Thymic involution and rising disease incidence with age

Abstract

For many cancer types, incidence rises rapidly with age as an apparent power law, supporting the idea that cancer is caused by a gradual accumulation of genetic mutations. Similarly, the incidence of many infectious diseases strongly increases with age. Here, combining data from immunology and epidemiology, we show that many of these dramatic age-related increases in incidence can be modeled based on immune system decline, rather than mutation accumulation. In humans, the thymus atrophies from infancy, resulting in an exponential decline in T cell production with a half-life of ∼16 years, which we use as the basis for a minimal mathematical model of disease incidence. Our model outperforms the power law model with the same number of fitting parameters in describing cancer incidence data across a wide spectrum of different cancers, and provides excellent fits to infectious disease data. This framework provides mechanistic insight into cancer emergence, suggesting that age-related decline in T cell output is a major risk factor.

Keywords: T cell; cancer; driver mutations; infectious disease; thymus.

Fig. 1.

Declining T cell production leads to increasing disease incidence. Our model assumes that immunogenic cells arise with the same probability at any age, and, after a period of being targeted, the population may overwhelm the immune system by crossing an immune escape threshold. This threshold is assumed to be proportional to T cell production, which decreases with age. This provides a prediction for the possible forms of disease incidence curves.

Fig. 2.

Infectious disease incidence. Log-linear plots of incidence (per 100,000 person-years) by age group for all ABC bacterial infections, West Nile virus (WNV) disease, and Influenza A, ordered from best fit to worst. Bacterial and viral diseases are shaded yellow and green, respectively. The two-parameter IM-II is in red, while the one-parameter IM-I is in orange. Incidence often decreases initially from birth due to an underdeveloped immune system in infants; therefore, models are fitted only to data points for ages greater than 18 y. Error bars show 99% CIs for all diseases.

Fig. 3.

Cancer incidence. Log-linear plots of incidence (per 100,000 person-years). Data taken from SEER (11). (A and B) Some cancer types rise exponentially fitting IM-I (A), while some cancer types rise like power laws, although can still be fit by IM-II (B). Fitting curves for IM-II and PLM are shown in red and green, respectively. (C) The top 20 best-fitting incidence curves as measured by Akaike Information Criterion (AIC) for IM-II. (D and E) Universal scaling functions for all cancers with defined pivot ages (84 out of 101 cancer types) plotted in gray with the top 20 incidence curves highlighted. Data shown for both genders (D) and gender-separated data (E), with dotted lines showing the model predictions for IM-I and IM-II. The gender-separated curves are fitted with higher independently determined values for α in males than females, reflecting the gender bias in T cell production (Methods). A purely exponential incidence curve would correspond to a pivot age of negative infinity, and therefore, for the purposes of plotting, we set a minimum pivot age of −50 y. Models are fitted only for ages greater than 18 y. Error bars show 95% CIs for all diseases.