Coffee leaf rust (CLR), caused by Hemileia vastatrix Berk. & Br., is a fungus in the order Pucciniales that is mainly dispersed by urediniospores, though other spores types have been observed (Talhinhas et al. 2017). Although H. vastatrix most strongly impacts the species Coffea arabica, all other cultivated species of Coffea can be impacted (Waller 1982). Since its first recorded outbreak in Ceylon in 1869, this phytopathogen has caused significant outbreaks in major coffee-growing regions around the world (McCook 2006). Most recently, a series of outbreaks spread through Central America and the Caribbean starting in 2012, with lasting impacts on production still visible in 2016 (Cerda et al. 2017). This reduced production and led to significant economic impacts, costing Central America over $600 million by 2014 and impacting the livelihoods of thousands of smallholder farmers and harvesters (McCook and Vandermeer 2015). In Guatemala, smallholder farmers lost an average of 71% of their crop, leading to a near doubling of migration rates out of coffee-growing regions (Dupre et al. 2022). Concerningly, in 2020, CLR was found in Hawaii, formerly the last major coffee-growing region without CLR (Keith et al. 2022).

Mechanisms that control the dynamics of local CLR spread can play a critical role in determining the intensity of epidemics on a larger spatial scale (Li et al. 2022; Mora Van Cauwelaert et al. 2023). While wind can play a role in dispersal (Bowden et al. 1971; Becker and Kranz 1977), rain is commonly identified as an important factor over many studies (e.g., Merle et al. 2020; Lasso et al. 2020, and reviews within). Rainfall plays a key role in local spread by transporting spores via several mechanisms and providing the moisture necessary for spore germination (Fitt et al. 1989). Rainwater splash, contact, or impact on infected leaves, have all been demonstrated to release spores from sporulating lesions (Nutman et al. 1960; (Rayner 1961a, b; Boudrot et al. 2016). Rainwater flow can disperse spores within a leaf or between leaves on the same plant (Bock 1962), while rain splash can disperse spores to other leaves and neighboring plants (Becker and Kranz 1977).

Rain splash from infected leaf litter and other plant debris can be a way of inoculum dispersal in plant-pathogen systems (Fitt et al. 1989; Konan and Guest 2002; Paul et al. 2004; Gomez et al. 2007; Rossi and Caffi 2012), and has been investigated for Mycena citricolor, which causes American leaf spot in coffee (Granadoes-Montero et al. 2020), but has not been investigated for CLR. Understanding the potential for CLR infection from leaf litter is important because leaf drop occurs naturally, as well as being induced by the disease (Waller 1982; Avelino et al. 2015). Though defoliation may decrease the inoculum on the plant, infected leaf litter could potentially serve as a stock for new dispersal, though this has not been demonstrated for this disease. Here, we report on a semi-controlled infection experiment testing whether simulated rain splash from infected leaf litter can transmit H. vastatrix spores to susceptible uninfected coffee plants (CLR litter treatment), as compared to paired plants with no infected leaf litter (control treatment).

We conducted the experiment from March to August 2022 in an outdoor laboratory of a field station (Finca Gran Batey) located in Utuado, Puerto Rico. Local average monthly temperature over the experiment period was 19.9 °C (1.3 S.D) and average monthly relative humidity was 84.4% (6.1 S.D.) (TWC 2023). We selected 24 uninfected C. arabica coffee seedlings (approximately 6 weeks old at the start of the experiment) of the susceptible Caturra variety and established them in separate 3-gallon (13.6 L) pots of 4:1 potting soil and compost mix. We grouped plants into 12 pairs of control and CLR litter treatment individuals. Plants had an average of 7.6 leaves (± 0.15 standard deviation) over the duration of the study and treatment groups did not differ significantly in total number of leaves (t = 1.740, df = 299, p = 0.08). Plant pairs were arranged in six rows of two pairs, with each pot center spaced 50 cm from row neighbors (Fig. 1). We rotated plant pair locations and individual plant positions within pairs weekly to homogenize environmental biases. All plants spent time within the inner core and outer edge of the arrangement, and each pair cycled along all sides of the experiment area.

Fig. 1
figure 1

Experimental setup. Picture on the left represents the rain shelter above the experimental plants. A sprinkler below the shelter roof was used to simulate rain events to test splash dispersal from infected leaf litter. The diagram to the right shows the arrangement of treatment plant pairs. Treatment pairs consisted of healthy plants with leaf litter infected by coffee leaf rust and healthy plants with clean leaf litter as control (filled and empty shading, respectively). Plant locations were changed weekly by shifting rows and switching pot orientation within pairs

Full size image

As experimental treatments, we placed either infected or uninfected coffee leaves on the soil surface of each pot. These leaves were harvested from plants outside the experiment, on a nearby farm. For the CLR litter treatment, we selected leaves that exhibited characteristic orange pustules with visible CLR spore clusters, which indicate potential to disperse viable spores. For the control plants, we used healthy leaves from uninfected bushes. We placed four leaves in each pot, arranged in a square around the central plant stem. We fixed these in place with bent wire to keep the underside of the leaf always facing upwards, which ensured that infected leaves had their lesions facing towards the plant. We checked leaves weekly and replaced leaves when they began to decay or when the infected lesions appeared to be washed clean of their spores. We also removed weeds and any other fallen leaves during this time.

We subjected plants to simulated rain events meant to recreate high-splash conditions, to document if CLR spread from leaf litter is possible. The plants were housed under a 5 m x 5 m mesh shelter that simulated rainfall within the shelter using two sprinkler heads installed beneath the roof and positioned approximately 3 m above the ground (Fig. 1). Sprinklers were set on a timer to turn on for 10 min each morning, with an approximate intensity of 80 mm.hr− 1. This value falls within the expected intensity for 10-min rainfalls historically occurring more than once a year for Utuado (Bonnin et al. 2008), and is comparable to the upper values of other splash studies using simulated rain (e.g., up to 60 mm.hr− 1, Madden et al. 1996). To simulate heavy rain conditions in addition to daily simulated rainfall, a technician watered each plant with a handheld sprayer once a week, holding the stream above the plant at a height of 1 m for 10 s. The shelter also allowed natural rainfall to permeate through the mesh roof, but only as a fine mist. We monitored the number of infected and total leaves on each plant weekly for 25 weeks.

We used a survival analysis of disease onset (Scherm and Ojiambo 2004) to compare the probability that individuals in the control and CLR litter treatment groups remained uninfected over time (i.e., their survival functions). We fit Kaplan-Meier survival curves calculating the cumulative survival rate over the experiment, based on the first observation of CLR for each plant. We then tested for difference between the curves using the log-rank test. These procedures were conducted and visualized with the packages “survival” v.3.4-0 and “survminer” v.0.4.9 (Kassambara et al. 2021; Therneau 2023), respectively, in the statistical software R v.4.2.2 (R Core Team 2022).

To quantify the intensity of dispersal, we modeled the number of infected leaves by treatment as a fixed effect in a generalized linear regression model (Poisson error distribution). Leaf infection count was offset by the total number of leaves in each plant, which normalizes modeled observations as a rate of infection per available leaves. To account for correlations over repeated measures of the same plant pairs and similarity between consecutive weeks, we added a random effect offset for pair group identity and a non-linear smoother with five basis functions (i.e., internal breakpoints) to account for correlation over time (weeks). We fit the model with the R package “mgcv” v. 1.8–41 (Wood 2011), and assessed model fit with the built-in function “gam.check” and by comparing model simulations to observed data, using the package “DHARMa” v. 0.4.6 (Gelman and Hill 2007; Hartig 2019). Model-estimated marginal means and multiple comparison corrections were calculated using packages “emmeans” v. 1.8.2 (Lenth 2019) and “ggeffects” v. 1.1.4 (Lüdecke 2018).

Our results demonstrate that CLR-infected leaf litter is a significant source of infection by H. vastatrix. By the end of the 25 weeks, all CLR litter treatment plants had become infected, while four control plants remained uninfected throughout the experiment (Fig. 2). The log-rank test indicates that the survival function of the two treatment groups differed significantly (χ2(1) = 4.4, p = 0.035). The median times to infection for CLR litter and control treatment groups were 9.0 and 11.5 weeks, respectively.

Fig. 2
figure 2

Plot of Kaplan-Meier survival curves depicting cumulative probability of survival, i.e., remaining uninfected by coffee leaf rust (CLR) over time, for the two treatment groups (green/solid line = control, orange/dotted line = CLR litter). Vertical lines indicate the median survival time for each treatment. The survival functions of the two curves were significantly different (χ2(1) = 4.4, p = 0.035). Crosses indicate right-censored plants; in this case, only four plants were considered censored because they survived to the end of the study

Full size image

In our analysis of infected leaves, we excluded the first four weeks of the experiment because infections did not appear on coffee plants until after these weeks. Across the subsequent 21 weeks, the model estimated plants with CLR litter treatment had 1.83 times more infected leaves than the control treatment (95% CI: 1.50–2.23, p < 0.001), with an average infection rate of 0.18 (95% CI: 0.13–0.25), which was higher than the control treatment average of 0.096 (95% CI: 0.067–0.14). CLR infection for both the CLR litter and control treatment groups took a significantly nonlinear shape (p < 0.001) that differed between the treatments (Fig. 3). The mean of the CLR litter treatment plants increased early in the experiment period and remained greater than 25% of total leaves infected from weeks 12 to 20 and was always higher than the control treatment mean throughout the study; though there was no overlap in the 95% confidence intervals only in weeks 12 to 15 (Fig. 3). The mean CLR infection of the control plants also increased but remained below 25% leaves infected. Both treatments decreased in infected leaves towards the end of the study period. The model of CLR leaf infections had an explained deviance of 38.7%.

Fig. 3
figure 3

Evolution of leaf infection rate by coffee leaf rust (CLR) and model estimation for control and CLR litter treatments up to week 25. Points represent individual plant infection rates, shifted randomly on the horizontal axis to show overlapping points (green/filled circle = control, orange/empty circle = CLR litter). Lines and areas represent average trend for each treatment group and 95% confidence interval, as estimated by the regression model (solid line = control, dashed line = CLR litter)

Full size image

The significant association between infected leaf litter with earlier initial infection and higher leaf infection incidence strongly suggests that H. vastatrix spores can be transmitted from infected leaf litter and contribute to epidemic growth, either by spreading CLR to uninfected plants, or increasing the infection on already diseased plants. Although rainfall has been demonstrated to wash spores off leaves and decrease inoculum (Avelino et al. 2020), our findings suggest that rain can still transmit spores from the leaf litter onto uninfected leaves and significantly increase infection. As H. vastatrix enters through the stomata on the underside of leaves, spores that are splashed from below may more easily contact the vulnerable part of the leaf. This may be a potential area for further research, as we are not aware of any previous work testing the relative importance of this dispersal pathway for CLR.

The polycyclic nature of rust (Avelino et al. 2004) means that later lesions could also be due to secondary infections dispersed from plants and leaves infected earlier in the experiment. These secondary infections would only be observed after the latent period between initial colonization and the appearance of sporulating lesions. This period varies by environment and region (Waller 1982), but may be indicated in our experiment by the five weeks before the first observed lesion (Fig. 2), which would suggest that the median time to infection of the CLR litter group (9.0 weeks) occurred prior to the earliest secondary infection. Infections later in the experiment may reflect additional dispersal processes arising from the ongoing infection, such as splashes from contaminated water drops falling from infected leaves (Bock 1962) or dry-dispersed spores resulting from water drop impacts on infected leaves (Rayner 1961b; Boudrot et al. 2016). On the other hand, control plants that became infected early could suggest rain splash projected spores from the leaf litter to nearby pots, spaced at least 50 cm away. Though this study did not recreate typical coffee planting spacing, future studies could investigate these dynamics in a field setting.

The differing progression of infected leaves between CLR litter and control treatment plants (Fig. 3) provide insight into the epidemic dynamics. While the early peak in infected leaves of the CLR litter treatment may be associated with initial infections from leaf litter (Fig. 3), the delayed peak in the control treatment plants could represent a secondary outbreak caused by an increase in environmental spore load from early infections. By the end of the study, rates of infection were decreasing for both groups (Fig. 3), reflecting a dilution effect that results from plants producing new healthy leaves and dropping diseased leaves (Waller 1982; Merle et al. 2019). While this decreases the relative infection level within the plant, it also increases the leaf surface area vulnerable to subsequent infection (Merle et al. 2019). Future studies should examine the relative importance of infection from leaf litter, as compared to other inoculum sources, considering dynamically changing inoculum stocks in leaf litter, infected and vulnerable plant foliage, and windborne dispersal over the growing season.

The objective of our study was to establish that CLR can be spread from infected leaf litter; however, our experimental conditions must be considered when applying these results. We used young plants to ensure no previous infection, which meant that most leaves were closer to the ground than in the field, where the lowest branches are typically higher than 30 cm. We also simulated a high rain intensity in our experiment, while successful dispersal from litter may be reduced under lower rain intensities. Although these factors could have increased exposure to rain splash dispersal, note that Paul et al. (2004) found that natural rain splash dispersal of Gibberella zeae spores (the agent of Fusarium head blight of wheat) from ground-level plant debris did not decrease substantially up to heights of 100 cm. Our experiment also maintained a stock of infected leaf litter and ensured that the sporulating undersides of leaves were always facing up. In the field, we expect the orientation and exposure of infected leaf litter to vary, and that leaf stock will differ by bush size, infection intensity, and season. Depending on environmental conditions, spores lose viability in the field within a few days to a week (Waller 1982), so this must also be taken into consideration.

Other environmental and management conditions may also mediate the effect of splash dispersal from leaf litter. Rainfall intensity can influence the distance of spore spread (Madden et al. 1996) and its effect should be measured in field conditions. Further, the kinetic energy of rainfall, a key determinant of spread, may potentially be influenced by the leaf characteristics of the intercepting shade canopy (Avelino et al. 2020), so its effect on leaf litter dispersal should be investigated in the context of the other effects of shade trees on CLR (Avelino et al. 2022). Future research should also clarify the relative importance of leaf litter to other inoculum sources to optimize management needs for most effectively controlling spread.

Although resistant varieties and appropriate fungicide application are important measures for mitigating infection spread (Avelino et al. 2015), our results show that practices that reduce rain splash dispersal from leaf litter could play a complementary role. Such measures could potentially help control other coffee leaf diseases, e.g., American leaf spot (Granados-Montero et al. 2020). However, future work should investigate the costs and benefits of possible alternatives and identify potential negative outcomes. The removal of infected leaf litter is not likely to be beneficial, because not only does this practice remove organic matter and requires intensive labor, but it also could promote further CLR spread by increasing worker contact with infected bushes during high rust periods (Becker and Kranz 1977; Waller 1982). Mulching can enhance soil properties in coffee (Nzeyimana et al. 2017) and could be used to cover infected leaf litter, but this practice may still disturb infected plants and should be considered carefully.

A potential practice that could be investigated further is the use of ground cover vegetation or intercrops. In other crop systems, this has been shown to decrease splash dispersal of spores (Ntahimpera et al. 1998) and suppress fungal disease incidence (Ristaino et al. 1997). In coffee agroecosystems, such practices can promote bee abundance (Fisher et al. 2017), support natural enemy insects (Rosado et al. 2021), and provide nitrogen fixation (Mendonça et al. 2017), depending on the vegetation or crop. Through increasing the diversity and structure of the ground cover, these measures could both reduce splash dispersal and create more diversified coffee agroecosystems that could provide broader benefits for conservation and farmer livelihoods (Iverson et al. 2019). Such cultural practices could be important for growers with limited resources, and would improve resilience against large-scale events such as climate and economic changes that have historically reduced regional capacities to use high-intensity inputs to respond to CLR (McCook and Vandermeer 2015).