    Slide 4 - The transmission bottleneck

    • HIV+
    • HIV-
    • >99% no infection
    • 80-90% of infection by 1 virus
    • Binomial
    • (not independent)
    • Abrahams, JVI 2009
  4. The transmission bottleneck

    Slide 5 - The transmission bottleneck

    • HIV+
    • HIV-
    • >99% no infection
    • 80-90% of infection by 1 virus
    • Are all viruses equally likely?
    • ?
  5. X

    Slide 6 - X

    • The transmission bottleneck
    • HIV+
    • HIV-
    • >99% no infection
    • 80-90% of infection by 1 virus
    • ?
    • Are all viruses equally likely?
  6. Couple is identified as HIV serodiscordant

    Slide 7 - Couple is identified as HIV serodiscordant

    • HIV negative partner is tested once per month
    • HIV negative partner seroconverts
    • Plasma collected from Donor and Recipient, median 45 days post estimated infection
    • HIV sequencing of both partners
    • N=137
    • N>1,000
    • Couples counseling and condoms reduce transmissions by 2/3
    • Susan Allen
    • Eric Hunter
  7. ...EPRGSDIAATTSNLQEQIGWMTSNPPIPV...

    Slide 8 - ...EPRGSDIAATTSNLQEQIGWMTSNPPIPV...

    • ...DPRGSDIAGTTSNLQEQIGWMTNNPPIPV...
    • ...EPRLSDIAGTTSNLQEQIGWMTNNPPIPV...
    • ...EPRGSEIAGTTSTLQEQIAWMTNNPPIPV...
    • ...EPRLSDIAGTTSNLQEQIGWMTNNPPIPV...
    • ...EPRGSDIAGTTSTLQDQIGWMTNNPPIPV...
    • ...DPRGSDIAGTTSNLQEQIAWMTHNPPVPV...
    • ...EPRLSEIAGTTSTLQEQITWMTNNPPIPV...
    • ...EPRGSDIAGTTSNLQEQIGWMTSNPPIPV...
    • ...DPRGSDIAGTTSTLQEQIGWMTNNPPIPV...
    • ...EPRGSDIAGTTSNLQEQIGWMTNNPPIPV...
    • ...EPRGSEIAGTTSTLQEQIGWMTNNPPIPV...
    • ...EPRLSDIAGTTSNLQEQIGWMTNNPPIPV...
    • ...EPRGSDIAGTTSTLQEQIGWMTNNPPIPV...
    • ...DPRGSDIAGTTSNLQEQIAWMTNNPPVPV...
    • ...EPRLSEIAGTTSTLQEQITWMTNNPPIPV...
    • Donor
    • Recipient
    • ID1
    • ID2
    • ID3
    • ID4
    • ID5
    • ID6
    • ID7
    • ID8
    • 99.8% identity
    • High level of identity by descent
  8. Slide 9

    • Zambian consensus more likely to be transmitted
    • % transmitted sites
    • per couple
    • Figure 1
  9. ...EPRGSDIAATTSNLQEQIGWMTSNPPIPV...

    Slide 10 - ...EPRGSDIAATTSNLQEQIGWMTSNPPIPV...

    • ...DPRGSDIAGTTSNLQEQIGWMTNNPPIPV...
    • ...EPRLSDIAGTTSNLQEQIGWMTNNPPIPV...
    • ...EPRGSEIAGTTSTLQEQIAWMTNNPPIPV...
    • ...EPRLSDIAGTTSNLQEQIGWMTNNPPIPV...
    • ...EPRGSDIAGTTSTLQDQIGWMTNNPPIPV...
    • ...DPRGSDIAGTTSNLQEQIAWMTHNPPVPV...
    • ...EPRLSEIAGTTSTLQEQITWMTNNPPIPV...
    • ...EPRGSDIAGTTSNLQEQIGWMTSNPPIPV...
    • ...DPRGSDIAGTTSTLQEQIGWMTNNPPIPV...
    • ...EPRGSDIAGTTSNLQEQIGWMTNNPPIPV...
    • ...EPRGSEIAGTTSTLQEQIGWMTNNPPIPV...
    • ...EPRLSDIAGTTSNLQEQIGWMTNNPPIPV...
    • ...EPRGSDIAGTTSTLQEQIGWMTNNPPIPV...
    • ...DPRGSDIAGTTSNLQEQIAWMTNNPPVPV...
    • ...EPRLSEIAGTTSTLQEQITWMTNNPPIPV...
    • Donor
    • Recipient
    • ID1
    • ID2
    • ID3
    • ID4
    • ID5
    • ID6
    • ID7
    • ID8
    • 99.8% identity
    • 137
    • 1712
    • Which sites are transmitted?
  10. Which sites are transmitted?

    Slide 11 - Which sites are transmitted?

    • E E
    • P P
    • … …
    • N T
    • S S
    • T T
    • … …
    • P P
    • V V
    • D
    • R
    • ID1, pos1
    • ID1, pos2
    • ID2, pos11
    • ID2, pos12
    • ID2, pos13
    • ID8, pos1711
    • ID8, pos1712
    • T
    • 1
    • 1
    • 0
    • 1
    • 1
    • 1
    • 1
    • N=228,362
    • Approach
    • Estimate with 454
  11. Slide 12

    • 5 couples; all sites in Gag, Pol and Nef
    • Todd Allen
    • Figure 2A
  12. Which sites are transmitted?

    Slide 13 - Which sites are transmitted?

    • E E
    • P P
    • … …
    • N T
    • S S
    • T T
    • … …
    • P P
    • V V
    • D
    • R
    • ID1, pos1
    • ID1, pos2
    • ID2, pos11
    • ID2, pos12
    • ID2, pos13
    • ID8, pos1711
    • ID8, pos1712
    • T
    • 1
    • 1
    • 0
    • 1
    • 1
    • 1
    • 1
    • N=228,362
    • Approach
  13. Estimating fitness with statistical power

    Slide 14 - Estimating fitness with statistical power

    • E E
    • P P
    • … …
    • N T
    • S S
    • T T
    • … …
    • P P
    • V V
    • D
    • R
    • ID1, pos1
    • ID1, pos2
    • ID2, pos11
    • ID2, pos12
    • ID2, pos13
    • ID8, pos1711
    • ID8, pos1712
    • T
    • 1
    • 1
    • 0
    • 1
    • 1
    • 1
    • 1
    • N=228,362
    • Approach
    • Use generalized linear mixed model for hypothesis testing.
  14. What is special about consensus?

    Slide 15 - What is special about consensus?

    • Consensus fitness??
  15. A fitness bias

    Slide 16 - A fitness bias

    • AA features related to fitness
    • Conservation
    • Predicted impact on structure
    • Interaction network properties
    • Relationship to immune escape
  16. A fitness bias

    Slide 17 - A fitness bias

    • AA features related to fitness
    • Conservation
    • Predicted impact on structure
    • Interaction network properties
    • Relationship to immune escape
    • Figure 2C
  17. A fitness bias

    Slide 18 - A fitness bias

    • AA features related to fitness
    • Conservation
    • Predicted impact on structure
    • Interaction network properties
    • Relationship to immune escape
    • Figure 2D-F
  18. Applied to full sequences

    Slide 19 - Applied to full sequences

    • E E
    • P P
    • … …
    • N T
    • S S
    • T T
    • … …
    • P P
    • V V
    • D
    • R
    • ID1, pos1
    • ID1, pos2
    • ID2, pos11
    • ID2, pos12
    • ID2, pos13
    • ID8, pos1711
    • ID8, pos1712
    • T
    • 1
    • 1
    • 0
    • 1
    • 1
    • 1
    • 1
    • Ln-odds (T)
    • 6.9
    • 7.2
    • 4.6
    • 5.3
    • 6.8
    • 8.3
    • 6.3
    • Transmission Index
    • 7.1
    • 6.3
    • 6.8
  19. Slide 20

    • Figure 5
  20. Slide 21

    • Individuals with weak viral populations are less likely to transmit to their partners
    • Figure 5
    • Table S3
  21. What’s going on?

    Slide 22 - What’s going on?

    • Bias toward consensus
    • Consistent with “archival” transmission [Derdeyn 2004; Sagar 2009]
    • Consistent with slower between than within host evolution [Alizon & Fraser 2013; Vrancken 2014]
    • Consistent with rapid reversion after transmission [Herbeck 2006]
    • Is it fitness or (in addition to?) “archived” sequences?
    • Continuous quadratic effect of Zambian conservation
    • Interaction with protein structure, compensation, immune escape
    • Prediction of which viruses and which donors will transmit
  22. What’s going on?

    Slide 23 - What’s going on?

    • The fitness hypothesis makes two key predictions
    • An interaction with infection risk
    • An interaction with reversion rates
  23. Interaction with risk

    Slide 24 - Interaction with risk

    • Prediction: increased risk will decrease bias
  24. Interaction with risk

    Slide 25 - Interaction with risk

    • Prediction: increased risk will decrease bias
    • Figure 3
  25. Interaction with risk

    Slide 26 - Interaction with risk

    • Increased fitness in addition to increased number of viruses
    • Figure 3
  26. Here’s the idea

    Slide 27 - Here’s the idea

    • Most viruses can clear a low bar.
    • Only the strong will clear the high bar.
  27. Here’s the idea

    Slide 28 - Here’s the idea

    • Most viruses can clear a low bar.
    • Only the strong will clear the high bar.
    • Brad Walker
    • (American Record Holder)
  28. Here’s the idea

    Slide 29 - Here’s the idea

    • Most viruses can clear a low bar.
    • Only the strong will clear the high bar.
    • Brad Walker
    • (American Record Holder)
    • Some guy
    • (Not likely to break record)
  29. What’s going on?

    Slide 30 - What’s going on?

    • The fitness hypothesis makes two key predictions
    • An interaction with infection risk
    • An interaction with reversion rates
  30. Transmission fitness linked to VL

    Slide 31 - Transmission fitness linked to VL

    • Analogue of Fig 4B
  31. The rate at which transmitted polymorphisms revert to consensus is a marker of fitness cost

    Slide 32 - The rate at which transmitted polymorphisms revert to consensus is a marker of fitness cost

    • Hypothesis: The polymorphisms transmitted to women are less fit than those transmitted to men
    • Implication: Reversion will be faster in women than in men
    • Reversion as a fitness signal
    • Figure 4A
  32. Reversion as a fitness signal

    Slide 33 - Reversion as a fitness signal

    • 7/8 features match transmission
    • P = 0.004
    • Table S2
  33. Reversion cannot explain the data

    Slide 34 - Reversion cannot explain the data

    • Figure S4
  34. Implications

    Slide 35 - Implications

    • Non-productive infection of target cells [Joseph & Swanstrom, 2014]
    • In order for general fitness to play a role, there must be a stage in which low fitness viruses infect target cells, but never establish systemic infection
    • Not “game over” once first target cell is infected
    • Window of opportunity for T-cell based vaccines
    • And for post-exposure prophylaxis (depending on kinetics)
  35. Implications

    Slide 36 - Implications

    • Reduction in viral fitness will lead to lower transmission rates
    • ART and Immune escape mutations?
    • CD4-targeted ART rollout? [Goulder, PNAS Dec 2014]
  36. Implications

    Slide 37 - Implications

    • Low biological infection risk leads to more severe disease upon breakthrough infection
    • Female vs Male
    • GUI vs Clear
    • MSM vs HSX?
    • Microbicide vs Placebo?
    • PrEP vs Placebo?
  37. Microbicide vs Placebo

    Slide 38 - Microbicide vs Placebo

    • Caprisa 004: Microbicides reduce infection
    • 1% tenofovir gel reduced infection 50% [Abdool Karim, Science 2010]
    • Are breakthrough viruses in the drug arm stronger than those in placebo?
    • p=0.01
    • Garrett, JAIDS 2014
  38. PrEP vs Placebo?

    Slide 39 - PrEP vs Placebo?

    • Partners PrEP: Drugs reduce infection
    • Oral tenofovir associated with 85% reduction (if used!) [Baeton NEJM 2012]
    • Will only see selection if transmissions actually occur despite adherence
    • If the bar is high enough, no virus makes it over and there is no selection.
  39. PrEP vs Placebo?

    Slide 40 - PrEP vs Placebo?

    • Partners PrEP: Drugs reduce infection
    • Oral tenofovir associated with 85% reduction (if used!) [Baeton NEJM 2012]
    • Will only see selection if transmissions actually occur despite adherence
    • If the bar is high enough, no virus makes it over and there is no selection.
  40. Summary

    Slide 41 - Summary

    • Fitness bottleneck at transmission
    • Features consistent with viral fitness predict transmission
    • Viral sequences and populations with high predicted fitness are more likely to establish infection
    • Possible Mechanism
    • Multiple target cells are infected, providing a substrate for fitness-related competition
    • Provides a window for drugs and T-cell vaccines to work
    • Clinical consequences
    • Anything that weakens the virus will reduce transmission rates
    • Vaccines and drugs that protect individuals from transmission may lead to more severe disease when breakthrough infection occurs
  41. Microsoft Research

    Slide 42 - Microsoft Research

    • Jonathan Carlson
    • David Heckerman
    • Jian Peng
    • Tian-Ho Lin
    • Charlie DeZiel
    • Emory
    • Eric Hunter
    • Susan Allen
    • Malinda Schaefer
    • Daniela Monaco
    • Martin Deymier
    • Zach Ende
    • Nikki Klatt
    • Dan Claiborne
    • Jessica Prince
    • U of Kwazulu-Natal
    • Thumbi Ndung’u
    • Ragon Inst
    • Todd Allen
    • Rebecca Batorsky
    • Aaron Seese
    • Oxford
    • Philip Goulder
    • John Fraser
    • UAB
    • James Tang
    • Paul Goepfert
    • Authors
    • IAVI
    • Jill Gillmour
    • Matt Price
    • Beth Israel Deaconess MC
    • Roger Shaprio
    • Rwanda-Zambia HIV Research Group
    • William Kilembe
    • carlson@Microsoft.com