Sampling and analysis were similar to that described by Rodríguez et al. (2013). Each other windrow was selected to sample, thus, only five windrows were sampled at three points: two next to the edges and one in the middle. Approximately 1 kg sample was taken at 20 cm depth. The three samples from the same windrow were mixed and labeled. Sampling was approximately every 6-8 days. Every time a sample was taken, outer and inner temperature was measured: outer temperature was measured by placing a mercury thermometer on the windrow and the inner temperature by using a thermocouple probe omega HH100 1 m inside the pile at the sampling point.

Sometimes, mixing or sampling was suspended because occasional rains, so the time interval for sampling was not homogeneous. Time zero was considered after the windrows were entirely deposited (so the first pile of that windrow could have been deposited 5 days or less before time zero).

Organic matter and humidity were determined according with NOM-021-RECNAT-2000 (SEMARNAT 2000). Samples were dried at 40 ºC to constant weight before analysis. Every analysis was carried out six times and only the mean is reported.

Compost DNA was obtained by following procedure of MPBio FastDNA Spin kit for Soil ® on compost samples when no significant change in organic matter content was observed for two weeks, when maturity is considered to be reached. Metagenomic analysis was performed by NGS service at CIAD-Mazatlan (www.ciadmazatlan.com) service by using Illumina ® NGS miniSeq The methodology used was an analysis of the microbiota by massive sequencing of the 16S ribosomal gene with the primers V3-338f and V3-533r with Illumina adapters to amplify the V3 region only. Finally, the samples were sequenced on the Miniseq device of Illumina in standard conditions (300 cycles, 2X150). The 16S ribosomal RNA sequences obtained were verified for the quality, chimeras were eliminated, after pairing and singleton elimination, paired sequences were searched against SILVA high quality ribosomal RNA databases http://www.arb-silva.de. Subsequently, species with less than 1 % were grouped as “others” to have a clear view of the dominant bacteria species.

Culturable bacteria were isolated by plating 50 μL of a compost suspension (1 g/10 mL of sterile saline solution) on soy trypticasein agar. Twenty colonies were randomly taken and cultured in Luria-Bertani broth overnight previous to measure plant-growth promoting and C23O enzyme activities. Species from sixteen isolates were identified by using MALDI-TOF of Bruker ®.

Phosphate solubilization was measured after placing 100 μL of overnight culture in 8 mL of phosphorite Pikovskaya broth (5.0 g/L phosphorite, 4 mg/L manganese sulfate, 2 mg/L ferrous sulfate, 0.2 g/L sodium chloride, 0.2 g/L potassium chloride, 0.5 g/L yeast extract, 0.3 g/L magnesium sulfate, 0.5 g/L ammonium sulfate, 10 g/L glucose). After 48 h incubation at 28 ºC, phosphate concentration in supernatant was measured by using Hanna kits. No inoculum was used as a control. Analysis were made by triplicate.

Indolacetic acid (IAA) production was evaluated by Salkowski reaction. Tubes containing 800 μL of tryptone culture medium (5 g/L glucose, 1 g/L dibasic potassium phosphate, 0.4 g/L ammonium nitrate, 0.2 g/L sodium chloride, 0.4 g/L magnesium sulfate, 20 g/L tryptone) with or without 0.1 g/L tryptophan (intermediate for IAA production) were inoculated with 10 μL of overnight culture. After 48 h incubation at 28 ºC, cell pellets were resuspended with Salkowsky reagent (1 mL 0.5 M ferric chloride, 50 mL water and 30 mL concentrated sulfuric acid). Tubes were incubated at room temperature for 30 min. Absorbance was measured at 530 nm. Concentration was calculated by interpolation within a calibration curve by IAA. Analyses were made by triplicate.

Siderophore production was determined by chromeazurol S reaction with ferric ion in agar plates. Siderophore reaction agar was prepared by adding 10 mL of solution 1 (0.1 mM ferric chloride, 0.6 mg/mL chromeazurol S, 0.728 mg/mL hexadecyltrimethylammonium bromide) to a mixture of 800 mL of solution 2 (37.8 g/L PIPES, 0.375 g/L monobasic potassium phosphate, 0.625 g/L sodium chloride, 1.25 g/L ammonium chloride, 2 % agar pH 6.8) with 70 mL of solution 3 (2.9 % glucose, 2.9 % mannitol, 0.7 % magnesium sulfate, 0.016 % calcium chloride, 0.0017 % manganese sulfate, 0.002 % boric acid, 0.00006 % cupric sulfate, 0.0017 % zinc sulfate, 0.0014 % sodium molybdate). After autoclaving, siderophore reaction agar was poured on Petri dishes: A color change to orange (from blue) indicates siderophore production. The diameter of the halo is an indirect measurement of the amount of produced siderophores. Analyses were made by triplicate.

For C23O enzyme activity measurement, overnight cultures were resuspended in Davis minimum medium and incubated 24 h at 28 ºC. Bacterial pellets were resuspended in 1 mL of 40 mM phosphate buffer (pH 7.5). Enzyme activity was measured spectrophotometrically by adding 50 μL of cell suspension on 1 mL of 40 mM phosphate buffer containing 2.5 mM catechol. Molar absorptivity of the product 36 000/M/cm was used (Hupert-Kocurek et al. 2012). Activity was reported by μmoles of products/min/mg protein. Total protein was determined by Lowry method (Ramírez-Sandoval and Velázquez-Fernández 2013). Analyses were made by triplicate.

Models and descriptive statistical analysis were carried out with Matlab R2017a.

Although the three thermal phases are to be expected during the composting: rising, thermophilic and lowering (EPA 1998), several works have reported no significant changes at laboratory or pilot scales. For instance, at laboratory scale, Arango-Osorno et al. (2016) did not observe changes during the entire process, i.e., the temperature was in the mesophilic range during the entire composting process (20-30 ºC). Contrastingly, at pilot scale, Rodríguez et al. (2013) found a higher temperature (45-70 ºC), but with no significant changes during the whole process. Similarly, in the present work, temperature ranged from 40 to 70 ºC, exception made of site 2 at 94 days, which was lower since it had been turned and watered just before sampling (Fig. 2). Thus, for sampling, no other windrow was sampled if it had been turned and watered in the same day or the previous to the sampling. This also suggests that the turning and water addition lower the compost temperature by 20 ºC. No rising phase could be detected, so, the temperature increases within the first five days, i.e., the time that takes to complete the pile.

Fig. 2

The differences between the temperature ranges at laboratory scale and pilot or field scales could be explained in terms of the windrow/pile depth and surface. At laboratory scale, compost piles have generally no more than 0.5 m depth, while at pilot scale they are 1 m or even deeper. The depth at laboratory scale allows better pile aeration and heat exchange, since most laboratory compost containers have no more than 20 cm distance between the core and the surface. At field scales, we measured external temperature by placing a mercury thermometer on the surface of the pile, and when it was introduced 20 cm within the pile, the difference was negligible or less than 2 ºC. This finding and the behavior of humidity and organic matter suggest that pilot scale is closer to field scale composting than laboratory scale, regardless of the feedstock. As can be observed in figure 3, external temperature ranged from 20 to 35 ºC, which is similar to those values observed at laboratory scale (Arango-Osorno et al. 2016).

Fig. 3

Inner and external temperatures had a low determination coefficient vs. time: R² = 0.50 and 0.52, respectively. Positive slope values reflect a temperature increase rate of 0.59 ºC/week and 0.24 ºC/week (Figs. 2 and 3). Therefore, temperature variations are larger than any observed correlation or rate, in other words, there is no significant correlation between time and inner or external temperature (or their difference). As the depth could affect inner temperature and aeration, it is logic to assume that it would also affect biodegradation of organic matter.

Regardless of the scale, compost organic matter levels, at the end of composting process are expected to be 30-50 %. Organic matter decay and maturation time was similar (Table I) among the five studied piles and with the values reported for pilot scale (Fig. 4; Rodríguez et al. 2013). Three mathematical equations were evaluated to model organic matter decay, as it has been described previously by Roncevic et al. (2005): one linear and two exponentials, corresponding to the following:

Fig. 4

M = M 0 - k t     (1)

M = M 0 e - k t (2)

M = M 0 e - k t (3)

Where M represents organic matter content in percent (dry basis) at time t; M ₀ , calculated initial organic matter (or at time 0) in percent; t, time in days and k, decay or degradation rate constant according to the respective model, in 1 / days only for linear and first exponential model (equations 1 and 2).

For comparison, averages or all the values (Fig. 4A) were used to evaluate the correlation (Table II). Average values were also used for contrast (Fig. 4B). They were calculated considering the corresponding time zero for each site regardless the site, in other words, values from different sites were averaged when they were within a time window of two days.

TABLE II

Model	R2	M0	ka
Using all data (n=28)
	-0.665	72.1	-0.348
	-0.593	70.5	-0.006
	-0.629	70.5	-0.083
Average values (n=10)
	-0.834	73.3	-0.366
	-0.822	74.3	-0.007
	-0.833	91.7	-0.086

For first and second models, units of k are 1 /days

Correlation coefficients are close or higher than 0.6, regardless of the averaging or the model. Considering that they arise from field-scale measurements having several variables that were not controlled, they reflect proper models. The use of average values led to higher correlation coefficient values (Table II). Although averaging reduces the number of time points for modeling, (n = 10), the n is still above 7, below which the n could have an important effect on the coefficient correlation. The values of correlation coefficient from the three models are very similar. Regardless of the n or model used, M ₀ value is 70-74 %, exception made of the third model described by Roncevic et al. (2005) for hydrocarbon degradation. Using that model, M ₀ is higher than the rest of all the models. Although the M ₀ value was calculated for the models, the values are very close to the real ones, which were 72-84 %. Thus, the third model using seems not to be adequate or advantageous. Moreover, R² and M ₀ values from the linear and exponential models are similar, suggesting that either model could be used. This can be observed graphically: when plotting the calculated values of organic matter from the exponential model within the time window of the present work, they look quite linear (Fig. 4B).

The linear model is easier and more pragmatic to use in field, thus, quick calculations can be done. Also, the kinetic constant k is very similar regardless the number of data points used, raw or average values, suggesting that even the data variability from this field-scale study does not impact on the kinetic parameters. Taking all this into account, it is possible to model the degradation kinetics of the organic matter decay at field-scale using a linear model, which will ease on-field calculations. Parameter k values for each site (Table III) range from -0.31 to -0.53 %/day, i.e., 2.2-3.7 % of organic matter is degraded per week. As far as we know, no other kinetic constants of organic matter decay at field-scale have been reported. Knowing the kinetic model and constants or organic matter decay will help prospect or predict compost degradation at the commercial level and ease the (on-field) calculations, which in its turn, will reduce the effort and time the analysis needs to be performed on the field. The fact that the data adjusts to a linear model seems to be due to a pseudo-zero order. Although seems independent of the organic carbon content, that could be caused by the slow process and high variability of the parameters involved such as temperature or other metabolic needs for bacterial biodegradation.

TABLE III

Bacterial genus	Phosphate (mg/L)	Indolacetic acid production (μg/mL)		Siderophore production	C23DO activity
		with Trp	without Trp
Bacillus	6.8	-	-	-	0
Bacillus	5.9	52.7	-	-	10.2
Bacillus	4.8	8.8	3.0	-	0
Bacillus	7.3	-	-	-	0
Bacillus	5.8	-	-	-	0
Bacillus	6.6	-	-	-	0
Bacillus	5.4	-	-	-	0
Bacillus	7.6	-	16.8	-	1.3
Lysinibacillus	6.0	12.3	4.9	-	0
Bacillus	6.7	-	-	-	0
Bacillus	6.6	-	19.8	-	0
Bacillus	8.7	-	25.9	-	18.5
Bacillus	13.1	-	29.5	-	0
Bacillus	7.6	-	6.8	-	0
Bacillus	11.0	-	-	-	0
Lysinibacillus	8.6	-	28.8	-	0

Trp = trypophan. Activity of C23DO is expressed in μmole/min/mg protein

The most abundant bacterial classes are Bacilli (5.6 %) and Anaerolineae (9.2 %, Fig. 5). Bacilli group contains heterotrophic bacteria. Recently, a species of Bacillus has been found to be termophillic and encode for a mannannase, a hemicellulase (Wang et al. 2018). Also, Anaerolineae, a class of Chloroflexi bacteria has been reported to encode for cellulase and precursors for biofilm (Xia et al. 2016). Thus, the presence of these classes could be explained by the presence of cellulose or hemicellulose present in the initial compost substrate. Moreover, as Xia et al. (2016) have pointed out, the Anaerolineae capability to form biolayers could be helpful to allow other bacteria or consortia to join together on matrices or particles to enhance biodegradation or allowing other bacteria to grow on the substrate surface. Despite that, Xia et al. (2016) observed that Anaerolineae relative abundance could decrease with the time. Contrastingly, Anaerolineae was the most abundant class in compost. Contrastingly, Anaerolineae was the most abundant class in compost. This suggests that the final compost itself could be used as an inoculum for subsequent bagasse composting. Further studies on this sense, beyond the aim of the present work, are required.

Fig. 5

Although obtaining an inoculum with the whole consortia would be the most adequate for downstream applications, when using conventional methods, only Bacillus and Lysinibacillus were isolated (Table III). Since some species from Bacillus genus have been described to be resistant to temperature, is not rare to observe them at the end of composting. Bacillus has 3.2 % relative abundance in the compost according to metagenomic analysis and Lysinibacillus were 0.37 % abundant. Keeping that in mind, this suggests that Bacillus are 8.6 times more abundant than Lysinibacillus. Indeed, when isolated in conventional culture media, Bacillus were 14 of the 16 species isolated, whereas Lysinibacillus are the other two. All of the isolates solubilize phosphate, and none of them produced siderophores. In the presence of tryptophan, eight, i.e. 50 % of them were able to produce the phytohormone, indoleacetic acid, whereas, only three isolates would produce the latter in the absence of the aminoacid precursor. Also, three were able to show C23O activity, an enzyme part of the pathway to degrade aromatic compounds, including lignin derivatives. Taken all this together, and since the metagenomic and capacity for promoting plant growth analysis were carried out with the final compost, it suggests that the microbiota would make the compost able to be used as a fertilizer. Moreover, microbiota from Bacilli class would still be able to degrade lignocellulosic residues as described above.