A total of 29 individual feedstuffs commonly used for dairy cattle feeding in Thailand and Vietnam consisting of fresh grasses, preserved roughages, whole crop ingredients and single concentrates were collected. Common and Latin names of the various feedstuffs are presented in Table 1. The feedstuffs were selected from various locations in Thailand and Vietnam with the assistance of recognized experts in ruminant nutrition (see acknowledgement). All feedstuffs, at the time of collection were used in practical feed formulation of dairy cattle. The samples were oven dried at 60°C until stable weight, stored in plastic bags and transported to Wageningen University & Research (Wageningen, the Netherlands) for analyses.

Upon arrival at Wageningen, feed samples were ground (1-mm screen) using a cross beater mill (Peppink 100 AN, Deventer, The Netherlands) and analysed in duplicate for residual dry matter, crude ash, crude fibre, starch and sugars. Crude protein (CP) was calculated from nitrogen (N×6.25) obtained by the Kjeldahl method [8]. The neutral detergent fibre (NDF; with heat stable α-amylase) was analysed according to Van Soest et al [9], acid detergent fibre (ADF), and acid detergent lignin (ADL) contents were determined according to Van Soest [10]. Volatile fatty acids (VFA) were analyzed using a gas chromatograph (Trace GC Ultra; Thermo Scientific, Milan, Italy) equipped with a flame ionization detector and an Agilent HP-FFAP column (Agilent Tech., Santa Clara, CA, USA; 30 m length, 0.53 mm i.d., 1 μm film) using hydrogen as carrier gas (25 kPa, constant pressure). Isocaproic acid was used as an internal standard.

Total cumulative in vitro gas and CH₄ production were measured using an automated gas recording system [11,12] with gas and CH₄ being measured over 72 h. Each of the 29 feedstuffs was ground over a 1-mm sieve using a Wiley mill (Peppink 100AN, Olst, The Netherlands). An amount (~0.5 g) of each feedstuff (Table 1) was precisely weighed into 250 mL fermentation bottles (Schott, Mainz, Germany) with each substrate weighed in triplicate bottles for each of two runs. Three bottles of blanks (rumen fluid without feedstuff) were used in each run. Rumen fluid was obtained from three rumen cannulated Holstein-Friesian, non-lactating cows before the morning feeding at the research farm of Wageningen University, the Netherlands. Cows were fed a low quality grass silage (net energy for lactation, 4.37 MJ/kg dry matter [DM]; CP, 99 g/kg DM; NDF, 675 g/kg DM) ad libitum and had free access to water. Approximately 250 mL rumen fluid was collected from the front ventral, middle ventral and cranial dorsal sac from each individual cow. The three rumen fluid samples were pooled and filtered through cheesecloth and subsequently mixed (1:2, v/v) with an anaerobic buffer/mineral solution [11] under continuous flushing with CO₂. Prior to inoculation, fermentation bottles were placed in a shaking water bath kept at 39°C and pre-flushed with CO₂. Sixty ml of buffered rumen fluid was added to the bottle before being connected to the fully automated gas recording equipment for 72 h at which time the bottles were opened and placed on ice. Six hundred μL of the solution was pipetted into a 1.5 mL Eppendorf tube and 600 μL internal standard solution (isocaproic acid) was added before mixing vigorously. After 5 min centrifuging at 14,000×g, a 0.75 mL sample of the supernatant was taken and mixed with an equal volume (1:1, v/v) of a stock solution composed of 25 mL of 85% (v/v) ortho-phosphoric acid dissolved in 200 mL Millipore water (Merck KGaA, Darmstadt, Germany) and 300 mL of a 4 g/L 4-methylvaleric acid (internal standard) for VFA analysis and stored at −20°C pending analysis.

The degraded OM (OMd) was determined as described by Williams et al [13] and calculated as the difference between incubated and residual OM after 72 h of fermentation.

Precisely 10 μL of the headspace gas was collected from each bottle and directly injected into a gas chromatograph to determine headspace CH₄ production at 0, 3, 6, 9, 12, 24, 30, 36, 48, 60, and 72 h, as described by Pellikaan et al [12,14]. Briefly, measured CH₄ production in individual bottles were expressed relative to the maximum production in each bottle and were fitted iteratively with a monophasic model. Methane production at each individual valve opening were then calculated, and cumulative CH₄ was calculated as the sum of the increase in headspace CH₄ production between two successive valve openings, and the amount of CH₄ vented from the bottle.

Gas and CH₄ production from all samples were corrected for the corresponding production by blank bottles at each time point [11,12]. Using the nonlinear least squares regression procedure of SAS Institute Inc [15], following equation [16] was fitted to the corrected cumulative data:

where, gas production (GP) (mL/g OM) is the cumulative produced gas or CH₄; n = total number of phases, i = number of phases, Ai (mL/g OM) is the estimated asymptotic gas or CH₄ production in phase i; Bi is a constant determining the switching characteristic of the curve in phase i; Ci (h) is the time at which half of the asymptotic gas or CH₄ production was reached in phase i and t (h) is the time of incubation.

A tri-phasic model (n = 3) was fitted to the cumulative GP following the procedure as described by Groot et al [16], where phases 1 and 2 are assumed to relate to the fermentation of the soluble and non-soluble fraction, respectively, and phase 3 is assumed to be related to microbial turnover. The time windows related to the asymptotes of GP for phase 1, 2, and 3 (A1, A2, and A3, respectively) were pre-set from 0 to 3, 3 to 20, and 20 to 72 h after the start of incubation of the substrate, respectively to enable the estimation of the various parameters (Bi and Ci). The aforementioned time points were determined by Van Gelder et al [17]. Data on CH₄ production were fitted according to the above-mentioned model where n = 1. The maximum rate of gas or CH₄ production (Rmax, mL/g OM/h) was calculated as described by Yang et al [18].

Total VFA in fermentation fluid at 72 h was calculated as the sum of Hac, Hpr, butyric acid (Hbu), valeric acid (Hva), isobutyric acid (iso-Hbu) and isovaleric acid (iso-Hva). Branched-chain volatile fatty acids (BCVFA) in fermentation fluid were calculated as the sum of iso-Hbu and iso-Hva. The non-glucogenic to glucogenic ratio (NGR) was calculated as follows [19]:

Prior to statistical analysis of the data related to the fresh grasses, triplicate in vitro results of each substrate were averaged per run. Effects of genus and species within each genus on model parameters were subjected to analysis of variance using the PROC MIXED procedure [14] using the following model:

where, Y_ijk = response variable (i.e. GP-72, CH4-72 production, fermentation kinetics parameters), μ is the overall mean, G_i is the effect of genus i (i = 1 to 3), S_j(i) is the effect of species j (j = 1 to 4) nested within genus i, R_k is the random effect of run k (k = 1 to 2) and e_ijk is the residual error term. Differences among species within each genus were determined using the least square means procedure and Tukey’s multiple comparisons [14]. Throughout, the level of statistical significance was pre-set at p<0.05 while a trend was declared at 0.05≤p<0.10.

Except for the OM contents, and as expected, the selected feedstuffs showed a large variation in chemical composition (Table 1). For all the selected roughages (i.e. fresh grasses, preserved roughages and whole crops), the greatest variation was found in the CP (14 to 264 g/kg DM) and sugar (0 to 327 g/kg DM) contents. The NDF and ADF contents were high in fresh grasses and preserved roughages, and except for alfalfa hay and cassava hay, NDF and ADF values were found to be ≥631 and ≥333 g/kg DM, respectively. Except for the low starch contents of corn silage, rice straw and whole crop sorghum and corn with cob, no starch was found in the other roughages (Table 1). In contrast to starch, sugars were detected in roughages and especially in fresh VA06 grass, cassava hay, whole crop sunflower and sugarcane, high sugar contents (≥128 g/kg of DM) were found.

Except for fresh VA06 grass and rice straw, the reported composition of the feedstuffs fall within ranges previously reports [20,21]. The CP value of fresh VA06 grass found here was somewhat lower than previously reported which may be explained by the low level of N fertilization commonly practised in Vietnam. Although values as high as 75 and 84 g/kg DM have been reported [22,23] the relative high CP content of rice straw (87 g/kg DM) compared to other reports [23,24] may be explained by the harvesting of relatively young paddy rice. The analyses showed that the rice straw also contained some starch which is most likely explained by the presence of paddy rice grains left in the straw after threshing.

As mentioned above, the values related to cell wall constituents (CF, NDF, ADF, and ADL) are very high in fresh grasses. Such high values indicate that the fresh grasses were cut at a mature stage [25]. Indeed, fresh Pennisetum grasses are traditionally cut at a height of >200 cm in SE Asian countries to obtain a high DM yield per cut. Harvesting of fresh grasses at an advanced maturity stage also negatively affects the CP content of the grasses [26,27]. To the authors’ knowledge, the chemical composition of green bean shells has not been reported previously. Its relative high CP content indicates that it may have great potential in ruminant nutrition.

Variables such as the cumulative in vitro gas production at 72 h (GP-72 h), OMd, and total volatile fatty acid (TVFA) were considered as indices of the in vivo FOM content of the fresh grasses. Total gas production (GP-72 h) was influenced by genus but not by species within genus (Table 2) and mean values across species within genus (Brachiaria, Panicum, and Pennisetum) were found to be 222, 226, and 245 mL/g OM, respectively. The asymptote GP associated with the soluble fraction (A1) was similar between Brachiaria and Panicum but greater values were found for Pennisetum because of the high A1 value of VA06 grass. This high A1 value can be explained by the high sugar content (i.e., 138 g/kg DM) of VA06 grass (Table 1). The asymptote GP associated with the non-soluble fraction (A2) was lowest in the genus Brachiaria and highest in Pennisetum. Parameter A2 was significantly influenced by species within the genus Panicum but post-hoc probability values on type I error indicated that the difference in A2 values between Guinea grass and TD59 only tended to be different. The asymptote GP associated with microbial turnover (A3) was similar between genera but within Brachiaria, the A3 value of especially Para grass was lower compared to the other 3 grass species within this genus. Within Pennisetum, the A3 value was highest for Napier grass. The relevancy of A3 values in relation to feed evaluation, however, can be disputed [28]. As such, the asymptote GP values associated with both the soluble and non-soluble fractions of the grasses (i.e. A1 and A2, respectively) versus GP-72 values can be considered as more suitable to estimate the FOM value in not only the selected fresh grasses, but also the other selected feedstuffs (Table 1).

The in vitro OMd of fresh grasses (Table 3) were affected by genus as well as species within genera. The highest OMd values were found within the genus Pennisetum, which is line with the A1 and A2 values of this genus. Within the genus Pennisetum, however, the OMd of VA06 grass was lower compared to that of King grass and Napier which does not agree with the corresponding A1 and A2 values. The latter observation extents to the genera Brachiaria and Panicum, where also a discrepancy between OMd and the corresponding A1 and A2 values was observed. These discrepancies are not easy to explain but it can be speculated that the variation in OMd does not properly reflect the FOM content of the grasses due to the relative long incubation time, i.e. 72 h [28]. The absolute amount of these VFA is of critical importance to determine the absolute amounts (g/d) of milk fat and lactose synthesized in the mammary gland of dairy cows. The total production of VFA produced in the rumen is primarily determined by FOM intake [7]. In contrast to OMd, the TVFA concentration after 72 h of incubation (Table 3) was not affected by genus. The TVFA concentration was found to be affected by species within genera, but significant differences between specific grass within the genus Brachiaria could not be identified by Tukey’s test. Under the assumption that the A1+A2 values are the most suitable indicators of the FOM content of the fresh grasses, it can be concluded that King grass and VA06 are the most promising grasses to increase the FOM content of dairy rations in Vietnam and Thailand.

Variables such as the half-time of A1 and A2 and the maximum rate of GP related to phase 1 and 2 (i.e. C1, C2, Rmax1, and Rmax2) can be considered to be relevant to provide some insight into the degradation rate of OM. The half time of asymptote GP of the soluble fraction (C1, Table 2) was affected by genera but not by species within genera; the half-time of the maximum rate of GP was reached ~40 min earlier in the genus Brachiaria compared to genera Panicum and Pennisetum. Such a difference in C1 is, however, considered of minor practical interest. The maximum rate of GP related to phase 1 (i.e., Rmax1) was neither affected by genera nor by species within genera. The latter result most likely indicates that the chemical composition of the soluble fraction and their fermentability in fresh grasses is similar between the various grass species. The soluble fraction of tropical fresh grasses typically contains compounds such as sugars and it is generally accepted that sugars are fermented rapidly [29].

The half time of asymptote GP of the insoluble fraction of fresh grasses (C2, Table 2) was influenced by both genera and species within genera. On average, C2 values were ~13% lower in Pennisetum compared to Brachiaria and Panicum species. Within Pennisetum, the C2 value of VA06 was lower than Napier grass and tended (p<0.10) to be lower compared to that of King grass. Within the genus Brachiaria, the C2 value of Para grass was found to not differ to that of Humidicola with both were lower compared to Mulato II and Signal grass who were not different. The results on the maximum rate of GP of the insoluble fraction (i.e., Rmax2) were found to be influenced by both genera and species within genera. The highest Rmax2 values were found when Pennisetum species were incubated, but unlike the corresponding C2 values, the Rmax2 values were found to be not different for the three grass species within the genus Pennisetum. Within the genus Brachiaria, the highest Rmax2 value was found when Para grass was incubated but the value of Para grass was found to be not different to that of Humidicola. Overall, it can be concluded that the highest fermentation rates were observed when VA06, King grass or Para grass were fermented.

It is well established that especially Hac is an important precursor of fatty acid synthesis in the mammary gland of dairy cows. Proprionic acid, however, is an important precursor for gluconeogenesis and thus of great importance in the intramammary process of lactose synthesis. The amount of lactose produced in the mammary gland of dairy cows is the major determinant in the amount of milk produced. Thus, a high ratio between Hac and Hpr will potentially result in a greater proportion of fat content in milk. Typically, FOM originating from slowly degrading roughage, shifts rumen fermentation towards greater proportions of Hac and will thus potentially result in greater proportions of milk fat. The proportions (Table 3) of Hac, Hpr, BCVFA, and the ratios of Hac to Hpr (A:P) and NGR were affected by both genera and species within genera. Across genera, the highest proportion of Hac was found when Brachiaria was incubated while at the same time the lowest proportions of Hpr were found. Consequently, the A:P was found to be the lowest when Brachiaria species were incubated. Within the genus Brachiaria, the shift in VFA profile towards Hpr was most profound when Signal grass was incubated while in Panicum and Pennisetum, the incubation of Hamil or VA06, respectively were most influential to lower the A:P. Overall, NGR was similarly affected, because NGR and the A:P were highly correlated (i.e., r = 0.96, n = 11, p<0.05) due to the dominance of Hpr in calculating NGR and A:P. The proportion of Hbu was not affected by species within genera but differences tended to be different between the grasses in Brachiaria. The proportions of Hva were not affected by genera but were higher after the incubation of Signal grass (Brachiaria), Hamil (Panicum) and Napier (Pennisetum). The proportions of BCVFA were affected by genera and the highest values were found when either Panicum or Pennisetum was incubated. Within genera, the highest BCVFA values were found for Para grass (Brachiaria), Guinea grass and Hamil (Panicum) and Napier (Pennisetum).

In an attempt to find a relation between parameters describing GP kinetics (i.e., C1, C2, Rmax1, and RMax2) and the profile of VFA, the proportions of Hpr (mol/100 mol VFA), A:P, and NGR were regressed against GP kinetics parameters. Upon visual inspection of the graphs of each regression (data not shown), however, it appeared that the data obtained from Mulato II and Signal grass, both belonging to the genus Brachiaria, had a profound effect on all regressions. In these two grasses, the highest Hpr values and thus lowest values on the A:P and NGR were found (Table 5). The observations on the VFA profile after incubation of Mulato II and Signal grass are difficult to explain. Perhaps, these two grasses contain specific compounds in their insoluble fraction that enhance the synthesis of Hpr during rumen fermentation. In case the various regressions were conducted without these two grasses, no significant correlations were found (r≤0.10, n = 9, p>0.05) between the proportions of Hpr, A:P, NGR, and GP kinetic parameters. Overall, the data indicate that in vitro VFA fermentation values are similar between tropical grasses and that the practical relevance of the differences is considered of minor interest to dairy cow nutrition. The similarity in VFA patterns between the various topical grasses is most likely related to the fact that all grasses were cut at a mature state.

Across the grass species within genera, cumulative CH₄ production measured after 72 h of incubation (CH₄-72; Table 2) expressed as mL/g OM were affected by genera and species within genera. Pennisetum species had the highest absolute CH₄-72 values (44.5 mL/g OM), subsequently followed by Panicum (39.8 mL/g OM), and Brachiaria (37.4 mL/g OM) species. Within the genus Brachiaria, the highest CH₄-72 was found for Para grass, whereas the highest CH₄-72 values were found in Guinea and Napier grass in Panicum and Pennisetum, respectively. Statistically significant differences in the asymptote CH₄ production between specific genera and specific grasses within genera were not detected but across the 11 grass species, the A values correlated well with the cumulative CH₄ production (i.e., r = 0.91, n = 11, p<0.001). As expected, CH₄-72 (mL/g OM) depended on GP-72, and when CH₄ production was expressed relative to total GP (CH₄-:GP-72), the percentage of CH₄ were not different between genera. Within the genus Pennisetum, however, CH₄-:GP-72 was affected by grass species and the highest value was found for Napier grass. The C value of CH₄ production tended to be affected by genera but within genera the values were found to be not different between grass species. The Rmax of CH₄ was influenced by both the three genera and grass species within genera with the highest rate found when Pennisetum species were incubated (p<0.05) while within the genus Brachiaria, the highest Rmax value was found when Para grass was incubated whereas incubation of Mulato II resulted in the lowest values (p<0.05).

It is generally accepted that a shift in VFA production from Hac to Hpr renders less hydrogen available for the synthesis of CH₄ [30]. The current data were found to be in line with this mechanism. Across the 11 grass species, CH₄-:GP-72 positively correlated with the proportion of Hac in TVFA (r = 0.74, p = 0.009, n = 11) and negatively with the proportion of Hpr in TVFA (r = −0.82, p<0.01, n = 11). This also explains why Brachiaria had a higher level of Hpr with lower amount of CH₄ production compared to Panicum and Pennisetum. However, as mentioned before the proportions of Hpr for both Mulato II and Signal grass can be considered as outliers and thus may interfere with proper interpretation of the aforementioned correlations. In case the data of Mulato II and Signal grass were omitted, the relative CH₄ production remained negatively correlated with Hpr (r = −0.83, p<0.01, n = 9) and positively with the A:P (r = 0.80, p = 0.01, n = 9). The values on CH₄ kinetics (i.e., C values and Rmax) were found to be unrelated to the proportions of Hac, Hpr as well as A:P and NGR. With the exception of Mulato II and Signal grass, the Hpr values ranged from 19.8 to 22.1 mol/100 mol TVFA whereas the values on A:P ranged from 3.0 to 3.4 mol/mol (Table 3). Thus, despite the fact that the variation in Hpr and the A:P accounted for a significant part the variation in CH₄ production, the practical relevance of the variation in the proportions of Hpr and A:P can be questioned.

The mean value GP-72 h across the 10 feedstuffs of this group was found to be 218 mL/g OM (Table 4). The lowest values on the asymptote GP associated with non-soluble fraction (i.e. A2) was found when cassava top hay and Pangola hay were incubated but the corresponding numerical value of rice straw was only ~10% higher compared to Pangola hay. Such low A2 values might be due to the very low starch content of these feedstuffs. The lowest A1 value was found when corn silage was incubated because of a lack of sugar to enable rapid fermentation in the first three hours. The OMd and TVFA values were found to be lowest in Pangola hay and greatest in sorghum.

The in vitro gas kinetics of preserved- and whole crop roughages are presented in Table 4. The greatest value on the half time of asymptote GP of the soluble fraction (C1) was found in sugarcane which is most likely explained by the very high sugar content of this feedstuff. A negative correlation in phase 2 was found between the maximum rate of GP of the insoluble fraction (Rmax2) and the corresponding C2 values (r = −0.68, p = 0.016, n = 10). This was obvious in rice straw as it had the lowest Rmax2 value and the lowest C2 value.

As can be seen in Table 5, the selected feedstuffs of preserved roughages had relatively similar values of Hac percentage while fairly large differences were observed for the whole crops. Sugarcane had the lowest Hac value and greatest Hpr and produced the lowest NGR and A:P. It should be noted that Hac and A:P are not only a good indicator of milk fat synthesis but also equally affect animal performance, therefore, sunflower could be considered a good feed ingredient to increase the fat percentage of milk because of its high values for Hac and A:P.

To determine the strength of the relationship between GP kinetic values and VFA profiles, linear, single regressions were performed. The C1 values were found to be negatively correlated with Hac, A:P and NGR (r = −0.76, −0.74, −0.76, with p = 0.01 for all) while the C1 values were positively correlated with Hpr (r = 0.80, p = 0.01). Thus, the feedstuffs with a higher C1 value are likely less suitable to increase the fat content of milk.

After 72 h incubation, CH₄ production (mL/g degradable OM) was greatest for sunflower. However, when expressed as percentage of GP-72, production was the greatest for rice straw. Methane proportion was positively correlated with Hac, A:P and NGR (r = 0.73, 0.75, 0.75 and p = 0.02, 0.01, 0.01, respectively) while it was negatively correlated with Hpr (r = −0.77, p = 0.01, n = 10). Generally, whole crop roughages had higher GP with a lower CH₄ proportion compared to preserved feeds.

In general, sorghum is the most interesting feedstuff among the preserved- and whole crop roughages when optimizing dairy rations due to its high FOM content (GP, A1, A2, OMd) and TVFA while it had the lowest CH₄ production when expressed as a % of total GP.

Across eight selected feedstuffs of the single concentrate group (Supplementary Table S1 and S2), it was noted that within approximately 20 h of incubation, more than half of the amount of gas was produced. The concentrate feeds were found to lead to numerically greater GP and OMd than other types of feedstuffs, assumingly due to higher contents of soluble carbohydrates, except for rice bran that had the lowest digestibility values among all the concentrate feeds.

Starch rich feedstuffs had high TVFA and Hbu percentage. Among the concentrate feedstuffs, rice bran had the lowest TVFA, whilst the highest molar proportion of Hac was found for green bean shells. The highest value of Hpr was observed for cassava waste whereas the highest values in TVFA, NGR, CH₄ production in terms of mL/g degradable OM and the highest rate of GP were reported in cassava peeled tuber.