Approximately 3 liters of raw Sumbawa mare’s milk were purchased directly from local producers located near the Curi Mori Karamabura partner site. The milk was obtained during routine morning milking and immediately placed into sterile containers in accordance with standard hygiene and sanitation procedures. After collection, the milk was portioned into sterile breast milk storage bags with a capacity of 150 mL using aseptic handling techniques. Throughout collection and initial handling, the temperature of the milk was maintained at approximately 4 ± 1 °C. The samples were then transported to the laboratory within about 2 hours using dry ice and subsequently stored at −80 °C until further processing and analysis.

Sumbawa mare’s milk was processed into three different forms: pasteurized liquid milk (SKC), powdered milk supplemented with 15% dextrin (SKD), and powdered milk without dextrin (SKT). Before treatment preparation, raw mare’s milk was pasteurized using the low-temperature long-time (LTLT) method at 63 °C for 30 min in a thermostatically controlled water bath (Memmert WNB 14, Memmert GmbH, Germany), followed by immediate cooling to 4 °C. For each treatment, 500 mL of pasteurized milk was used. The powdered products were prepared separately using a vacuum-drying technique. A digital vacuum drying oven was selected because it is more suitable for laboratory-scale processing and provides lower oxygen exposure and milder thermal conditions than spray drying, which may help preserve the detectable microbial profile prior to sequencing. In the SKD treatment, the milk was supplemented with 15% dextrin and whipped to produce stable foam prior to the drying process. The procedure followed the method described in our previous study (Sari et al., 2025). Prior to processing, the milk was weighed and mixed with 1% Tween 80 as a food-grade emulsifier, while dextrin at a concentration of 15% was added according to the treatment design.

The use of 15% dextrin was determined based on preliminary optimization experiments carried out in our laboratory. These trials indicated that this concentration achieved a suitable balance between drying performance, powder stability, and the prevention of protein aggregation during dehydration. Similar carrier concentrations have also been reported in dairy powder processing applications (Tița et al., 2024). In addition, Tween 80 (1%) was added to improve the dispersion of milk components, support stable foam formation prior to drying, and reduce phase separation during dehydration. The use of such emulsifiers is commonly reported in vacuum- and spray-drying systems involving carbohydrate–protein matrices (Exarhopoulos et al., 2025).

Drying was carried out using a digital vacuum drying oven (VOV-91, B-One Scientific, Korea) operated at 70 °C under a vacuum pressure of −66 cmHg for 7 hours. The resulting powders had final moisture contents of 4.33 ± 0.14% for powdered milk supplemented with 15% dextrin (SKD) and 4.54 ± 0.13% for powdered milk without dextrin (SKT). After drying, the resulting flakes were milled using a laboratory grinder (FGD-20, FOMAC, Indonesia) for 1–2 minutes to obtain a fine and homogeneous powder. Three milk batches were prepared, and each treatment was produced in triplicate to ensure consistency during the drying process. One representative sample from each treatment was subsequently selected for sequencing. Each selected sample was analyzed twice during the sequencing workflow to confirm technical consistency. The powdered samples were then stored in a desiccator at room temperature and kept protected from light until further analysis.

Prior to DNA extraction, the three horse milk samples—SKC (pasteurized liquid milk), SKD (powdered horse milk with dextrin), and SKT (powdered horse milk without dextrin)—were gently homogenized to ensure an even distribution of microbial cells. An aliquot of 1–2 mL from each sample was then transferred into sterile microcentrifuge tubes and centrifuged at 10,000 × g for 10 minutes to concentrate the microbial biomass. After centrifugation, the supernatant was carefully removed, while the pellet containing microbial cells was retained for genomic DNA extraction. All procedures were carried out under sterile conditions to minimize the risk of external contamination.

Three types of Sumbawa mare’s milk samples—SKC, SKD, and SKT—were analyzed to characterize their microbial communities. For each treatment, one independently prepared milk batch was selected for full-length 16S rRNA gene sequencing. To improve sequencing reliability and maintain technical consistency, two technical replicates were included during the next-generation sequencing (NGS) workflow. These replicates served exclusively for quality-control purposes and were not treated as independent biological replicates.

Genomic DNA was extracted using the ZymoBIOMICS DNA Miniprep Kit (Zymo Research, D4300) following the manufacturer’s instructions, which included mechanical lysis through bead-beating to facilitate effective disruption of microbial cells. All extraction steps were carried out in a sterilized laminar flow hood using DNA-free consumables to reduce the risk of contamination. A negative extraction control (blank control) was included in each extraction batch and processed through library preparation and sequencing to detect any potential background contamination. DNA concentration and purity were evaluated using a Qubit fluorometer and a NanoDrop spectrophotometer, respectively.

Sequencing was performed using the Oxford Nanopore Technologies (ONT) platform, which enables long-read sequencing of the full-length 16S rRNA gene encompassing the V1–V9 regions, thereby providing improved taxonomic resolution (Bahram et al., 2019). All procedures, including DNA extraction, library preparation, and sequencing, were conducted at PT Genetika Science Indonesia. Library preparation followed the standard ONT protocol using official ONT kits (Indonesia, 2023). Sequencing runs were carried out on a MinION device operated with MinKNOW software version 23.04.5, while basecalling was performed using Guppy software version 6.5.7 with the high-accuracy model(Wick et al., 2019). Prior to taxonomic classification, quality control and data preprocessing were conducted. The resulting FASTQ files were first examined using NanoPlot version 1.41.0 to evaluate read length distribution and sequence quality metrics. Low-quality reads were subsequently filtered using NanoFilt version 2.8.0, applying a minimum Q-score threshold of ≥7 and a minimum read length of ≥500 bp, ensuring that only high-quality reads were retained for downstream analysis (Coster et al., 2018); (Nygaard et al., 2020).

Taxonomic assignment was carried out using Centrifuge software (Kim et al., 2016). The reference database included bacterial and archaeal sequences obtained from the NCBI 16S RefSeq database (https://ftp.ncbi.nlm.nih.gov/refseq/TargetedLoci/). The use of full-length 16S rRNA gene sequences allowed improved taxonomic resolution, enabling putative identification of microbial taxa at the species level.

Relative abundances of microbial taxa were derived from the 16S rRNA sequencing data and used to generate community profiles at the phylum, genus, and species levels. To ensure biological relevance and minimize analytical noise, only taxa with a relative abundance of ≥1% were considered. Within this threshold, analyses were further restricted to the ten most abundant taxa (Supamri et al., 2025)

Alpha diversity metrics, including the Shannon, Simpson, and Inverse Simpson indices, were calculated to characterize patterns of microbial richness and evenness among treatments. For each treatment, one representative sample was selected for sequencing and analyzed twice during the sequencing workflow to confirm technical consistency. Descriptive statistics (mean ± standard deviation) were calculated from these duplicate measurements. Because the dataset was derived from repeated technical measurements of the same sample rather than from independent samples, the alpha diversity results were interpreted descriptively to identify trends in microbial diversity among treatments. Sankey diagrams were constructed from the relative abundance matrix to visualize microbial community dynamics across treatments, with branch width representing the number of operational taxonomic units (OTUs). All analyses and data visualizations, including relative abundance calculations, stacked bar plots, alpha diversity boxplots, Sankey diagrams, and principal component analysis (PCA), were performed using OriginPro 2024 (OriginLab Corporation, Northampton, MA, USA).

The analysis of microbial composition revealed that ten phyla were detected with relative abundances exceeding 1%, whereas the remaining taxa occurred below this threshold. Among these groups, Bacillota represented the predominant phylum across all treatments (Figure 1a). Its relative abundance was particularly high in the SKC (98%) and SKD (95%) samples, indicating a consistent dominance of the core microbial community. In contrast, in SKT, the abundance of Bacillota drastically decreased to 60%, accompanied by an increase in the proportion of Cyanobacteriota (31%), Pseudomonadota (5%), and Planctomycetota (2%). In addition, other taxa such as Actinomycetota (1.3%) and Bacteroidota (1.2%) were also noticeably present in SKT.

At the genus level, five taxa were identified with a relative abundance above 1%, while the other genera were below that threshold (Figure 1b). SKC and SKD showed very strong dominance by Lactobacillus, at 96% and 86%, respectively. In contrast, SKT displayed a more even and complex microbial community distribution, with various dominant genera such as Lysinibacillus (20%), Bacillus (19%), and Brevibacillus (16%). Other genera with proportions above 1% in SKT included Thalassoporum (12%) and Paenibacillus (4%).

Seven taxa at the species level had a relative abundance above 1%, while other species were detected in lower amounts (Figure 1c). SKC and SKD showed relatively similar dominant compositions, marked by the presence of Lactobacillus helveticus (SKC: 32%; SKD: 30%), L. gallinarum (SKC: 39%; SKD: 40%), and L. crispatus (9% in both samples). In contrast, the SKT samples exhibited a shift in microbial community composition, characterized by the predominance of Brevibacillus parabrevis (26%), Thalassoporum komareki (12%), and Lysinibacillus boronitolerans (10%). Lower but still notable abundances were also observed for members of the Bacillus cereus group (4%) and L. helveticus (4%).

These observations suggest that the form of horse milk and the drying process substantially influence the structure and diversity of the microbial community. This interpretation is supported by the microbial community profiles shown in Figure 1, which illustrate clear differences among the treatments.

Because the dataset was derived from technical replicates, the statistical analyses presented here should be interpreted as exploratory. Alpha diversity indices calculated from the 16S rRNA gene sequencing data are summarized in Table 1 and illustrated in Figure 2. The number of observed species (Figure 2a) increased markedly across treatments, from 190 in SKC to 963 in SKD and 1185 in SKT. This pattern indicates that powdered horse milk samples were associated with a higher detected taxonomic richness under the sequencing and analytical conditions applied in this study compared with the liquid milk sample (SKC). Overall, these findings suggest that processing horse milk into powdered form is linked to a broader detection of microbial taxa within the analytical framework used in this research.

The Shannon diversity index (Figure 2b) likewise increased across treatments, rising from 2.02 ± 0.06 in SKC to 2.40 ± 0.99 in SKD and reaching 3.43 ± 0.23 in SKT. This trend suggests a gradual increase in microbial diversity and evenness from pasteurized liquid milk to powdered milk without dextrin under the applied processing conditions. In contrast, the Simpson index (Figure 2c) displayed the opposite trend, decreasing from 0.18 ± 0.03 in SKC and 0.17 ± 0.01 in SKD to 0.07 ± 0.00 in SKT. Because the Simpson index is inversely related to diversity, the lower value observed in SKT indicates a more even and diverse microbial community compared with SKC and SKD.

A similar trend was observed for the Inverse Simpson index (Figure 2d), which increased substantially from 5.61 ± 0.99 in SKC and 5.93 ± 0.50 in SKD to 14.26 ± 0.57 in SKT. This pattern suggests a more complex microbial composition in SKT under the applied experimental conditions. The consistency among the alpha diversity metrics indicates that the different milk treatments, particularly powder production without dextrin, were associated with distinct patterns of microbial richness and evenness. Overall, these results suggest that the drying process and dextrin supplementation may influence the detectable microbial diversity of horse milk.

The Sankey diagram depicts the hierarchical distribution of microbial taxa from the phylum to the species level across the different treatments (Figure 3). To facilitate meaningful comparison among samples, only taxa with a relative abundance of ≥1% were considered in the visualization. In SKC (Figure 3a), the microbial community was largely dominated by members of the family Lactobacillaceae, particularly the genera Lactobacillus, Lactococcus, and Enterococcus. The SKD sample (Figure 3b) exhibited an intermediate pattern, sharing characteristics with both SKC and SKT. By contrast, SKT (Figure 3c) showed a more diverse taxonomic composition, including representatives of the families Bacillaceae and Paenibacillaceae. Dominant genera detected in this treatment included Bacillus, Paenibacillus, Lysinibacillus,Brevibacillus, Thalassoporum, and Gloeothece under the sequencing and processing conditions applied in this study.

The Venn diagram (Figure 4) presents the distribution of operational taxonomic units (OTUs) detected in SKC, SKD, and SKT, highlighting both shared and unique microbial members among the treatments. Only OTUs that passed the quality-filtering step were considered in the analysis. Among the samples, SKT showed the highest OTU richness (944 OTUs), followed by SKD (310 OTUs) and SKC (72 OTUs), suggesting a greater number of detectable taxa in the powdered horse milk samples under the sequencing and processing conditions applied in this study. A total of 74 OTUs were common to all three treatments, representing the core microbial community based on presence–absence within this dataset. In addition, 151 OTUs were shared between SKT and SKD, 28 OTUs between SKC and SKD, and 16 OTUs between SKC and SKT.

These quantitative patterns align with the alpha diversity results presented in Table 1, where SKT showed the highest observed OTU richness, followed by SKD and SKC. Although several dominant lactic acid bacteria—including Lactobacillus gallinarum, L. helveticus, and L. crispatus—were identified in all treatments, their relative abundances varied noticeably among the different processing conditions. Samples from SKC and SKD were characterized by a greater relative contribution of lactic acid bacteria, whereas SKT showed increased representation of spore-forming genera such as Bacillus and Brevibacillus. This pattern suggests a shift in the detected microbial composition associated with the drying process without dextrin.

Principal Component Analysis (PCA) revealed a separation pattern of microbial community structures among treatments based on relative abundance profiles derived from technical replicates. Two principal components, PC1 (34.01%) and PC2 (33.61%), explained 67.62% of the total variation in the data. SKT samples were predominantly distributed in the positive quadrants of both PC1 and PC2, indicating a distinct ordination pattern under the applied processing and analytical conditions. SKC was positioned in the positive PC1 and negative PC2 quadrant, while SKD was located in the negative quadrants of both components. This distribution reflects processing-associated differences in detected microbial composition, with minimal overlap observed between the samples. within this exploratory ordination space. This pattern is clearly illustrated in Figure 5.

Substrate form and the incorporation of functional carbohydrates such as dextrin have been reported to influence microbial community structures in fermented dairy products (Kaur et al., 2022); (Ge et al., 2025). In the present study, clear differences in microbial diversity were observed among the treatment groups. These variations reflect patterns associated with the different milk treatments. Powdered horse milk without dextrin (SKT) showed the highest microbial diversity, characterized by a relatively balanced distribution at the genus level and the presence of less commonly reported genera, including Lysinibacillus, Thalassoporum, and Brevibacillus, within the dataset. By contrast, both pasteurized liquid milk (SKC) and powdered horse milk supplemented with dextrin (SKD) were strongly dominated by Lactobacillus, which comprised more than 85% of the detected community. The greater microbial diversity observed in SKT may be related to the absence of dextrin during the drying process. Without a protective carbohydrate matrix, the relative dominance of heat-sensitive lactic acid bacteria may have decreased, allowing stress-tolerant and spore-forming taxa to become more detectable under the applied sequencing conditions (Selmi et al., 2026). In addition, the low water activity typically associated with drying may contribute to the persistence of genera such as Bacillus, Brevibacillus, and Lysinibacillus, which are commonly reported as tolerant to harsh environmental conditions (Saha et al., 2025).

This pattern of dominance aligns with findings reported in probiotic goat milk and fermented camel milk, where Lactobacillus species often prevail due to their tolerance to acidic conditions and competitive advantage during fermentation (Monareng et al., 2025); (Shori, 2024). In the present dataset, SKC largely maintained microbial characteristics typically associated with fresh milk. The SKD samples exhibited a slightly more heterogeneous community structure, whereas SKT showed a greater presence of spore-forming and stress-tolerant taxa under the drying conditions applied in this study. These observations are in agreement with previous reports suggesting that thermal processing and storage conditions can affect microbial survival and resilience (García et al., 2023); (Hebishy et al., 2023). The compositional patterns observed were further reflected in the alpha diversity metrics (Observed, Shannon, Simpson, and Inverse Simpson), which showed higher microbial richness and evenness in SKT within the analyzed dataset.

These observations suggest that powder production without dextrin was associated with the detection of a broader range of microorganisms within the analytical scope of this study. From an ecological perspective, the higher diversity observed in SKT indicates a more complex microbial community within the dataset, where spore-forming and stress-tolerant taxa occur alongside lactic acid bacteria under the experimental conditions applied (Sibanda et al., 2024). Such community patterns may reflect potential ecological resilience rather than demonstrating confirmed functional advantages (Tan et al., 2022).

The differences observed in microbial community profiles may also be influenced by physicochemical factors associated with the drying process. Dextrin functions as a carbohydrate carrier capable of forming a semi-crystalline matrix around microbial cells, which may help reduce thermal and osmotic stress during dehydration (Bamidele & Emmambux, 2021). In the SKD samples, this matrix may have contributed to the persistence of lactic acid bacteria, particularly Lactobacillus spp., thereby supporting a moderate level of diversity within the analyzed sample. By contrast, the absence of dextrin in SKT likely exposed microorganisms to greater stress conditions, which may have favored the detection of stress-tolerant or spore-forming taxa such as Bacillus and Brevibacillus. Although water activity (a_w) was not directly measured in this study, the drying process was conducted at 70 °C and produced powders with low final moisture contents. Under such low-moisture conditions, bacterial cellular proteins and membrane integrity may be affected, whereas spore-forming microorganisms generally remain more resistant (Tong et al., 2025); (Shymialevich et al., 2024). Therefore, the observed microbial differences may be associated with the drying conditions applied, although this interpretation remains tentative because a_w was not determined experimentally. These differences in microbial composition may also influence the quality characteristics of the final product. The greater persistence of lactic acid bacteria in SKD may be relevant to maintaining microbial characteristics commonly associated with fermented milk products commonly associated with fermented milk, including a more stable microbial profile and possible contributions to flavor development. In contrast, the higher representation of spore-forming taxa in SKT may be relevant to product stability during storage, as some species of Bacillus and Brevibacillus are known to survive harsh processing conditions and may contribute to spoilage in dried dairy products (Medjahdi et al., 2025). Consequently, the microbial structure observed in the powdered horse milk samples likely reflects interactions between the presence of dextrin and the drying conditions applied in this experiment.

From a microbial ecology standpoint, the higher alpha diversity detected in SKT may indicate a broader metabolic potential within the microbial community identified in this dataset; however, such interpretations remain descriptive and exploratory. The simultaneous presence of both spore-forming and non-spore-forming taxa under the applied processing conditions suggests a pattern consistent with survival under environmental stress rather than evidence of confirmed ecological adaptation (Manyi-Loh & Lues, 2025); (Toor et al., 2024). Additionally, the identification of microbial taxa uniquely detected in SKT, including Lysinibacillusboronitolerans and Thalassoporumkomareki, represents analytical findings within the current dataset, and their functional roles were not experimentally verified in this study (Tuesta-Popolizio et al., 2021).

Taxonomic shifts were also reflected in the dominance of particular taxa across the treatments. Although several core microbial groups were detected in all samples, their relative abundances varied within the limits of the technical replicate dataset. Samples from SKC and SKD were primarily characterized by the presence of species such as L. helveticus and L. gallinarum. In contrast, SKT showed relatively higher detection of Brevibacillusparabrevis and Bacillus spp., which may be associated with the processing conditions applied. These descriptive observations are consistent with the alpha diversity patterns reported earlier, although they should not be interpreted as statistical evidence of population-level differences. Previous research has also shown that the use of additives can influence patterns of microbial dominance in processed food systems, providing contextual support for these observations (Xiong et al., 2022). Taken together, the compositional differences observed among treatments indicate that each sample type exhibited a distinct microbial profile under the applied experimental conditions, with SKT displaying the broadest taxonomic representation. These results highlight processing-related trends rather than confirmed biological effects and suggest that food processing practices, such as those applied to horse milk in this study, can influence the detected structure and relative abundance of microbial communities (Blanco-Doval et al., 2024); (García et al., 2023).

Several limitations of the present study should be considered. First, the microbial analyses were based on technical replicates derived from a limited number of processed samples; therefore, the patterns observed should be interpreted as analytical trends associated with processing conditions rather than as indicators of population-level biological variability. Second, key physicochemical parameters—including water activity (a_w), drying kinetics, and temperature gradients—were not measured directly, which restricts the ability to relate the observed microbial shifts to specific processing factors quantitatively. Furthermore, microbial profiling was performed using full-length 16S rRNA gene sequencing, which provides detailed taxonomic information but does not permit direct assessment of microbial functionality, viability, or metabolic activity. As a result, the ecological and biotechnological interpretations presented here remain exploratory and would benefit from further validation through studies that include biological replication, functional characterization, and genome-resolved analytical approaches. Despite these limitations, the present findings provide preliminary descriptive information relevant to the milk powder industry, particularly in the selection of drying conditions and carrier additives to help maintain desirable microbial groups and improve the microbiological stability of horse milk powder products.