Inferences on the acquisition of multidrug resistance in \emph{Mycobacterium tuberculosis} using molecular epidemiological data

G. S. Rodrigues, Andrew Francis, Scott A. Sisson, Mark M. TanakaarXiv 2017 · DOI · arXiv

TL;DR

Researchers used Approximate Bayesian Computation (ABC) to model how multidrug-resistant tuberculosis (MDR-TB) spreads in Bolivia. Their findings reveal that the vast majority of MDR cases are driven by the transmission of existing resistant strains rather than new resistance evolving within patients, and that resistance to isoniazid develops significantly faster than to rifampicin.

What Problem Does This Paper Address?

Multidrug-resistant tuberculosis (MDR-TB)—defined as resistance to the two most powerful "first-line" drugs, rifampicin and isoniazid—is a global health crisis. Public health officials need to know: is resistance primarily "brewing" inside individual patients due to treatment failure, or is it spreading from person to person?

Answering this is difficult because epidemiological data is usually just a "snapshot" in time. We see the current resistant strains, but we don't see the decades of transmission and mutation that led to them. Furthermore, the mathematical models required to describe these processes are so complex that standard statistical methods (like maximum likelihood estimation) become computationally impossible to solve.

Methods

The authors developed a discrete-time stochastic model that simulates a TB epidemic over 50 years. The model tracks:

  • Transmission: How the bacteria move between susceptible and infected individuals.
  • Treatment: The transition from untreated to treated states.
  • Evolution: How bacteria acquire resistance to one or both drugs.
  • Genetic Fingerprinting: The mutation of "Variable Number Tandem Repeat" (VNTR) loci, which act as molecular markers to track specific bacterial lineages.

To bridge the gap between their complex simulations and real-world data, they used Approximate Bayesian Computation (ABC). Instead of calculating a traditional likelihood function, ABC works by running thousands of simulations with different parameters and keeping only those that produce "synthetic" data closely matching the observed Bolivian dataset. They used a two-stage importance sampling approach and neural network regression to refine their estimates.

Datasets

The study utilizes a molecular epidemiological dataset from Bolivia (Monteserin et al. 2013).

  • Sample Size: 100 bacterial isolates collected in 2010.
  • Features: Each isolate includes a drug resistance profile (Sensitive, INH-resistant, or MDR) and a genetic fingerprint based on 15 VNTR loci.
  • Composition: 78 were sensitive, 8 were resistant only to isoniazid (INH), and 14 were MDR. Notably, no isolates were resistant only to rifampicin (RIF).

Key Insights and Ideas

  • Transmission is the Engine: The most striking insight is that over 99% (median estimate) of MDR-TB cases in the sample resulted from the transmission of already-resistant strains, not from new resistance developing within the current patients.
  • The "Gateway" Drug: Resistance to isoniazid (INH) is acquired roughly an order of magnitude faster than resistance to rifampicin (RIF). This confirms laboratory observations at an epidemiological scale.
  • Direct MDR Acquisition: The data supports a "direct" route to multidrug resistance. While we often think of resistance as a step-by-step process (Drug A then Drug B), the model suggests bacteria can effectively become MDR in a single step within a host, likely because resistant subpopulations exist at low levels even before treatment begins.
  • The Cost of Resistance: Even though resistant strains have a "fitness cost" (they transmit about 20% less effectively than sensitive strains), they still dominate the MDR landscape through sheer transmission volume.

Key Results

  • Mutation Rates: The estimated rate for acquiring INH resistance ($\rho_{INH} \approx 1.14 \times 10^{-3}$) was nearly seven times higher than for RIF resistance ($\rho_{RIF} \approx 1.67 \times 10^{-4}$).
  • Source of MDR:
    • Transmission: 99.75% (Median contribution).
  • Single-step evolution: 0.23%.
  • Two-step evolution: 0.00%.
  • Model Comparison: By comparing "Submodels," the authors showed that a model assuming equal rates for INH and RIF resistance failed to accurately recreate the observed data, reinforcing the asymmetry of TB drug resistance.

Limitations and Future Work

  • Snapshot Data: The study relies on a single point in time. Longitudinal data (samples taken over several years) would provide much stronger constraints on the model.
  • Simplified Latency: The model disregards "latent" TB infections (where the bacteria are dormant) to keep the simulation manageable. Since TB can stay dormant for years, including this could change the timing of the evolutionary "clock."
  • Fixed Parameters: Several variables, such as the 10% fitness cost per drug resistance, were fixed based on previous literature rather than estimated directly from the Bolivian data.
  • Population Size: The model assumes a constant population of 10,000 susceptible individuals; scaling this to a national level would require significantly more computational power.

Summary updated

Abstract

We investigate the rates of drug resistance acquisition in a natural population using molecular epidemiological data from Bolivia. First, we study the rate of direct acquisition of double resistance from the double sensitive state within patients and compare it to the rates of evolution to single resistance. In particular, we address whether or not double resistance can evolve directly from a double sensitive state within a given host. Second, we aim to understand whether the differences in mutation rates to rifampicin and isoniazid resistance translate to the epidemiological scale. Third, we estimate the proportion of MDR TB cases that are due to the transmission of MDR strains compared to acquisition of resistance through evolution. To address these problems we develop a model of TB transmission in which we track the evolution of resistance to two drugs and the evolution of VNTR loci. However, the available data is incomplete, in that it is recorded only {for a fraction of the population and} at a single point in time. The likelihood function induced by the proposed model is computationally prohibitive to evaluate and accordingly impractical to work with directly. We therefore approach statistical inference using approximate Bayesian computation techniques.
BibTeX
@misc{rodrigues2017inferences,
  author = {G. S. Rodrigues and Andrew Francis and Scott A. Sisson and Mark M. Tanaka},
  title = {{Inferences on the acquisition of multidrug resistance in \textbackslash\{\}emph\{Mycobacterium tuberculosis\} using molecular epidemiological data}},
  year = {2017},
  eprint = {1704.04355},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1704.04355},
}
Figure