The Pre-Balkhash gerbil data cover about 2% of Kazakhstan. That the early 20th century human plague cases are nonetheless predicted so well on a larger (national) scale is probably partly related to the synoptic nature of the patterns involved. Indeed, climatic fluctuations synchronize gerbil abundance over large areas, which may itself be crucial to the spread of plague , as well as making the impact on humans greater simply by generating larger areas with high infection risk. Moreover, the relative explanatory power of a purely climate-driven model, such as that derived here, is likely to increase with the spatial scale over which it is applied, as the endogenous dynamics in otherwise uncoupled populations would then cancel out, increasing the relative strength of the larger-scale climate signal in the mean density. Human population density patterns (which reflect landscape productivity) also suggest that a large proportion of cases should occur in the southeast part of the country, in the region of the study site.
The impact of sylvatic plague on humans may operate through several scales and mechanisms [10, 16, 20]. First, it is obvious that more infected rodents and fleas are likely, everything else being equal, to translate into a greater number of humans coming into contact with the bacterium. Further, the more primary human infections there are, the more likely it is for an epidemic to start and for a particularly virulent mutant to establish in humans. However, everything may not be equal, as plague activity feeds back on itself through evolutionary processes. This may impact the resulting epidemiology when secondary hosts are infected and leads us to suspect that epidemics in naïve hosts could be more virulent when occurring after a prolonged period of high plague activity, even though (or indeed because) increased resistance in the primary hosts is also being selected for. This might suggest why the current pandemic is less virulent than those of centuries past.
No quantitative records of plague prior to 1904 are available for the region. But the 11th to 14th century period of high plague activity predicted by our study site coincides with plague outbreaks in Kaifeng and other parts of China from circa 1320 to the 1350 s [35, 36] and notably with an outbreak near Lake Issyk-Kul (about 200 km from Pre-Balkhash; Figure 1) in 1338-1339 , the discovery of which suggested to archaeologists that the Black Death pandemic originated in this region .
The associations between the climatic indexes, predicted plague and NDVI suggest that periods of high plague activity should have a degree of correlation with increased precipitation and thus productivity in water-limited grasslands, as also suggested by earlier studies in Kazakhstan  and the USA . Historical records reflecting the relative strength of the Han Chinese states versus neighboring pastoralist groups  are consistent with periods of high productivity preceding the two major plague epidemics (Figure 3d; see also Methods and Additional file 1). Resource scarcity is often assumed to be a primary driver of historical conflicts , and poor grazing conditions during cold periods probably forced Central Asian nomads south. However, territorial expansions and (successful) warfare depend not only on population size relative to its resources but also on the absolute size of the population from which armies can be recruited. We thus observe that the two main periods of border expansion, migration and warfare by Central Asian nomad pastoralists  found in Chinese records, and known in European history from the Hun invasions of the 5th century and the Mongol expansions of the 13th, are consistent with periods of high productivity in Central Asian grasslands having occurred prior to the great plague pandemics (Figure 3d). Also, Mongol expansions  might explain why plague appeared in Kaifeng (then recently occupied by Mongols) but not in Europe in the 1200 s, whereas during the 1300 s, after the Mongol conquest of Eastern Europe, plague appeared both in China and Europe .
Not much is known about the evolutionary rate of Y. pestis virulence in its natural reservoirs. But on a theoretical basis, we may suspect that periods of high plague activity (i.e., high host densities and transmission probabilities) select for more virulent types, and periods of low activity weed out strains killing the host too quickly . Thus, it might be noteworthy that the centennial scale minima and maxima (i.e., averaging over moving windows of from 2 to 400 years in length) suggest essentially three periods of higher virulence having been selected for, and that these periods are consistent with the three large pandemic periods (Figure 3d). Moreover, depending on scale and the relative importance of virulence 'retreats' (i.e., plague minima) versus 'booms' (i.e., maxima), the Third Pandemic started with a markedly lower expected virulence than the other two and would therefore perhaps not have spread so far out of its original habitat had it not been for human activities. Interestingly, the largely pneumonic and virulent Manchurian epidemics of 1910-1911 and 1920-1921  did appear during a short peak of high expected virulence.
There are some peaks that do not correspond to known plague events. These may point to weaknesses in the data or model or to unrecorded events (both likely, considering the paucity of data back in time), but it should also be noted that high plague activity in the reservoir is likely (as for other zoonoses) to be a necessary but not sufficient condition for large human outbreaks: plague in humans cannot be sustained without plague activity in the reservoir, but we do not expect every peak in rodents to generate a human epidemic. Moreover, there is no instantaneous link between Europe (or China) and the part of Central Asia with which our models are concerned. Thus, mechanisms for long-distance transmission by human or nonhuman hosts are also necessary, and our model of course cannot account for the local dynamics of plague foci persisting in Europe after the first waves of infection by some (presumably) virulent strain(s) of Y. pestis. However, as plague evidently has not been able to persist indefinitely in European hosts, it is notable that plague died off in rural Europe under periods of decreasing predicted plague peaks and virulence in Central Asia, though several European cities continued to act as independent plague foci for a considerable time, possibly due to the urban rodent reservoirs . Thus, the apparent patterns are striking enough on several scales to suggest that the plague system has indeed been influenced by climate fluctuations over the past 1500 years and that this has had consequences for human history.
Even though the first, controversial isolates of Y. pestis DNA in European plague victims appear to have been satisfactorily replicated [1, 40–45], there is ongoing debate about whether Y. pestis was a major cause of Justinian's plague and the Black Death . Our findings suggest a consistent impact of climate on modern plague, are consistent with Y. pestis being present in medieval plague victims [40, 44, 45], and tentatively suggest that changes in host densities would select for fluctuations in pathogen virulence consistent with the historical trends. However, we must also caution that our climate-forcing models would arguably apply to other Central Asian zoonoses that could potentially have emerged and spread due to the same climate forcings.
Looking towards the future, plague is still a concern because of its potentially disrupting effects on local health systems and on international trade, and because of the risk of epidemics that are difficult to control, especially in dense human settlements with poor public health provisions [7, 34]. While Y. pestis is currently not a major human pathogen, its status as a reemerging disease, combined with growing risks of antibiotic resistance  and its potential use as a bioweapon , makes plague a disease worth studying for its own sake. In addition, its reputation has motivated surveillance and research on scales that makes it a valuable model system for understanding zoonotic diseases in the past, present and future. The predicted regional climate changes are highly uncertain, but we note that increasing temperature (T) and monsoon (S) proxy evidences, and lower than average glacial accumulation, are related to higher plague estimates (Y). In other words, our results suggest that when the gerbils' dry habitats experience higher than average rainfall and warm springs, large plague outbreaks in gerbils, and high plague risk in humans, tend to follow. During the past decades, glacial accumulation (G) has been increasing and monsoon strength (S) has been about average, but the tree-ring temperature index, which is statistically the most closely and positively related to plague dynamics, has shown a rising trend which is expected to continue , and glacial accumulation may also fall as temperature rises . Hence, predicted continuation of global warming associated with an increase of the hydrological cycle may enhance the probability of human plague outbreaks in Central Asia, especially if socioeconomic instability undermines the efforts of plague prevention in less developed though highly populated regions.
However, much work remains to be done. The epidemiological dynamics of the nonhuman hosts and the arthropod vectors are incompletely known, as are the differences between bacterial strains. The most common vectors have been studied in detail, but the sylvatic patterns are sometimes puzzling, suggesting some unknown mode of long-range dispersal or an unrecognized mode of persistence in the environment. Interactions with the habitat structure of the rodent host community may reflect some of the mechanisms through which climate, topography and vegetation (it is hard to avoid the impression that forest belts seem to act as boundaries) seem to form a barrier to plague occurrence, a barrier that seems to break down under some circumstances to allow the large-scale epidemics of history.