Multiroute pathogen transmission is different

Abstract

Pathogens can spread via multiple transmission routes. The epidemiology of such pathogens exhibits rich complexity and produces counterintuitive dynamics.

Epidemics are complex systems. They emerge from the interaction of host, pathogen, and environment. Consequently, since at least the time of Bernoulli, scientists and policy-makers have relied on data-driven models to understand how, when, and why small outbreaks grow into epidemics. Such models are necessary because the kinds of nonlinear, context-dependent, and unstable dynamics associated with epidemics—indeed all complex systems—are challenging to intuit. When pathogens transmit via multiple routes, e.g., sexual transmission and needle sharing, gaining an intuition for their dynamics without the use of models goes from challenging to nearly impossible.

The more we look, the more it seems multiroute transmission is the rule, not the exception, for pathogens. From well-established examples like HIV and hepatitis C to more recent discoveries, including Zika, Ebola, and mpox viruses, the list of pathogens that transmit readily via at least two routes is growing rapidly (1). Despite robust empirical evidence for the prevalence of multiroute pathogens, most of our epidemiological models focus on single routes. Understanding why this modeling gap matters, and why models with two or more transmission routes behave so differently from those with only one, requires new modeling approaches.

When studying infectious disease transmission, epidemiologists typically report a quantity called the basic reproduction number, written as R0, and pronounced R-naught. The R0 for a pathogen is the expected number of secondary infections resulting from a single infectious individual who is introduced into an otherwise entirely susceptible population. If this number is greater than 1, then in an infinitely large population, we would expect an epidemic (typically defined as an infinite number of infected individuals connected to the initial case). Despite its near-universal application, R0 does not really tell us much about how, when, and why small real-world outbreaks grow into large real-world epidemics (2). Understanding the shortcomings of R0 can help us better understand the unintuitive nature of multiroute pathogen transmission.

Growing from a small number of cases into a large number of cases requires quite a bit of luck on the part of the pathogen. This luck typically comes from one of two places: either lots of introductions (violating a key assumption underlying R0) or superspreading (as we explain below, also violating an assumption behind R0). For a multiroute transmission pathogen, one of the routes will typically generate more of this “luck” than others. However, once established, other transmission routes can become more prevalent (1, 3). The cartoon in Fig. 1A shows how repeated introductions and/or superspreading via one transmission route can facilitate future spread via secondary transmission routes.

Fig. 1. Pathogen transmission activity.

(A) Infectious individuals move from population i, which is experiencing a large epidemic, into population j, which is not. Blue squares: Infections via route a. Orange triangles: Infections via route b. Three of the introductions lead to only a few additional cases, but the fourth results in a superspreading event that seeds many new infections, including a considerable number via route b. (B) Simulated distributions of secondary infections with a fixed R0 and increasing levels of superspreading. To model superspreading, we sample from a negative binomial distribution with mean (R0) and an overdispersion parameter. Values for the overdispersion span the range observed in naturally occurring human pathogens. (C) Time series of new cases resulting from the introductions and transmission changes plotted in (A). Even though the transmission rate via the orange triangle route appears much less than that of the teal square route, a switch occurs between dominant transmission routes.

How does superspreading violate the assumptions behind R0? Only a small number of infected individuals can be associated with superspreading events. If everyone caused a large number of infections, it would just be average spreading. Said differently, superspreading leads to substantial variability in secondary infections. The bar plots in Fig. 1B show the change in the maximum, median, mode, and mean number of simulated secondary infections as the R0 is held constant, but the rate of superspreading is increased. Remembering that R0 is the mean number of secondary infections, note that for high amounts of superspreading, the mean does not tell us much about the shape of the distribution. The pace and tempo of cases, the probability of transmission chains going extinct, and many other aspects of the dynamics are completely different when superspreading occurs. For example, in the case of sexual transmission, the distribution of secondary infections can be very heavy-tailed, often following a finite-size power law, where the role of the second moment can be more important than the first moment, i.e., the R0, when quantifying epidemiological risk. Coupling superspreading to a second transmission route will result in hard-to-intuit dynamics.

If you are starting to see how repeated introductions and superspreading could play similar epidemiological roles from the perspective of multiroute transmission dynamics, then you are building up the right kind of intuition. Superspreading can seed multiple, largely independent transmission chains. In Fig. 1C, we see the time series of news cases for the scenario illustrated in Fig. 1A (a pathogen spreads through introductions and via two transmission routes, one of which results in superspreading). Occasionally, a superspreading event will occur. The resulting secondary infections will each start chains of transmission. While each of these individual chains is unlikely to generate many infections, given enough spins of the roulette wheel, there will be large clusters. One outcome is that the secondary, less transmissible, route can come to dominate.

In this issue of Science Advances, Miura et al. (4) develop a multiroute transmission model to study mpox transmission in the Democratic Republic of the Congo (DRC). Similar to Zika and Ebola, sexual transmission of mpox was a recent discovery (5). Unlike the other two viruses, scientists now believe the dominant transmission route for mpox is sexual contact (5). Further confounding our understanding of mpox is that the age distribution of cases and perceived severity seems to shift regularly. These changes have led researchers to hypothesize roles for waning immunity following cessation of smallpox vaccination (which provides cross-protection against mpox), evolutionary changes in the mpox virus to exploit alternative transmission routes, and/or shifts in human behavior (6). A possibility explored by Miura et al. is that the multiroute nature of mpox transmission alone could account for the observed epidemiological patterns.

Using their multiroute transmission model, Miura et al. demonstrate that the dominant transmission route associated with mpox can shift from sexual to nonsexual transmission without requiring evolutionary changes in the pathogen or behavioral changes on the part of the host. In their model, the global epidemic dynamics of mpox are sustained by less frequent superspreading events generated via sexual transmission that seed many independent transmission chains of nonsexual transmission. Given sufficient time, and with enough immunity from infection, sexual transmission begins to fade away, while nonsexual transmission persists (if only because it started later). Eventually, both kinds of transmission will stop, but there will be a period when nonsexual transmission dominates (even though the R0 for nonsexual transmission is likely far below 1).

Mathematical epidemiologists have previously reported similar shifts in the dominant transmission route and on the counterintuitive nature of multiroute transmission (1, 3, 7). What distinguishes the work of Miura et al. is the real-world data they used to parameterize their models. Although not formally fit to data, they used empirical information on both sexual and nonsexual contact networks from the DRC and recently measured transmission parameters for mpox to constrain the state space of their analysis. The shape of the sexual contact network, and the resulting potential for superspreading, is critically important. As a result, their conclusion, that the dominant transmission route for mpox should change from sexual to nonsexual and that the result will be an increase in the proportion of nonadult cases, carries significant weight from both a basic science and public health intervention perspective.

Three findings from Miura et al. stand out for the broader epidemiological modeling and public health communities. First, the presence of multiple transmission routes can result in scenarios in which multiple mechanisms coexist, even if, individually, the routes would not persist. Second, incorporating real-world social network data does not just make models more realistic; it also leads to entirely new kinds of epidemiological scenarios. Finally, public health interventions are impacted from the work in Miura et al. As discussed in the paper, it may be difficult to control outbreaks once the dominant transmission route switches from sexual to nonsexual. The intuition here is that the nonsexual transmission will be quite spread out in space and, just as importantly, vaccinating high-risk individuals, e.g., with large numbers of sexual contacts, will be of limited effect because sexual transmission has already slowed or stopped due to the buildup of infection-derived immunity. Sexual transmission slows because of an inverse “rich-get-richer” problem. Highly connected individuals are likely to get infected earlier in an epidemic.

Building on Miura et al., future studies should examine whether results based on data from the DRC generalize to other settings. Conducting such research relies on high-quality social network, mobility, and sociodemographic data, highlighting the need for sustained investment in the generation and curation of such datasets. As it relates to mpox, it will be important to leverage counterfactual modeling approaches and statistical/machine learning methods to strengthen the evidence for a switch in the dominant transmission route that is not driven by evolutionary, socio-behavioral changes, and/or confusion with other pathogens like chicken pox. Ideally, such efforts will integrate genomic data from the mpox virus and incorporate waning immunity from both past mpox infection and smallpox vaccination (8).

As we continue to collect increasingly high-resolution molecular and socio-behavioral data on infectious disease outbreaks, we will undoubtedly find more evidence of multiple–transmission route pathogens. It seems plausible that in the near future, most infectious disease epidemiological studies will need to consider multiple transmission routes. Recent modeling efforts have begun advancing the kinds of mathematical and computational tools that more fully embrace epidemics as complex systems and that we will need to study multiroute transmission pathogens (9–11). Continuing to develop new modeling approaches and challenging models with high-resolution, real-world data is more important than ever in our study of pathogen transmission. As such, advancing both the science and practice of epidemiology requires international, multisector collaboration and the work of multidisciplinary teams.