SCOTTI: Efficient Reconstruction of Transmission within Outbreaks with the Structured Coalescent

• SCOTTI combines genetic information from infection samples with epidemiological information about patient exposure to infection.Source: Author Summary · p. 1
• SCOTTI efficiently accounts for within-patient genetic variation and non-sampled patients, including asymptomatic patients.Source: Author Summary · p. 2
• These features enable accurate transmission reconstructions in complex scenarios and can help identify and limit transmission sources and routes and prevent disease spread.Source: Author Summary · p. 2

Introduction

• Threshold-based genetic methods cannot generally distinguish direct transmission from intermediate-host transmission or infer direction, while pathogen phylogenies can mislead when within-host variation creates discordance.Within-host diversity can make transmitted isolates differ from those sampled from the same host and can bias infection-time estimates; incomplete bottlenecks, non-sampled hosts, and multiple infections can also obscure transmission history.Source: Introduction · p. 2
• Many joint transmission-and-evolution methods use epidemiological dates but ignore within-host variation, while methods that model within-host diversity often assume every outbreak host was detected and sequenced.These assumptions can be impractical when host detection or sequencing is incomplete or uncertain.Source: Introduction · pp. 2, 4

SCOTTI

• SCOTTI accounts for within-host diversity, non-sampled hosts, and multiple infections by modelling each host as a pathogen population and transmission as migration.It uses an approximation to the structured coalescent and is implemented as an open-source package for BEAST2; it does not model transmission bottlenecks, which may disadvantage it with small inocula but may help with large inocula.Source: Introduction · p. 4
• SCOTTI typically showed higher accuracy than Outbreaker on simulated data, while the two methods usually gave very different interpretations of outbreak dynamics.The paper also applies the methods to FMDV and Klebsiella pneumoniae outbreak datasets.Source: Introduction · p. 4

SCOTTI: A new Approach to Reconstructing Transmission Events

• SCOTTI treats each host as a pathogen population, lineage migration as transmission, and uses BASTA approximations to reduce computational demand.Simulations used a multispecies coalescent model with transmission bottlenecks, while inference used the structured coalescent.Source: SCOTTI: A new Approach to… · pp. 4, 6

Accuracy of Inference on Simulated Data

• SCOTTI exceeded Outbreaker across simulated scenarios except early sampling, and its 95% credible sets contained true sources 90–100% of the time.With early sampling, SCOTTI’s lack of transmission bottlenecks led it to over-assign non-sampled sources.Source: Accuracy of Inference on Simulated Data · pp. 6, 9, Figure 4 · p. 8

Analysis of FMDV and Klebsiella pneumoniae outbreaks

• In the FMDV outbreak, SCOTTI’s highest-probability transmission origins matched prior findings and assigned ≈66% posterior probability to non-sampled intermediates linking the August and September clusters.Outbreaker instead inferred some direct links that conflicted with exposure data and earlier transmission findings.Source: Analysis of FMDV and Klebsiella… · p. 10, Figure 5 · p. 11
• In the K. pneumoniae outbreak, SCOTTI inferred non-sampled sources for most patients and most sampled-case common ancestors, unlike Outbreaker’s sampled-host transmission picture.SCOTTI inferred separated but related sub-outbreaks; four gaps in host exposure intervals implied non-sampled intermediates, recurrent introductions, or environmental contamination.Source: Analysis of FMDV and Klebsiella… · pp. 10, 12

Discussion

• Across broad simulations, SCOTTI was more accurate overall than Outbreaker, particularly when using multiple samples per host, which Outbreaker does not allow.Multiple samples can inform within-host coalescent rates and potentially the proportion of mixed infections.Source: Discussion · p. 12
• By modeling each host as a pathogen population and transmission as lineage migration, SCOTTI inferred transmission with good accuracy despite within-host variation and non-sampled hosts.Its structured-coalescent framework explicitly models within-host evolution and non-sampled hosts.Source: Discussion · pp. 12, 14
• Partial inaccuracies in exposure-time data modestly reduced SCOTTI’s accuracy, while absent exposure information is expected to compromise it more deeply than Outbreaker.SCOTTI uses exposure times to identify plausible direct transmissions and rule out those with non-overlapping exposure times.Source: Discussion · p. 12
• In the K. pneumoniae outbreak, SCOTTI and Outbreaker yielded different transmission interpretations and estimates of non-sampled hosts’ contributions.Figure 6 shows different inferred transmission trees, including SCOTTI’s reconstruction with non-sampled hosts.Source: Figure 6 · p. 13, Discussion · p. 14
• SCOTTI ignores transmission bottlenecks, which may bias inference with strong bottlenecks and early sampling, but may suit outbreaks with mixed infections and large inocula.Future extensions could incorporate infection-time priors, compartmental epidemiological models, and host geography or structure.Source: Discussion · p. 14
• SCOTTI addresses incomplete host sampling, mixed infections, and within-host variation, supporting transmission-history reconstruction across bacterial and viral outbreaks.Such histories can inform strategies to fight infectious-disease spread.Source: Discussion · pp. 12, 14

Approximate Structured Coalescent Model

• SCOTTI models each host as a pathogen population and transmission as migration, using BASTA’s approximation to infer transmission histories.The model conditions on user-provided host exposure intervals; non-sampled hosts can have unlimited exposure, and migration rates and effective population sizes are assumed equal across exposed hosts.Source: Approximate Structured Coalescent Model · pp. 14–15
• SCOTTI approximates the genealogy probability by integrating over migration histories and assumes lineages migrate independently between events.It updates lineage-host probabilities across intervals and accounts for host introductions and removals; uniform migration rates and population sizes reduce computational demand.Source: Approximate Structured Coalescent Model · pp. 15–17

Simulations of Pathogen Evolution

• Simulations used two outbreak transmission histories, weak or strong transmission bottlenecks, and one or two samples per host across 64 settings.Each setting produced 100 datasets; SCOTTI and Outbreaker were assessed by how often they recovered the simulated infection origin as the highest-posterior origin and by its average posterior probability.Source: Simulations of Pathogen Evolution · pp. 17–19
• The simulation model varied sampling time, alignment length, missing hosts, and accuracy of SCOTTI’s exposure intervals.The simulated transmission bottleneck was not modeled in either SCOTTI or Outbreaker.Source: Simulations of Pathogen Evolution · p. 18

Foot and Mouth Disease Virus and K. Pneumoniae data

• SCOTTI and Outbreaker were applied to the 2007 UK FMDV and 2011–2012 Nepali hospital K. pneumoniae outbreaks.SCOTTI used 10^8 MCMC iterations and zero to two non-sampled hosts; Outbreaker used only the oldest sample for multiply sampled hosts, and convergence was checked for both methods.Source: Foot and Mouth Disease Virus and K.… · p. 19

S1 Text

• Supplementary Text provides further details of the methods and analyses, including supplementary figures and tables.Source: S1 Text · p. 19

S1 Data

• Supplementary Data is a file containing information to replicate the results.Source: S1 Data · p. 19

Graphical representations

• Nested trees combine host exposure intervals, transmission events, sampled sequences, and pathogen phylogeny; beanbag and Maypole trees emphasize transmission alone or transmission alongside evolution.Beanbag trees show donor-to-recipient arrows, while Maypole trees color phylogenetic lineages by host; standard phylogenetic trees relate sampled sequences.Source: S1 Text · p. 24

Simulation results

• SCOTTI shows higher accuracy than Outbreaker in simulations that include one or three non-sampled hosts.The scenarios include two simulated transmission histories and vary the number of non-sampled hosts.Source: S1 Text · p. 28
• With more genetic variation and thus phylogenetic signal, both SCOTTI and Outbreaker show increased inference accuracy, while SCOTTI has overall higher accuracy than Outbreaker.Simulations include long infection times averaging 10 rather than 2 Ne generations and abundant genetic data of 15,000 rather than 1,500 base pairs.Source: S1 Text · p. 29
• When exposure intervals are twice their true length, SCOTTI is not considerably affected and remains more accurate than Outbreaker.The simulations use inaccurate introduction and removal times in two transmission histories.Source: S1 Text · p. 30
• With late sampling near clearance, SCOTTI is more accurate than Outbreaker; with early sampling near infection, SCOTTI’s accuracy falls below Outbreaker’s.The early-sampling condition produces a noticeable decrease in SCOTTI accuracy, while Outbreaker has high error rates under late sampling.Source: S1 Text · p. 31