Coffee is one of the most important crops around the world because it is crucial for the economy of at least 60 countries and the major source of income for 100 million people (^{Talhinhas et al., 2017}). In Mexico, coffee is a strategic crop since its production employs more than 500,000 producers from 14 states and 480 municipalities (^{SAGARPA, 2017}). The major coffee cherry producing states are Chiapas (357 733 t), Veracruz (192 341 t) and Puebla (112 228 t). In Puebla, coffee production is valued at 778 526 thousand pesos (^{SIAP, 2018}). However, the coffee crop is affected by rust, a disease caused by Hemileia vastatrix. The disease symptoms appear on coffee leaves, initially in the form of chlorotic spots that become necrotic as the infection advances. The effect varies from slight defoliation to up 50% yield loss (^{Avelino et al., 2015}, ^{Zambolim, 2016}; Talhinhas et al., 2017). Preventive and systemic fungicides containing copper, epoxiconazole or pyraclostrobin are applied in the diverse coffee producing regions to control the pathogen (Zambolim, 2016; Talhinhas et al., 2017). In Mexico, SENASICA ran a control program based on a successful Surveillance System that contained the 2012-2013 regional epidemic outbreak (^{PVEF-CAFETO, 2018}). In the Sierra Norte de Puebla, coffee is traditionally grown with minimum use of inputs, which means that production is under organic management. However, this has contributed to the disease occurring year after year, causing variable production losses. The varieties most commonly cultivated are Typica, Garnica and Mundo Novo (^{Benitez-García et al., 2015}). Diverse studies have demonstrated the potential use of biofungicides. These products are made from plants and microorganisms fungi control. In Mexico, Acremonium byssoides, Calcarisporium arbuscula, C. ovalisporum, Sporothrix guttuliformis, Fusarium pallidoroserum, Verticillum lecanii, Simplicillium sp. and Lecanicillium sp. have been identified as having the potential to control H. vastatrix (^{Carrión and Rico-Gray, 2002}; ^{Gómez-De La Cruz et al., 2018}). Other studies conducted under greenhouse and laboratory conditions have indicated the potential of biofungicides based on Cinnamomum zeylanicum, Cymbopogon nardus, Cymbopogon citratus, Corymbia citriodora, Melaleuca alternifolia, Thymus vulgaris, Azadirachta indica, Syzygium aromaticum and Allium sativum to control H. vastatrix. This kind of products offer a management alternative for the production of organic coffee, which is valued worldwide (^{Borges et al., 2012}; ^{Silva et al., 2014}). The use of biofungicides to control rust is a low cost and low environmental impact option that reduces the risk for H. vastatrix to develop resistance compared to the use of synthetic fungicides (Avelino et al., 2015; ^{Ibañez and Blackman, 2016}; Talhinhas et al., 2017). However, it is necessary to analyze the effect that formulations based on plants or microorganisms have on disease progress in the field in order to come up with recommendations that can be incorporated into integrated management practices. For this reason, the objective of the present study was to evaluate the effectiveness of biofungicides based on Azadirachta indica (neem), Melaleuca alternifolia (tea tree) and the combination of Bacillus subtilis with Azadirachta indica and Syzygium aromaticum (clove) to control rust caused by H. vastatrix on two coffee tree varieties.

In 2017, experiments were established at three sites in the Huehuetla municipality, Puebla: a) Lipuntahuaca (20° 04’ N and 97° 37’ W; 546 masl); b) Chilocoyo (20° 04’ N and 97° 39’ W; 907 masl); and c) Cinco de Mayo (20° 07’ N and 97° 37’ W; 608 masl). In this municipality, annual precipitation is 2900-3600 mm; the precipitation in the driest month is higher than 40 mm, which is a semi-warm and humid climate where the temperature varies from 18 to 24 °C (^{INEGI, 2009}). The effect of three biofungicides on coffee plants of the Garnica and Typica varieties was evaluated in each experiment (Table 1).

The experiments were established in the field in sites surrounded by coffee tree fields to which no fungicides were applied and where the disease occurs year after year. Five-month old plants grown in seedbeds with no fungicide treatment were taken from the nursery for the experiment. Infection was natural. The experiments in each location were established using a treatment design of subdivided plots with three replications (^{Villaseñor-Mir et al., 2012}). Locations were established in the large plot (3), treatments with and without biofungicide in the intermediate plot (4), and Garnica and Typica (2) in the small plots (Villaseñor-Mir et al., 2012; ^{Santa-Rosa et al., 2016}). Each experiment unit consisted of three plants. Each location had nine plants of each variety per treatment, for a total of 18 plants of the two varieties in each treatment. Disease progress was monitored on all the leaves of each of the 72 plants per location.

The infection caused by H. vastatrix occurred naturally. The first signs and symptoms were observed by inspecting the leaves with a 40 X magnifying glass and under a stereoscopic and compound microscope (^{SENASICA, 2019}). The disease progress was monitored by conducting seven evaluations and estimating the following variables (^{Campbell and Neher, 1994}): 1) Incidence, calculated as the percent of infected leaves compared to the total of leaves per plant; 2) severity in leaves, calculated as the percent of damaged foliar area on each leaf using a diagrammatic scale (^{SINAVEF, 2013}); 3) total severity, evaluated based on the average severity observed on each leaf of the plant, on each evaluation date.

The disease was evaluated two days after the first application (dpa), at eight dpa, 16 dpa, 29 dpa, 36 dpa, 43 dpa and at 51 dpa. Based on the total severity values, the area under the disease progress curve was estimated (AUDPC) (^{Simko and Piepho, 2012}) by the following equation:

ABCPE=∑i=1n-1Yi+Yi+12*ti+1-tI

Where:

Yi= disease intensity

t= evaluation period (time)

n= number of evaluations

The incidence and severity data were subjected to a validation of assumption of normality using Shapiro-Wilk’s test and homogeneity of variance using Levene’s test (^{Zar, 1999}). Since the tests indicated that the data were not normal, the incidence and severity percents were converted to the arcsine square root of the percent value. The transformed data were subjected to an analysis of variance using ANOVA with the SAS statistical program version 9.4 (Statistical Analysis System, Cary, North Carolina) to identify the differences between locations and treatments, as well as the interactions. In addition, analyses of variance were conducted for each evaluation date. The means were compared using Duncan’s multiple range test (p≤0.05) (^{Steel and Torrie, 1985}). The analysis included ANOVA of the area under the disease progress curve (AUDPC) based on the severity which was estimated using the transformed data, as well as the estimation of the apparent infection rate with the transformed percentages of infected foliar area using ln(x/(1-x)) and considering the time during which the disease was monitored (^{APS, 2019}).

Statistically significant differences (p≤0.01) among locations and biofungicides were observed for both rust incidence and severity. However, there were no differences in the rust incidence among the varieties, although the analysis of variance showed differences in severity (Table 2). At two dpa, the control plants and the plants treated with a mixture of B. subtilis + A. indica + S. aromaticum showed 9.5% and 8.7% incidence values, respectively, followed by the plants that were treated with A. indica (14.2%) and M. alternifolia (18.9%) (Figure 1A). This pattern was also observed at eight dpa and was more associated with natural expression of the disease on aggregation spots than to the effect of the biofungicides (^{Quiñones-Valdéz et al., 2015}). Starting at 16 dpa, the incidence levels increased in all the treatments, but no significant differences were detected. However, the apparent infection rate (r) made it possible to differentiate the level of the disease progress between 8 dpa and 16 dpa in each treatment, where the highest rate of infection was observed in plants that were sprayed with a mixture of B. subtilis + A. indica + S. aromaticum (r=0.16), followed by the control (r=0.14), by plants treated with A. indica (r=0.08) and by plants treated with M. alternifolia (r=0.04). This indicates that the biofungicide based on M. alternifolia reduced the disease progress compared to the other treatments.

Starting at 29 dpa and up to 51 dpa, the plants treated with M. alternifolia had the lowest level of rust incidence (p≤0.05 on each evaluation date) compared to the other treatments. This behavior was observed in both varieties and could have been caused by the accumulated effect of the first and second biofungicide applications. At 51 dpa, the incidence in plants treated with M. alternifolia was 34.4%, while the incidence in the controls was 43.5%; the differences in the final incidence between the control plants and the plants sprayed with B. subtilis + A. indica + S. aromaticum (44.2%) and A. indica (41.5%) were not significant (Figure 1A).

Figure 1. Progress of rust caused by Hemileia vastatrix in young coffee plants treated with biofungicides in the field, in Huehuetla, Puebla. A) Incidence, B) Total severity. Total incidence and severity were statistically compared by using the transformed values (* = p≤0.05; ns=not significant). The figure shows the non-transformed values. The arrow indicates the time when the fungicide was applied. TE=Control, AI= Azadirachta indica, MA=Melaleuca alternifolia and BA+AI+SA= Bacillus subtilis + A. indica + Syzygium aromaticum. The bars represent the standard error of the mean of each evaluation. 

These results indicate that the application of the biofungicide made of M. alternifolia reduced the infection caused by the fungus on leaves and at the plant level at the three locations. Similar results were reported by ^{Borges et al. (2012}), who observed a lower level of coffee rust incidence when they applied essential M. alternifolia oils compared to the controls and the plants sprayed with A. indica. Regarding the total severity, starting at 16 dpa and up to 51 dpa, the plants treated with M. alternifolia showed a lower level of severity with values of less than 2.5%. The plants treated with a mixture of B. subtilis + A. indica + S. aromaticum, and the plants treated with A. indica and the control plant had 3.2 and 3.8% severity. The differences in the severity values among treatments were not significant from 2 dpa and up to 43 dpa. However, in the last evaluation (51 dpa), significant differences were observed among the treatments (p≤0.05), because the control plants reached the highest severity values (3.8%), followed by the plants treated with B. subtilis + A. indica + S. aromaticum and the plants treated with A. indica that had 3.3% and 3.6% severity, respectively; these were, in turn, statistically the same (p≤0.05), while the plants treated with M. alternifolia had the lowest percent of damaged foliar area (2.5%) (Figure 1B).

These results are in agreement with the results documented by ^{Borges et al. (2012}), who reported that M. alternifolia was more effective in reducing disease progress than A. indica, while ^{Haddad et al. (2009}) reported a lower level of disease intensity when applying Bacillus sp. isolates. ^{Medice et al. (2007}) reported 35.0-62.0% reduction in Asian rust severity (Phakopsora pachyrhizi) in different varieties that were treated with essential oils of thyme (Thymus vulgaris), citronella (Cymbopogon nardus), eucalyptus (Corymbia citriodora) and neem (Azadirachta indica). In our experiments, the biofungicide made of M. alternifolia reduced H. vastatrix incidence by 20.9% compared to the control plants. M. alternifolia, A. indica y B. subtilis + A. indica + S. aromaticum reduced the disease severity by 35.1%, 5.5% and 13.7%, respectively, compared to the severity in the control plants. Borges et al. (2012) obtained similar results where the incidence and severity was reduced by 12.1% and 55.4%, respectively, using essential tea tree oils (M. alternifolia). No statistically significant differences were found in the severity of the plants that were sprayed with essential neem oil (A. indica) regarding the control plants. In these experiments, the application of M. alternifolia limited the disease progress and fungal infection but did not prevent increases over time.

The analysis of variance of AUDPC based on the severity and the apparent infection rate indicated significant differences among treatments (p≤0.05). Although the biofungicide made of M. alternifolia reduced the disease progress, the mean comparison test did not show significant differences regarding the control, which could be due to a variation in the levels of severity recorded during the first evaluations as a result of the initial infection process and the recent application of the biofungicide. The comparison of the AUDPC means indicated that the application of A. indica did not limit the disease progress and that it was even higher than that of the control (Table 3). However, the means test conducted on the apparent infection rate indicated significant differences among treatments (p≤0.05), where M. alternifolia had the lowest infection rate (r=0.028), while the control had the highest value (r=0.046). In percentage terms, the reduction in the apparent infection rate compared to that of the control was 39.1% with M. alternifolia.

^{Haddad et al. (2009}) reported that the application of Bacillus sp. was effective in controlling H. vastatrix. They also mentioned that certain isolates did not show significant differences due to unfavorable environmental conditions for this biological control agent. It is likely that the response of the isolate in the formulation of the biofungicide based on Bacillus sp. was caused by its low adaptative ability to the environmental conditions at the experiment sites. This means that formulations must be prepared with isolates obtained from each coffee producing area, since using commercial formulations made from microorganisms to control the disease can lead to lower effectiveness caused by the effect of the environmental conditions in each area.

The incidence and severity had significant differences among locations (p<0.01). The Chilocoyo location had the highest final disease incidence and severity values, while Lipuntahuaca had the lowest level of incidence and severity (Figure 2). Plants of the Garnica variety had 29.0% to 55.0% incidence, while the plants of Typica had 24.7% to 48.2% incidence. According to the analysis of variance, Typica and Garnica varieties had no significant differences in the incidence in each of the locations (Figure 2A). Regarding the average severity, the Typica variety reached values between 1.7% and 3.5%, while the Garnica variety had values between 2.9% and 4.8%. These varieties had significant differences in severity only in the Lipuntahuaca location (p≤0.01), where the Garnica variety was more susceptible to the disease (Figure 2B). According to reports from the Coffee Rust Surveillance Program, varieties Garnica and Typica have been the most severely affected compared to other varieties such as Geisha, Costa Rica and Catimor (^{PVEF-CAFETO, 2018}).

The disease was not eradicated with the use of biofungicides, but the applications of M. alternifolia reduced the infection caused by the fungus at the plant level. They also showed a tendency to reduce the disease after two applications; that is, 29 days after the first application; also although it did not prevent an increase, it was the treatment with the lowest apparent infection rate (r=0.028, p≤0.05). By using M. alternifolia, the percentage of disease reduction compared to the control, at the end of the evaluation, was 20.9% for incidence, 35.4% for severity, 14.2% for AUDPC and 39.1% for apparent infection rate. The Typica and Garnica varieties did not show significant differences in H. vastatrix incidence and severity, except in the Lipuntahuaca location, where the Garnica variety had the highest level of foliar damage. The use of M. alternifolia extracts to control rust in young coffee plants could be combined with integrated management practices as a compatible alternative aligned with the principles of sustainable organic coffee production. However, these results suggest that further research must be done, considering better inoculum control, alternative doses, reproductive-age plants and periods of higher regional epidemic intensity.

Figure 2. Behavior of coffee rust in young plants of two varieties evaluated per location in the field in Huehuetla, Puebla. A) Incidence, B) Severity. The incidence and severity were statistically compared using transformed values. The bars show non transformed values. TY=Typica, GA=Garnica. Bars with the same letter in each locality are statistically the same (Duncan, p≤0.05). The bars represent the standard error of the mean at each evaluation.