The first phase of the experiment (original study) [11] was carried out at the Dairy Cattle Research Laboratory of the University of São Paulo (LPBL/USP, Pirassununga, Brazil), and the second phase (current retrospective study) was performed at the Dairy Research Unit and Department of Animal Sciences of University of Florida (DRU/UF, Gainesville, FL, USA).

Soybean (Glycine max cultivar Agroeste^® 3610 I PRO, Xanxerê, Brazil) was seeded in an experimental field localized 21˚58’19.8" south and 47˚28’15.9" west, at 620 m of altitude, subtropical humid (Cwa), and 1200 mm of mean annual rainfall. Soybean plants were harvested at three different phenological stages (R5: beginning seed, R6: full seed, and R7: beginning maturity [12] at 90, 113, and 124 days after seeded respectively). For each stage, 30 kg of forage material was harvested from each of five different plots previously determined in the field, hand-mixed to get a composite sample of 150 kg of fresh forage for each stage and was chopped (Cremasco^® ECT-4000) targeting 20 mm of the theoretical length of cut. Fresh forage samples of each plot and phenological stage were collected for nutrient characterization.

A dose of 40 g/kg fresh forage of molasses was top-dressed on 2400 g of the composite sample and hand-mixed before placing into experimental silos, while the control group was not treated with molasses. The experimental silos (10 silos/treatment - 60 in total; silos were not used as experimental units in the current retrospective study) were made with polyvinyl chloride pipe with 0.5 m height and 0.1 m i.d. to achieve 650 kg/m³ fresh forage. All silos were sealed with plastic canvas (sailcloth 200 µm; Superlona black/white, Electro Plastic, Varginha, Brazil), taped within 2 h after harvest, and stored for 90 d.

For chemical profile characterization, fresh forage samples were individually dried at 55˚C for 72 h in a forced-air oven and ground to pass through a 1-mm sieve. The 1-mm ground samples were analyzed for absolute DM (method 950.15; correction of DM at 105˚C) [13], CP (N × 6.25; Kjeldahl method 984.13) [14], ether extract (EE; method 920.39) [14], NDF, acid detergent fiber (ADF), and lignin (sulfuric acid method) concentrations by sequential analysis [15] using a heat stable α-amylase [16] and sodium sulfite (TE-149 fiber analyzer, Tecnal Laboratory Equipment Inc., Piracicaba, Brazil). Non-fiber carbohydrate (NFC) concentration was calculated as NFC = 1000 − (NDF + CP + EE + ash) [17].

Chemical composition of whole-plant soybean fresh forage (average weight g/kg DM) is in Table 1. As expected, concentration of DM increased from 253 to 353 g/kg DM, as phenological stage progressed. Organic matter concentration ranged from 925 to 937 g/kg DM, as was very similar among phenological stages. Concentration of NDF and ADF decreased from 610 to 541 and 462 to 415 g/kg DM, respectively, as phenological stage progressed from R5 to R7. In addition, CP and EE concentrations increased with phenological stage.

At opening, the top 10 cm of WSS was discarded from each experimental silo and the remaining sample was homogenized for further analysis. For fermentation profile characterization, 15 g from each experimental silo was mixed with 250 mL of distilled water for 30 s and the silage juice was evaluated for pH and buffering capacity [18] using a pH meter (MB-10, Marte Cientifica^®, Santa Rita do Sapucaí, Brazil). Silage juice was centrifuged (2000 × g for 15 min.), and the supernatant aliquot (0.9 mL) was mixed (1:1) with 450 µL sulfuric acid (1 N), for ammonia nitrogen (NH₃-N) concentration analysis by colorimetric phenol-hypochlorite method [19]. For lactic acid concentration, a supernatant aliquot (1 mL) was frozen until further analysis [20]. For the concentration of organic acids, a supernatant aliquot (1.6 mL) of silage juice was mixed (4:1) with 0.4 mL ortho-fosforic acid (20%) and frozen until analysis. Briefly, organic acids were determined using a gas chromatograph equipment (GC-2014 Plus, Shimadzu©, Kyoto, Japan) equipped with automatic injector (AOC-20i at 300˚C), capillary column (Stabilwax-DA™: 30 m length, and 0.25 mm i.d.; Restek^©,

¹R5: beginning seed; R6: full seed; R7: beginning maturity; *not detected.

Bellefonte, USA) and flame ionization detector using helium as the carrier gas with a linear speed of 42 cm³/s. The temperature ramp of the column started with a gradient from 40˚C to 120˚C, at a rate of 40˚C/min, followed by ranges of 120˚C to 180˚C, at 10˚C/min, and from 180˚C to 240˚C, at 120˚C/min, holding temperature at 240˚C for 3 min at the end of analysis. Data were analyzed using the GCsolution software (v.2.42.00; Shimadzu Corporation, Kyoto, Japan).

Remaining samples from each experimental silo were dried at 55˚C for 72 h in a forced-air oven and ground to pass through a 2-mm sieve and were composited per treatment (6 samples in total, 1 for each treatment used in the present trial) for the DM, CP, and NDF in situ degradability assays. Composited 2-mm ground samples were ground to pass through a 1-mm sieve and analyzed for DM (method 950.15) [13], CP [N × 6.25; total N was analyzed by the Dumas dry combustion method (method 968.06) [13] using a CHNS analyzer (Vario Micro Cube; Elementar, Hanau, Germany)], and NDF (method 2002.04) [13] in an ANKOM 200 Fiber Analyzer (Ankom Technologies, Macedon, NY).

The addition of molasses slightly altered chemical concentrations of DM, NDF, CP, and lignin throughout the late compared to early phenological stages, whereas greatly changes were observed for fermentation parameters regardless of the phenological stage (Table 2).

Ruminal in situ procedures were conducted at the University of Florida (Gainesville, FL) under a protocol approved by the University of Florida, Institute of

¹R5: beginning seed. ²R6: full seed. ³R7: beginning maturity. ⁴Control: no additive. ⁵Molasses: 40g/kg fresh forage. ⁶Dry matter. ⁷Neutral detergent fiber. ⁸Crude protein. ⁹ meq. of alkali required of changing the pH from 4 to 6 per 100 g of dry matter [18].

Food and Agricultural Sciences, and Animal Care Research Committee (protocol #201709849). A completely randomized design with a 3 × 2 factorial arrangement of treatments was used considering phenological stage (R5, R6, and R7) and the addition of molasses (0 and 40 g/kg fresh forage) as main effects and rumen environment (cow) as experimental units. Ruminal in situ measurements were performed using three rumen-cannulated, mid-lactation, multiparous Holstein cows fed ad libitum a diet containing (DM basis) corn silage (382 g/kg), alfalfa hay (40 g/kg), dry ground shelled corn (273 g/kg), soybean meal (145 g/kg), citrus pulp (91 g/kg), minerals and supplements (68 g/kg). Cows were consuming this diet for at least 21 d prior to the experiment.

Dacron polyester cloth bags (R1020, 10 × 20 cm and 50 ± 10 μm pores; Ankom Technology, Macedon, NY, USA) containing 5 g of dried and 2-mm ground composited WSS samples (6 samples in total, 1 for each treatment), yielding a ratio of sample mass to bag area of 16.6 mg/cm² were incubated in duplicate within each cow. A total of 96 bags were incubated within each cow (6 treatments in duplicated for each of the 8 time-points). Average of duplicated bags of each treatment generated the kinetic parameter estimates on each rumen environment (cow).

Bags were placed in a nylon laundry bag (30 × 40 cm) and incubated in the ventral rumen of each cow in reverse chronological order for 240, 120, 72, 48, 24, 12, 6, and 3 h. Additionally, two empty bags for each time point (16 bags in total/cow) were incubated to correct bag infiltration or losses. After the incubation period, bags were removed from the rumen, soaked in cold water to stop fermentation, and rinsed in a washing machine using the mode rinse and spin cycle, set with room temperature water for a 30-min cycle (Roper RTW4516F*, Whirlpool Corp., Benton Harbor, MI). Each treatment had two bags prepared, soaked, and washed together with the rest of in situ bags (0 h bags). After washing, bags were dried in a forced-air oven at 60˚C for 48 h, and the difference between the initial and final DM amount was considered as DM disappearance. Rumen incubated residues from duplicates within each treatment, timepoint and cow were composited, ground to pass through a 1-mm Wiley mill screen, and analyzed for CP (0, 3, 6, 12, 24, 48, and 72 h) and NDF (0, 3, 6, 12, 24, 48, 72, 120, and 240 h) concentrations as previously described (method 968.06 [16] and method 2002.04 [16], respectively). The NDF residue at 240 h was considered as uNDF concentration (DM basis).

The degradation of DM, CP and, NDF were calculated without correction for microbial protein contamination and calculated by the exponential model according to the Equation (1) [21], and the parameters estimated by PROC NLIN of SAS (version 9.3, SAS Institute Inc., Cary, NC, USA) as follows:

D t = A + B × ( 1 − exp − k d × t ) , (1)

where D t is degradation (% DM) at time t; A is an intercept representing the proportion of soluble fraction (% DM); B is the insoluble but potentially rumen degradable fraction (% DM); k d is the fractional rate of soluble degradation of the fraction B (h); t is the incubation time (h).

The effective ruminal degradability (ERD) of DM, CP, and NDF were calculated according to the Equation (2) [21],

E R D = A + B × ( k d k d + k p ) (2)

where A, B, and k d are the same parameters as described earlier, and k p is a passage rate set to 3 %/h [21].

Data on fractions A, B, and C, k d , ERD, D t , and uNDF concentration were analyzed as a completely randomized block design with the three cows designated as blocks. Average of duplicated bags of each treatment that generated the kinetic parameter estimates on each block (cow) were the experimental units. Analysis was performed by a mixed model procedure (PROC MIXED in SAS; SAS Institute Inc., Cary, NC, United States) according to the following model:

Y i j k = μ + S i + M j + c k + S M i j + e i j k ,

with c k ≈ N ( 0 , σ c k 2 ) and e i j k ≈ N ( 0 , σ e 2 ) ; where Y i j k is the value of the dependent variable, µ is the overall mean, S i is the fixed effect of the phenological stage (i = 1 to 3), M j is the fixed effect of molasses (j = 1 and 2), c k is the random effect of cow (k = 1 to 3), S M i j is the fixed interaction effect, and e i j k is the residual error. N stands Gaussian distribution, σ e 2 is the residual variance, and σ c k 2 is the variance associated with the random effect of cow. Degrees of freedom were adjusted based on Kenward-Roger’s method, an approach specifically proposed for small sample setting as a precaution for inflated error variance [22]. Significance was declared at P ≤ 0.05.

Molasses did not affect (P ≥ 0.15) fraction B or the of DM, but both were greater (P = 0.001) at R7 compared to earlier phenological stages (Table 3). A molasses by phenological stage interaction was detected (P ≤ 0.03) on fractions A and C, and ERD of DM. Addition of molasses increased (P ≤ 0.05) fraction A regardless of phenological stage, but differences among phenological stages were only observed when WSS was not treated with molasses. Molasses addition increased (P ≤ 0.05) ERD while decreasing (P ≤ 0.05) fraction C of DM at R5 and R6, but not R7 phenological stage.

Molasses enhanced (P ≤ 0.05) the ERD of CP at R5 and R6, but not at R7 (Table 4). The addition of molasses increased (P = 0.01) fraction A and decreased (P =

¹A: r ^a-cMeans in the same row with different superscripts differ (P < 0.05). uminal rapidly degradable fraction; B: ruminal slowly degradable fraction; C: ruminal indigestible fraction; kd: degradation rate constant of fraction B; ERD: effective ruminal degradability at ruminal passage rate of 0.03 h⁻¹ by Ørskov and McDonald [21]. ²Control: no additive; ³Molasses: 40 g/kg as-fed forage. ⁴R5: beginning seed; ⁵R6: full seed; ⁶R7: beginning maturity. ⁷MOL: molasses, PS: phenological stage, and INT: molasses and phenological stage interaction effects.

^a-bMeans in the same row with different superscripts differ (P < 0.05). ¹A: ruminal rapidly degradable fraction; B: ruminal slowly degradable fraction; C: ruminal indigestible fraction; kd: degradation rate constant of fraction B; ERD: effective ruminal degradability at ruminal passage rate of 0.03 h⁻¹ by Ørskov and McDonald [21]. ²Control: no additive; ³Molasses: 40 g/kg of as-fed forage. ⁴R5: beginning seed; ⁵R6: full seed; ⁶R7: beginning maturity. ⁷MOL: molasses, PS: phenological stage, and INT: molasses and phenological stage interaction effects.

0.01) fraction B, with no differences (P ≥ 0.43) on fraction C or the of CP. Later phenological stages decreased (P = 0.001) fraction A and increased (P ≤ 0.01) fraction B and k_d, with no effect (P = 0.18) on fraction C of CP. There were no interaction effects (P ≥ 0.11) on fractions A, B, C, or k_d of CP.

The addition of molasses decreased (P = 0.01) fraction C, increased (P = 0.001) ERD, and had no effect (P = 0.51) on the k_d of NDF (Table 5). Harvesting WSS at a later phenological stage decreased (P = 0.001) fraction C and k_d, increasing (P = 0.001) the ERD of NDF. Moreover, an interaction between the addition of molasses and phenological stage was detected (P ≤ 0.04) on fractions A and B of NDF, and uNDF. Fractions A and B of NDF were greater (P ≤ 0.05) at R7 and

^a-dMeans in the same row with different superscripts differ (P < 0.05). ¹A: ruminal rapidly degradable fraction; B: ruminal slowly degradable fraction; C: ruminal indigestible fraction; kd: degradation rate constant of fraction B; ERD: effective ruminal degradability at ruminal passage rate of 0.03 h⁻¹; undegraded NDF at 240 h of ruminal incubation (g/kg DM) by Ørskov and McDonald [21]. ²Control: no additive; ³Molasses: 40 g/kg of as-fed forage. ⁴R5: beginning seed; ⁵R6: full seed; ⁶R7: beginning maturity. ⁷MOL: molasses, PS: phenological stage, and INT: molasses and phenological stage interaction effects.

R5, respectively, while uNDF was decreased (P ≤ 0.05) at R5 and R6 when silages were treated with molasses.

Degraded DM, NDF, and CP concentrations in WSS at each incubation timepoint are presented in Figure 1.