The Impact of Whole‐Animal Fluid Preservation on the Observed Gut Microbiome of Vertebrates: Implications for the Use of Museum Specimens in Microbiome Research

ABSTRACT

The vertebrate gut houses diverse microbial communities that provide insights into their host's ecological and evolutionary histories. Nevertheless, microbiome research has not been distributed equally across host taxonomy, geography and timescales. The millions of fluid‐preserved specimens stored in natural history museums worldwide represent a potentially untapped resource for microbiome information. However, it is unknown how fluid preservation and long‐term storage change the composition and diversity of the original microbial community across a variety of host taxa. Here, we present the largest study to date aimed at addressing this question. Specifically, we identified an optimal method for extracting DNA from preserved samples using commercially available kits. Next, for 11 host species representing four vertebrate classes, we compared the gut microbiomes between animals dissected freshly and those collected simultaneously but subsequently fixed in formalin and stored in 70% ethanol for 1 year, similar to museum conditions. In a secondary analysis in amphibians, we compared our collected samples with those from decades‐old historical museum specimens. We found that while fluid preservation altered the community composition and reduced the diversity of the recovered microbiome inventories, host species identity predominated in shaping the gut microbiome, and differences across species and geographic localities were retained after preservation. Historical specimens had microbiomes that were the most different from fresh specimens, suggesting that over time, changes in the microbiome of populations have occurred, or preservation effects have compounded. Considering these findings, we discuss the potential for use of fluid‐preserved museum specimens in future microbiome studies.

1. Introduction

The vertebrate gut harbours dense and diverse microbial communities that are distinct across species in a manner often congruent with host phylogeny (Brooks et al. 2016; Ley et al. 2008; Ochman et al. 2010; Phillips et al. 2012). These communities perform a suite of biological services, providing physiological benefits to their hosts and facilitating diversification in host ecological strategies, such as diet choice and seasonal acclimation (Kohl and Carey 2016; Kohl, Weiss, et al. 2014; Sommer et al. 2016). While gut microbiome research has rapidly increased in recent years, the majority of studies have focused on humans and laboratory‐reared model organisms (Pascoe et al. 2017; Woodhams et al. 2020). Broadening focus to a wider variety of wild host taxa will enhance understanding of host biology and the ecological and evolutionary implications of host–microbe interactions (Colston and Jackson 2016; Hird 2017). Garnering this information ultimately relies on the ability to collect representative samples from the host species of interest. Currently, gut microbiomes are typically analysed from swabs or faecal samples collected from live animals or after euthanasia and dissection of the intestinal tract. This process requires capture and, often, sacrifice of the animal, which may be difficult or impossible depending on the species' rarity or protected status, and which can cause undue handling stress. Additionally, these methods only provide a snapshot of the microbiome at the time of sampling, and longitudinal studies, especially in natural systems, can be challenging (Björk et al. 2019).

Faecal samples stored in frozen repositories that were collected before the advent of modern microbiome sampling techniques have recently been used to study microbiomes of the past and glean notable insights into host ecology (Björk et al. 2022; Risely et al. 2021). Similarly, museum specimens may represent an untapped historical resource that can be used for modern microbiome studies. With an estimated three billion specimens representing two million species stored in public collections worldwide, natural history museums are rich repositories of biological diversity (Yeates et al. 2016). Specimens in which the whole animal is preserved in fluid (e.g., ethanol or formalin) retain gut contents (Sasa and Monrós 2000; Harmon et al. 2019), and utilising these specimens for microbiome research could open large avenues for future research by increasing sampling accessibility across host taxonomic, geographic and temporal scales. However, it is unclear how museum methods of animal preservation and storage may affect recovered microbial inventories. For example, museum fluid preservation often involves fixation of specimens in formalin which can degrade DNA, complicating extraction for downstream analyses (Miething et al. 2006). Studies with fresh samples have shown that sample storage length and medium, such as ethanol or nucleic acid stabilising reagents, affect aspects of sample integrity, including DNA concentration and resulting microbial community composition (Vlčková et al. 2012; Hale et al. 2015, 2016). However, despite these sample storage effects, individual and host species identity tend to be the strongest factors shaping the observed gut microbial community (Song et al. 2016; Carroll et al. 2012; Blekhman et al. 2016). Outside of the gut, a recent study recovered unique microbial communities in different tissue types (bone, liver and skin) of ethanol‐preserved museum snake specimens (~56–178 years old), which structure by individual (Zacho et al. 2021). Understanding if similar trends are true for the gut microbiome of museum‐preserved specimens will enhance their utility for future studies.

Some studies have assessed the impact of whole‐animal fluid‐preservation on the gut microbial community, generally concluding that preservation alters the diversity and composition of the observed bacterial gut microbiome (Heindler et al. 2018; Greiman et al. 2020; Bodawatta et al. 2020; Chalifour et al. 2022). However, these studies are limited in having either focused only on a single host species (Bodawatta et al. 2020; Chalifour et al. 2022) or having compared recently collected samples to those from museum specimens collected during an entirely different time period (Heindler et al. 2018; Greiman et al. 2020). Studies from single host species make it difficult to determine whether preservation effects are minimal relative to microbiome differences across host species. Those comparing the microbiome of animals collected at different time periods are unable to disentangle the effects due to preservation from actual changes in the microbiome of a population that may occur over time or due to annual variation (Degnan et al. 2012). Preservation time periods investigated have also varied. In studies comparing contemporary samples to those from natural history collections, the length of specimen preservation has varied from 6 to 117 years (Heindler et al. 2018; Greiman et al. 2020). In cases where freshly dissected and preserved animals were collected simultaneously as part of the study design, fresh specimen preservation time ranged from 2 weeks to 9 months (Bodawatta et al. 2020; Chalifour et al. 2022). Studies focused on the microbiomes of preserved specimens would also benefit from more consideration of how to optimise study methodology. For example, DNA extraction methods can vary in their efficacy for a particular sample type (Claassen et al. 2013), and the microbiome diversity metrics used for analyses can differ in their sensitivity to factors like amplification biases (Hu et al. 2021).

Here, we compared the gut microbiome of 11 animal species representing four vertebrate classes, and within each species, all animals were collected from the same populations simultaneously and divided into two sample conditions: freshly dissected or fluid‐preserved for 1 year prior to sample collection. With these data we (1) identified an optimal method to extract DNA from formalin‐fixed specimens for microbiome studies, (2) determined the effects of fluid preservation on the microbiome and (3) tested if species‐specific differences in microbiomes were retained despite storage. We coupled this approach with an analysis of samples from a natural history museum collection for a subset of species to determine how microbial communities may change over longer periods of preservation.

2. Materials and Methods

2.1. Animal and Sample Collection

We chose 11 study species (Table 1) from four vertebrate classes that are abundant and geographically widespread in Pennsylvania (PA, USA). During the summer of 2018, we collected 12–20 adult individuals of each species from two sites in western PA: Pymatuning Laboratory of Ecology (PLE; University of Pittsburgh, Linesville, PA) and Powdermill Nature Reserve (PNR; Carnegie Museum of Natural History, Rector, PA). Animals were captured by hand through visual encounter surveys or live trapping. For most species, individuals were captured from a single site; however, individuals were collected evenly from both sites for two amphibians ( Lithobates clamitans and Plethodon cinereus ; Table 1).

TABLE 1.

List of 11 study species from four vertebrate classes and their sample sizes across treatment groups. The number of samples collected from freshly dissected individuals, individuals preserved for 1 year, or historical specimens from the Carnegie Museum of Natural History (CMNH) are listed. The total number of samples from individuals collected from either of two geographic sites, Pymatuning Laboratory of Ecology (PLE) and Powdermill Nature Reserve (PNR), is listed for each group. Species identifiers used in figures and tables throughout the manuscript are listed in parentheses after each species name.

Class	Species name and (abbrv.)	Freshly collected (N)	Preserved 1 year (N)	Historical (N)
Ray‐FinnedFishes	Bluegill Lepomis macrochirus (LM)	9 PLE	10 PLE	0
	Yellow Perch Perca flavescens (PF)	10 PLE	10 PLE	0
Amphibians	American toad Anaxyrus americanus (AA)	10 PLE	10 PLE	10 PNR
	Green frog Lithobates clamitans (LC)	5 PLE 5 PNR	5 PLE 5 PNR	10 PNR
	Dusky salamander Desmognathus ochrophaeus (DO)	10 PLE	10 PLE	10 PNR
	Slimy salamander Plethodon glutinosus (PG)	6 PNR	7 PNR	10 PNR
	Red‐backed salamander Plethodon cinereus (PC)	5 PLE 5 PNR	6 PLE 4 PNR	10 PNR
	Eastern newt Notophthalmus viridescens (NV)	8 PNR	8 PNR	10 PNR
Reptiles	Painted turtle Chrysemys picta (CP)	6 PLE	6 PLE	0
	Garter snake Thamnophis sirtalis (TS)	8 PLE	9 PLE	0
Mammal	Deer mouse Peromyscus maniculatus (PM)	10 PLE	10 PLE	0

Upon capture, animals were transported to field station laboratories and euthanised chemically using MS‐222 or isoflurane inhalation depending on the species, followed by pithing. Individuals from each species were then randomly assigned to a sample condition: fresh dissection or fluid preservation. Animals assigned to the fresh group were immediately dissected to remove the contents of the large intestine from the gastrointestinal tract, except in mice, where contents were collected from the cecum. The gut section was removed from the animal and held against the inside of a sterile tube using forceps. We applied pressure to the outside of the gut tissue in a downward motion using a dissection spatula to push the contents into the tube. Between individuals, we changed gloves, washed all dissection supplies with antibacterial soap and water, and dried and wiped them with 100% ethanol. Samples were stored in tubes without preservative and kept on dry ice until return from the field, after which they were stored at −80°C.

Animals assigned to the fluid preservation group were injected throughout the body cavity with 10% formalin until full and plump and placed in glass jars of 10% formalin for a fixation period of 24 h. Due to their small body size, salamanders were fixed in formalin without injections. After fixation, individuals were placed in new glass jars of 70% ethanol with other individuals of the same species for long‐term storage in the laboratory. These methods were meant to mimic natural history museum preservation, which typically involves initial specimen fixation in formalin and long‐term storage in ethanol (Pisani 1973). After 1 year of storage at room temperature, preserved individuals were dissected exactly as the fresh animals to collect the contents of the large intestine, or cecum in the case of mice. Samples were then stored in tubes without preservative at −80°C until processing.

We collected samples from 10 individuals of each of the six amphibian species from fluid‐preserved specimens in the Amphibian and Reptile collection at Carnegie Museum of Natural History (CMNH; Table 1 and Table S1) and we term this sample condition ‘historical’. These specimens were originally captured at PNR in the 1980s during a pitfall trap study (Meshaka Jr. 2009; Martin et al. 2021). Pitfall traps were filled with formalin to fix specimens, and they were subsequently stored in the museum in 70% ethanol for at least 36 years at the time of contemporary microbiome sampling. We dissected each specimen and collected samples of large intestine contents as described above. Specimens were dissected at the museum and samples were transported to the laboratory on dry ice and subsequently stored at −80°C in tubes without preservative.

2.2. Comparison of DNA Extraction Methods and Sequencing

Before extracting DNA from all samples, we chose one species from each vertebrate class (Lepomis machrochirus—fish, Anaxyrus americanus —amphibian, Chrysemys picta —reptile, Peromyscus maniculatus —mammal) to compare three DNA extraction methods which combined the use of three commercially available kits from QIAGEN. For the method we termed ‘PW’, we extracted DNA from samples using the QIAamp PowerFecal DNA Kit following the manufacturer's protocol. For the method we termed ‘FM’, we extracted DNA from samples using the QIAamp DNA FFPE Tissue Kit, which is designed to extract DNA from formalin‐fixed paraffin‐embedded tissues. We skipped steps 1–9 of the manufacturer's protocol (because our samples were not embedded in paraffin) and began with step 10 following the rest of the protocol. For the method we termed ‘CL’, we extracted DNA from samples following the FM method but added a DNA cleaning step using the DNeasy PowerClean CleanUp Kit following the manufacturer's protocol. Samples of gut contents were briefly thawed, homogenised and divided into three equal portions, one for each DNA extraction method. After extraction, DNA concentration was calculated spectrophotometrically using a BioTek Microplate reader and frozen at −20°C until sequencing.

We sent 20 μL of DNA per sample to the DNA Services Facility at the University of Illinois, Chicago for PCR, library preparation and amplicon sequencing of the bacterial 16S rRNA gene on the Illumina MiSeq platform (Caporaso et al. 2012). The V4 region of the gene was amplified using primers 515F and 806R (Caporaso et al. 2012) and sequenced, generating 2 × 250 base‐paired end reads. Raw sequence reads were processed using QIIME2 version 2019.7 (Bolyen et al. 2019). Sequences were filtered for quality and clustered into amplicon sequence variants (ASVs) using the DADA2 pipeline (Callahan et al. 2016). ASVs were assigned taxonomy using the SILVA database classifier release 132 (Quast et al. 2012), and any sequences identified as chloroplast, mitochondria or archaea were removed from analysis. A phylogenetic tree of ASVs was built using FASTTREE (Price et al. 2010). We created separate ASV tables for each of the four species and computed Bray‐Curtis distance across samples for each using QIIME2.

2.3. DNA Extraction and Sequencing for All Samples

We chose to extract DNA from the full sample set and two extraction blanks using the CL method (see Results). We sent 20 μL of extracted DNA per sample to the Microbial Analysis, Resources and Services facility at the University of Connecticut for 16S rRNA amplicon sequencing following the same methods described above. To eliminate any potential facility or sequencing batch effects, we sent a new aliquot of DNA for sequencing from samples previously analysed during our DNA extraction method comparison. Only sequences from this second run were used for subsequent analyses. Raw sequences were processed in QIIME2 as described above. Our extraction kit control samples returned a total of 4601 sequences representing 137 ASVs. To identify and remove ASVs from our dataset likely to be extraction kit contaminants, we used the package ‘decontam’ in R version 3.6.3 (Davis et al. 2018). We utilised the prevalence‐based method with the two DNA extraction blanks as negative controls and the threshold probability for detection as a contaminant set to the default of 0.1. The program identified 81 ASVs as potential contaminants (Figure S1) which were removed from the dataset. We then calculated diversity metrics in QIIME2 including four metrics each of bacterial community alpha diversity (number of observed ASVs, Shannon diversity, Faith's phylogenetic diversity and Pielou's evenness) and beta diversity (Jaccard, Bray‐Curtis and unweighted and weighted UniFrac distances). These metrics allow us to compare the impacts of preservation on a variety of microbiome attributes (e.g., richness vs. evenness, presence/absence vs. relative abundance of microbes and phylogenetic vs. non‐phylogenetic metrics) (Lim and Bordenstein 2020).

2.4. Statistical Analyses

2.4.1. Comparison of DNA Extraction Methods

We compared the effect of extraction method, sample condition and their interaction on sample DNA concentration and total number of sequences returned per sample using ANOVAs in R. We used the package ‘emmeans’ to make pairwise comparisons between extraction methods and sample conditions within extraction methods using Tukey's HSD test. Regardless of DNA concentration and average read counts across groups, samples that return very few reads are unlikely to be usable for analyses, and are likely to be removed during rarefaction prior to diversity analyses, a commonly used practice in the microbiome field (Weiss et al. 2017). To inform which extraction method yielded the most samples with sufficient read counts deemed usable for analysis, we classified each sample as either having > 100 or < 100 high‐quality sequences returned and performed a Fisher's exact test to identify differences in the proportion of samples in these classifications by extraction method. We chose 100 sequences as our threshold because studies investigating the effects of sequencing depth on microbiome datasets conclude that this value is the minimum threshold above which biological conclusions do not meaningfully change (Caporaso et al. 2012; Kuczynski et al. 2010).

Next, we compared the microbial community composition of samples extracted using different methods. Extracting DNA from fresh gut content samples with the QIAGEN QIAamp PowerFecal DNA Kit is considered the field standard (Lim et al. 2018), and thus we considered the community composition of our freshly collected samples extracted with the PW method to represent the ‘best’ approximation each species' microbiome. We compared the community composition of preserved samples extracted with the other methods to these communities to determine which extraction method produced a community most similar to this standard. Specifically, we obtained the pairwise Bray‐Curtis distances between samples and compared the average pairwise distances between fresh samples extracted with the PW method to fresh samples extracted with the FM or CL method and preserved samples extracted with the PW, FM or CL method within each of the four species using ANOVAs with Tukey's post hoc HSD test.

2.4.2. Microbiome Comparison Between Freshly Collected Specimens and Specimens Preserved in Fluid for 1 Year

We filtered our ASV table to include only freshly collected samples and those preserved for 1 year and rarefied the table to 478 sequences per sample, which was the number of sequences in the sample with the lowest read count. Read counts above 100 have been demonstrated to be sufficient to recover significant microbiome differences (Caporaso et al. 2012; Kuczynski et al. 2010). However, to ensure our conclusions would not change if we rarefied samples to a higher depth but retained less samples, we performed two additional analyses in QIIME2. First, using the ‘diversity alpha‐rarefaction’ command, we created an accumulation curve to explore the relationship between the number of observed ASVs and sample sequencing depth. We compared values at 10 depths with 10 iterations per depth up to a maximum of 10,000 reads. We found that while the number of observed ASVs had not yet plateaued at our rarefaction depth of 478, conclusions drawn regarding relationships between metadata groups (e.g., taxonomic classes) were the same at every depth (Figure S2), indicating the rarefaction value should not meaningfully change alpha diversity conclusions. To ensure our beta diversity analyses performed at a depth of 478 would be representative of those that could be performed at higher depths, we utilised a Procrustes analysis (Gower 1975) in QIIME2 with the ‘diversity procrustes‐analysis’ command to compare relationships between principal coordinate analyses (PCoA). Specifically, we compared PCoAs generated using 478 sequences per sample with those generated using 1000 or 10,000 sequences per sample. A depth of 1000 represents a read count where ASVs are still rapidly accumulating and 10,000 represents a depth where ASV counts have largely plateaued (Figure S2). These analyses were performed for each of the four beta diversity metrics and for each comparison we calculated M ² and p‐values using 999 Monte Carlo permutations. p‐Values were significant (p < 0.05) for every comparison (Table S2), indicating our rarefaction depth should not change our conclusions about sample similarity for any diversity metric. Remaining analyses were thus performed at 478 reads per sample to retain all samples.

2.4.2.1. Beta Diversity Analyses

To understand the impacts of host taxonomic class (ray‐finned fish, amphibian, reptile and mammal), host species identity and sample condition (freshly collected, preserved 1 year) on the composition of the gut microbiome, we performed PERMANOVAs on the four distance matrices in the PRIMER7 software with the PERMANOVA+ add‐on (Anderson 2008). For each distance matrix, we conducted a main test to determine the significance of these factors and their interactions, with host species as a nested variable within host class, followed by a pairwise test to determine significant differences between pairwise comparisons of host class, host species and sample conditions within each host class or species. For all tests, 999 permutations were used, and p‐values were FDR corrected.

To test if samples were more similar to one another based on host species identity or sample condition, we compared average pairwise microbiome distance values across several groups for each species: freshly collected samples to all other freshly collected samples of the same species, samples preserved for 1 year to all other samples preserved for 1 year of the same species, freshly collected samples to preserved samples of the same species, preserved samples to all other preserved samples of different species within the same host class, preserved samples to all other preserved samples of different species in different host classes. We conducted this test using each of the four distance matrices for each species and assessed significance between groups using a pairwise permutation test in R which corrected p‐values for multiple comparisons using the FDR method.

To determine if relationships of microbiome similarity across host species were retained after sample preservation, we performed two analyses. First, we made separate ASV tables for each sample condition and merged reads within each table by species using the ‘feature‐table group’ command in QIIME2. We then created new distance matrices for each beta diversity metric at the species level for each sample condition. In R, we used the ‘vegan’ package to perform a Mantel test between distance matrices of fresh and preserved sample types for each of the four metrics. Next, we used UPGMA clustering via the ‘diversity beta‐rarefaction’ command in QIIME2 to create a dendrogram of species' microbiome similarity within each sample condition. For each sample condition, this analysis was repeated four times using each distance matrix. We compared the congruence of dendrograms by calculating cophenetic correlation coefficients using the R package ‘dendextend’. For each beta diversity metric, we calculated this coefficient comparing trees from freshly collected samples to those created with samples preserved for 1 year.

We next tested for patterns of phylosymbiosis in our dataset, in which more closely related species have more similar microbial communities (Brooks et al. 2016), and if these relationships were impacted by fluid‐preservation. We performed a generalised linear model in R with the average pairwise microbiome distance between species pairs as the dependent variable, and the phylogenetic distance between species pairs, sample condition and the interaction of these two factors as the independent variables. We obtained average pairwise microbiome distances between species from Bray‐Curtis and Jaccard distance matrices using the ‘feature‐table group’ command in QIIME2. We obtained the phylogenetic distance between species pairs using TimeTree (Kumar et al. 2022).

We also analysed for potential effects of geographic site on microbiome composition using samples from the green frog ( L. clamitans ) and red‐backed salamander ( P. cinereus ). We performed PERMANOVAs on each of the four microbiome distance matrices, including samples from only these two species, using the ‘adonis2’ function in the ‘vegan’ package in R, with species, site, sample condition and interactions of these variables as factors. Because we identified significant interactions between site and species, we next performed PERMANOVAs using the same method for each species separately to determine the effects of site, sample condition and their interaction within each species. For each model, we used 999 permutations, FDR corrected p‐values, and set the ‘by’ argument to ‘terms’.

2.4.2.2. Alpha Diversity Analyses

We compared differences in the four alpha diversity metrics across host classes, species and sample conditions. The alpha diversity metrics were not initially normally distributed, so we normalised the data using transformations. For right‐skewed variables (number of observed ASVs and Faith's phylogenetic diversity) we used a cube root transformation, and for left‐skewed variables (Shannon diversity and Pielou's evenness) we subtracted values from a constant (10 for Shannon diversity and 1 for Pielou's evenness) and used a cube root transformation on the resulting values (Mangiafico 2016). Next, we performed nested ANOVAs in the ‘car’ package in R with each alpha diversity metric as the dependent variable, and species nested within class, sample condition and their interactions as the independent variables. For variables with significant effects of host class or species and significant interactions between sample condition and host class or species, we used the package ‘emmeans’ to make pairwise comparisons across groups using Tukey's HSD tests. Specifically, we identified which host classes, host species and sample conditions within each class or species were significantly different from one another.

2.4.2.3. Bacterial Taxonomic Relative Abundance

Within each species, we identified specific microbial taxa at the phylum and genus level with relative abundances that were significantly different between sample conditions. To identify differentially abundant taxa, we used ANCOM‐II (Mandal et al. 2015) version 2.1 in R, a method found to produce consistent results across studies and produce results that agree best with the intersection of other methods (Nearing et al. 2022). Within models, we set the ‘zero_cut’ argument to 0.90 which excludes taxa from analyses that have a proportion of zeroes higher than this fraction and is suggested to obtain the most robust results (Nearing et al. 2022). We set the ‘lib_cut’ argument to 0 to retain all samples in the analysis and corrected p‐values using the BH FDR method. We identified taxa as differentially abundant using the 0.8 level detection threshold.

2.4.3. Comparisons With Historical Fluid‐Preserved Amphibian Specimens

We performed a second set of analyses with only amphibian species, including freshly collected samples, samples from specimens preserved for 1 year, and historical samples. We created an ASV table with these focal samples rarefied to 478 sequences per sample, which retained all samples in the analysis and calculated the same alpha and beta diversity metrics as previously described.

To assess the effects of host species, sample condition and their interaction on microbiome community composition, we performed PERMANOVAs in R using the ‘adonis2’ function in the ‘vegan’ package with the same settings previously described. If significant effects of sample condition were shown, we then performed pairwise comparisons between sample conditions, using the ‘pairwise.perm.manova’ function in the ‘RVAideMemoire’ package with 999 permutations and FDR corrected p‐values. We also performed separate PERMANOVAs for each sample condition to test the effect of species within each condition.

We repeated Mantel tests, UPGMA clustering, alpha diversity and bacterial taxa relative abundance analyses in this dataset exactly as previously described, including an additional sample condition for each host species (historical). For relative abundance data, we compared historical relative abundances to the relative abundances of freshly collected samples only.

3. Results

3.1. Comparison of DNA Extraction Methods

The concentration of extracted DNA differed based on both the method of extraction and sample condition. Specifically, samples extracted with the FM method had significantly higher DNA concentrations than those extracted with either the CL or PW methods (Figure S3A, ANOVA p < 0.001) and among samples extracted with the FM method, samples from preserved specimens yielded higher DNA concentrations than freshly collected samples (Figure S3A, ANOVA p < 0.001). Extraction method did not significantly impact the total read count of samples after sequencing (Figure S3B, ANOVA p > 0.05). Freshly collected samples had consistently higher total read counts than samples from preserved specimens, and this trend was strongest in samples extracted using the PW method (Figure S3B, ANOVA p < 0.001). Extracting DNA with the FM method significantly reduced the proportion of samples that returned at least 100 sequences (Figure S3C, 100% of samples > 100 reads in CL and PW, and 89.2% in FM, Fisher's exact test p < 0.001), the minimum value for analysis that should not obscure results after rarefaction (Caporaso et al. 2012; Kuczynski et al. 2010).

When comparing the microbial community composition across extraction methods, we found that for the fish ( L. macrochirus ), amphibian ( A. americanus ) and mammal ( P. maniculatus ), extracting preserved samples with the FM or CL methods returned communities that were significantly more similar to freshly collected samples extracted with the PW method (our standard), compared to preserved samples extracted with the PW method (Figure S3D,E,G, ANOVA p < 0.05 for all). Additionally, fresh samples are amenable to extraction with FM and CL methods. Specifically, when comparing the community composition of fresh samples extracted with the PW method (standard) to other fresh samples extracted using the FM or CL methods, we observed more (in the mammal or amphibian) or equal (in the fish) community similarity as we did when comparing these ‘standard’ samples to preserved samples extracted with any method (ANOVA p < 0.05 for the amphibian and mammal). In the reptile species ( C. picta ), we did not observe any difference in microbial community similarity to fresh samples extracted with the PW kit across any of our sample condition and extraction method combinations (Figure S3F), indicating all methods returned similar communities in this species.

In summary, although extracting samples with the FM method yielded the highest concentration of DNA, this kit significantly reduced the number of samples returning at least 100 reads. The CL and FM methods performed equally as well in returning microbial communities with compositional similarity to the sample condition and methods we considered to be representative of a species' microbiome. Thus, to maximise reads per sample while maintaining reliable compositional results, we extracted our full sample set with the CL method.

3.2. Microbiome Comparison Between Freshly Collected Specimens and Specimens Preserved in Fluid for 1 Year

3.2.1. Beta Diversity Analyses

Host taxonomic class, host species and sample condition (i.e., fresh or preserved 1 year) were all significant factors shaping the observed gut microbiome (Figure 1, Figure S4 and Tables S3–S6, PERMANOVA p = 0.001 for all tests). We consistently observed that host taxonomy exerted a stronger influence than sample condition on microbial community composition. Based on test statistics, host taxonomic class had the greatest effect size, and, in three of four metrics, host species identity had a larger effect size than sample condition (Tables S3–S6). There were also significant interactions between sample condition and host class and species. Pairwise comparisons between sample conditions within host classes and species revealed effects of fluid preservation on the microbial community composition of all amphibian and mammal species (Tables S3–S6, PERMANOVA p < 0.05 for all). Fish microbiomes were similarly sensitive to the effects of preservation, though the bluegill ( L. macrochirus ) microbiome was not significantly affected by preservation in terms of Jaccard distance between samples (Table S3, PERMANOVA p > 0.05). Reptile microbiomes were the least sensitive to preservation effects. Specifically, garter snake ( Thamnophis sirtalis ) microbiomes were not impacted by preservation based on any metric, and painted turtle ( C. picta ) microbiomes were only impacted by preservation based on unweighted UniFrac distance between samples (Table S5, PERMANOVA p < 0.01).

FIGURE 1.

Gut microbial community composition structures across host taxonomy and sample condition. Principal coordinate analyses plots of (A) Jaccard and (B) Bray‐Curtis distances across the gut microbiome inventories of 11 animal species (colours) in four vertebrate classes obtained from samples from freshly collected animals (circles) or animals that were fluid‐preserved for 1 year (squares). Percentages represent the proportion of variation explained by each axis. Abbreviations of species names are listed in Table 1.

Samples from preserved specimens tended to be more similar to fresh samples from their own species than samples from preserved specimens of other species, and this trend was detected both within and outside the focal species' taxonomic class (Figure 2 and Figure S5). Relationships of species' microbiome similarity were also generally retained after sample preservation. Specifically, there was a significant correlation between distance matrices at the species level for freshly collected samples and those from samples preserved for 1 year for each beta diversity metric (Figure S6, Mantel test p < 0.05 for all). Beta diversity metrics that are naïve to phylogenetic relationships of microbial taxa performed better at maintaining relationships of species' microbiome similarity across sample conditions than metrics that do incorporate phylogenetic relationships (Figure S6, Cophenetic correlation [CC] between dendrograms using: Jaccard distance CC = 0.94, Bray‐Curtis distance CC = 0.79, Unweighted UniFrac distance CC = 0.63 and Weighted UniFrac CC = 0.63).

FIGURE 2.

Microbiome inventories from preserved specimens tend to be more similar to microbiomes of fresh specimens of the same species than preserved specimens of other species. Average pairwise microbiome distance between several group pairings in (A) bluegill (LM), (B) dusky salamanders (DO), (C) garter snakes (TS) and (D) deer mice (PM). For each species, the following comparisons of average pairwise Jaccard distance are shown: Fresh samples to all other fresh samples within the focal species (‘Fresh only’), samples from preserved specimens to all other samples from preserved specimens within the focal species (‘Pres. only’), fresh samples to samples from preserved specimens within the focal species (‘Pres. – Fresh’), samples from preserved specimens of the focal species to samples from preserved specimens from all other species within the same host taxonomic class (‘Pres. – Pres. same class’), and samples from preserved specimens of the focal species to samples of preserved specimens from all species in different host taxonomic classes (‘Pres. – Pres. diff. class’). Within a panel, groups labelled with different letters have average pairwise distances that are statistically significantly different from one another based on permutation tests. Because mice (PM) are the only species from the mammalian class, there is no comparison to samples from the same taxonomic class.

We found significant positive associations between microbiome distance and host phylogenetic distance (Figure 3, GLM, p < 0.001 for each metric), and no interactive effects with sample condition (GLM, p > 0.05), indicating closely related host species tend to have more similar microbial community composition than more distantly related species, and that this relationship is detectable after sample preservation.

FIGURE 3.

Positive associations between microbiome distance and host phylogenetic distance are retained during preservation. Host phylogenetic distance calculated as the divergence time of two species and (A) microbiome dissimilarity calculated as pairwise Jaccard distance or (B) microbiome dissimilarity calculated as Bray‐Curtis distance. The analyses were conducted with freshly collected samples (red) and samples from preserved specimens (yellow). Trendlines from linear models are shown with shading indicating 95% confidence intervals.

In the green frog ( L. clamitans ) and red‐backed salamander ( P. cinereus ), which were collected evenly from two sites, there was no effect of capture site on microbiome composition in a combined model (Figure S7A and Tables S7–S10, PERMANOVA p > 0.05 for all). However, due to interactive effects between host species and capture site (Tables S7–S10), we performed separate models for each host species. The salamander showed a significant effect of capture site on microbial community composition in three of four metrics (Figure S7B, Tables S7–S10), and capture site was significant in shaping the frog's microbiome in one of four metrics (Figure S7C and Tables S7–S10). In both species, there were no interactive effects between capture site and sample condition (Tables S7–S10), suggesting that preservation did not obscure geographic effects.

3.2.2. Alpha Diversity Analyses

Microbiome alpha diversity differed significantly across host classes and species, and the effects of fluid preservation on alpha diversity varied across host taxonomy (Figure 4 and Figure S8–S10). Microbiome inventories from fluid‐preserved specimens tended to have fewer unique ASVs than those from freshly collected specimens (Figure 4, ANOVA p < 0.01). There was a significant interactive effect between sample condition and host class (ANOVA p < 0.01) but not host species for this metric (ANOVA p > 0.05). Pairwise tests for differential effects of sample condition within host classes demonstrated significantly fewer ASVs in fluid‐preserved amphibians and mammals (Figure 4, Tukey's HSD p < 0.01), but not in fishes and reptiles (Figure 4, Tukey's HSD p > 0.05). For Shannon diversity, Faith's phylogenetic diversity and Pielou's evenness, there were no significant main effects of sample condition (ANOVA p > 0.05). However, there were significant interactive effects between sample condition and host class (ANOVA p < 0.01) but not species (ANOVA p > 0.05). Pairwise tests found significantly lower Shannon diversity and Pielou's evenness in preserved mammalian specimens, while fluid‐preserved fish yielded inventories with greater Pielou's evenness compared to freshly collected samples (Figures S8 and S10; Tukey's HSD p < 0.05 for all). Faith's phylogenetic diversity was significantly higher in preserved reptile specimens compared to freshly collected specimens, but this effect was not observed in the three other classes (Figure S9, Tukey's HSD p < 0.01).

FIGURE 4.

Alpha diversity of the observed gut microbiome varies across sample host taxonomy and sample condition. (A) Host taxonomy at the level of class and (B) host taxonomy at the level of species. At the top of each panel, letters represent host classes or species that are significantly different from one another. In panel B, differences between species were tested only within their own taxonomic class and these letters cannot be compared to other classes. Letters A and B were used for amphibians, letters C and D were used for fish, letter E was used for reptiles, and no tests were conducted within mammals because there is only one species. In both panels, significant differences between sample conditions (fresh, preserved 1 year, or historical) within each class or species are shown with symbols. In each comparison, an asterisk (*) over the preserved (yellow) or historical (brown) group indicates that these groups were significantly different from the fresh (red) group. A plus‐sign (+) over the historical group indicates that this group was significantly different from the preserved group. Abbreviations of species names are listed in Table 1.

3.2.3. Bacterial Taxonomic Relative Abundance

The number of bacterial phyla that differed in relative abundance between fresh and preserved specimens ranged from zero in painted turtles ( C. picta ) to seven in bluegill ( L. macrochirus ), and for bacterial genera this value ranged from 32 in painted turtles ( C. picta ) to 119 in dusky salamanders (Desmognathus ocrophaeus) (Table 2 and Table S11). There were no bacterial phyla or genera that differed consistently between sample conditions across all species (Table 2, Table S11 and Figure 5). Further, preservation seemed to have variable effects on the relative abundances of particular bacterial taxa across vertebrate hosts. For example, some bacterial phyla (e.g., Gemmatimonadetea and Dependentiae) differed consistently across the hosts in which they were impacted such that they were present in low relative abundance in freshly collected samples but absent in inventories from preserved specimens (Table S11). Other phyla (e.g., Deferribacteres and Fusobacteria) had relative abundances that were higher in fresh specimens compared to preserved specimens in some host species but exhibited the opposite relationship in other host species (Table S11 and Figure 5). Our findings suggest that the specific impacts of specimen preservation on the detected relative abundance of bacterial taxa vary across host taxonomy, and therefore, are difficult to generalise across groups.

TABLE 2.

Overview of results from bacterial taxonomic relative abundance analysis. Panel A displays the number of bacterial phyla or genera that differed between freshly collected samples (F) and those from specimens preserved for 1 year (P) for each host species. For amphibians, two columns are shown depicting the comparison between fresh and preserved samples and fresh samples compared to those from historical museum specimens (H). Panel B depicts the number of bacterial phyla and genera that differed between sample conditions in multiple host species. For the comparison between fresh and preserved samples, the number of possible host species that a bacterial taxon can differ in has a maximum of 11, while for fresh specimens compared to historical specimens the maximum number of host species is six (amphibian analysis only).

(A) Q: How many phyla and genera had relative abundances that differed between sample conditions in each host species? F = Fresh P = Preserved H = Historical	Host class	Host species	No. diff. phyla			No. diff. genera
	Ray‐ finned fishes	L. macrochirus	7			91
		P. flavescens	2			82
			F v. P		F v. H	F v. P		F v. H
	Amphibians	A.americanus	2		9	67		91
		L. clamitans	3		9	82		73
		D. ochrophaeus	5		5	119		159
		P. cinereus	3		5	41		99
		P. glutinosus	4		6	51		79
		N. viridescens	3		3	65		113
	Reptiles	C. picta	0			32
		T. sirtalis	4			70
	Mammal	P. maniculatus	4			37
(B) Q: How many phyla or genera differed between sample types in two or more host species? sp. = host species	No. individual phyla diff. in X species	2 sp	3 sp	4 sp	5 sp	6 sp	7 sp	8 sp	9 sp	10sp	11 sp
	F v. P (all 11 species)	4	3	1	0	0	0	0	0	0	0
	F v. H (6 amphibians)	5	6	0	1	0	NA	NA	NA	NA	NA
	No. individual genera diff. in X species	2 sp	3 sp	4 sp	5 sp	6 sp	7 sp	8 sp	9 sp	10sp	11 sp
	F v. P (all 11 species)	106	35	23	9	3	1	0	0	0	0
	F v. H (6 amphibians)	108	35	28	10	8	NA	NA	NA	NA	NA

FIGURE 5.

Recovered relative abundances of bacterial phyla in the gut exhibit highly variable responses to preservation across host taxa. The relative abundance of any phyla present at greater than 1% mean relative abundance in any group is shown. All other phyla are grouped together as ‘other’. On the x‐axis, F = freshly collected samples, P = samples from specimens preserved for 1 year, and H = samples from historical specimens. Shaded boxes horizontally aligned with phyla names indicate phyla that showed significant differences in relative abundance between sample conditions in each species based on ANCOM II analysis. Arrows in each shaded box represent the direction of change in abundance of the phyla in the preserved or historical group compared to the mean relative abundance of the fresh group. Amphibian species have two columns of boxes in which the first column represents a comparison between fresh samples and samples from specimens preserved for 1 year, and the second column represents comparisons between fresh samples and historical samples. All other species have one column of boxes representing the comparisons between fresh samples and samples from specimens preserved for 1 year. Abbreviations of species names are listed in Table 1.

3.3. Comparisons With Historical Fluid‐Preserved Amphibian Specimens

3.3.1. Beta Diversity Analyses

Host species and sample condition (i.e., fresh, preserved for 1 year or historical) were again significant factors shaping the composition of the gut microbiome, and all three sample conditions were significantly different from one another (Figure 6, Figure S11 and Tables S12–S15, PERMANOVA p = 0.001 for all tests). Here, sample condition exerted a stronger influence on the observed gut microbiome than host species based on effect sizes (Tables S12–S15); however, within each condition, host species always remained a significant factor shaping the microbiome (Tables S12–S15, PERMANOVA p < 0.05 for all tests).

FIGURE 6.

The gut microbiome of six amphibian species is shaped by host species and three sample conditions. Principal coordinate analysis plots show dissimilarity between gut microbiome samples of six amphibian species from freshly collected samples, samples from specimens preserved for 1 year, and samples from historical museum specimens using (A) Jaccard distance and (B) Bray‐Curtis distance. Percentages on all axes represent the proportion of variation explained by each axis. Abbreviations of species names are listed in Table 1.

Relationships between species' in microbiome composition were largely retained after both preservation for 1 year or in a historical museum collection. Specifically, there was a significant correlation between distance matrices at the species level for freshly collected samples and those from samples preserved for 1 year for each beta diversity metric except Unweighted UniFrac (Figure S12). Comparing matrices between fresh samples and those from specimens in the historical collection, there were significant correlations for Jaccard and Weighted UniFrac distance, but not Bray‐Curtis and Unweighted UniFrac distance (Figure S12). When comparing relationships between species' microbiomes using UPGMA clustering, relationships were generally similar comparing fresh species to those from preserved specimens (Figure S12, Cophenetic correlation [CC] between dendrograms using: Jaccard distance CC = 0.99, Bray‐Curtis distance CC = 0.94, Unweighted UniFrac distance CC = 0.90 and Weighted UniFrac CC = 0.46). These relationships were less congruent when comparing fresh samples to those from historical museum specimens (Figure S12, Cophenetic correlation [CC] between dendrograms using: Jaccard distance CC = 0.73, Bray‐Curtis distance CC = −0.35, Unweighted UniFrac distance CC = −0.19 and Weighted UniFrac CC = 0.70).

3.3.2. Alpha Diversity Analyses

There were significant effects of sample condition on microbiome alpha diversity in all metrics (ANOVA p < 0.001 for all tests). Samples from historical specimens exhibited significantly lower diversity than freshly collected samples or samples from specimens preserved for 1 year (Figure 4 and Figures S8–S10, Tukey's HSD p < 0.01 for all tests). There was no significant interaction between host species and sample condition.

3.3.3. Bacterial Taxonomic Relative Abundances

A greater number of bacterial taxa differed in relative abundance between fresh and historical sample conditions than we had previously identified when comparing fresh samples to those from specimens preserved for 1 year (Figure 5, Table 2 and Table S11). Additionally, when comparing fresh and historical specimens, there was more consistency across host species in how the relative abundance of bacterial taxa was impacted. There was no single bacterial phylum that differed in relative abundance between sample conditions in all host species; however, there were several phyla that were consistently in higher abundance in fresh samples across all hosts in which differences were detected (Deferribacteres, Dependentiae, Gemmatimonadetes, Planctomycetes and Verrucomicrobia) and others that consistently showed the opposite pattern (Deinococcus‐Thermus and Patescibacteria) (Figure 5 and Table S11). There were eight bacterial genera that differed in relative abundance between fresh and historical samples in all six amphibian species (Table 2 and Table S11). Two of these genera were present in freshly collected samples of all species but not detected in historical samples (uncultured Gemmataceae and Candidatus Xiphinematobacter) (Table S11), while six had higher relative abundances in historical samples of all species (Caldalkalibacillus, Escherichia‐Shigella, Halomonas, Oceanobacillus, Staphylococcus and Xanothomonadaceae bacterium Sty a‐1) (Table S11).

4. Discussion

4.1. Relative Effects of Fluid Preservation and Host Factors on the Observed Gut Microbiome Across Species

While other studies have begun to explore how fluid‐preservation may impact the microbiome (Heindler et al. 2018; Greiman et al. 2020; Bodawatta et al. 2020; Chalifour et al. 2022), ours is the first to compare the effects across multiple host taxa in individuals collected from the same population at the same time. Thus, our study represents the most comprehensive assessment of the impacts of specimen fluid‐preservation on the vertebrate gut microbiome to date and helps resolve the relative impacts of preservation vs. host identity in shaping these gut microbial community inventories.

In general, we found that fixing specimens in formalin, followed by storage in ethanol for 1 year, altered the observed gut microbiome compared to samples collected from fresh individuals during the same collection event based on alpha diversity, beta diversity and relative abundance of bacterial taxa. It is perhaps unsurprising that fluid‐preserving specimens altered observed gut microbial communities, as it has been well documented that sample storage method can impact gut microbiome composition and diversity (Vlčková et al. 2012; Hale et al. 2015, 2016), and preservation of whole animals can result in a reduction of gut microbial diversity and loss of specific bacterial taxa from the community (Bodawatta et al. 2020; Chalifour et al. 2022). The observed changes in the microbiome post‐preservation may be due to formalin‐induced degradation of DNA (Miething et al. 2006), which could occur differentially across bacterial taxa (Krehenwinkel et al. 2018), as well as the impact of fixatives on PCR amplification (Giannella et al. 1997), potential contaminants introduced during handling and storage of specimens (Llamas et al. 2017), or a combination of such factors.

While there were general impacts of fluid‐preservation on the microbiome, not all host species' microbiomes were equally sensitive to this process. Mammal and amphibian microbiomes were the most sensitive to fluid‐preservation, with all species in these classes exhibiting reduced diversity and altered composition of the microbiome. The composition of fish microbial communities were generally sensitive to the effects of preservation, but their communities did not tend to exhibit decreases in diversity after preservation. Reptile microbiomes were the least sensitive to preservation. Their microbiomes did not decrease in diversity after preservation, and snake microbiomes ( T. sirtalis ) did not differ in composition between groups, while turtle microbiomes ( C. picta ) differed in composition based on only one beta diversity metric. The identity of specific bacterial phyla and genera with relative abundances that were impacted by preservation, and the direction of their difference, also differed across host species. Several factors could explain the differences in sensitivity of host species' microbiomes to the effects of fluid preservation, including pre‐existing characteristics of microbial communities such as diversity or composition. For example, the reduced sensitivity of the reptile microbiome to preservation may be explained by the fact that external effects on microbial communities can disproportionately remove rare taxa (Wegner et al. 2013) which are significant contributors to overall diversity (Cao et al. 2021), and reptiles had the least diverse microbial communities compared to other host taxa. Additionally, samples from preserved specimens tended to have reduced sequencing depths compared to those from fresh animals. While we did rarefy all samples—which is currently the best method to control for uneven sequencing effort (Schloss 2024)—community composition can still be sensitive to the effects of sequencing depth even after rarefaction (Ramakodi 2021). Hosts with naturally less diverse microbial communities, such as reptiles, may be less sensitive to these effects of sequencing depth because more of the variation in their communities can be captured at low depth.

Physical characteristics of hosts and bacteria may also be important factors in the susceptibility of microbial communities to fluid‐preservation induced changes. Natural rates of DNA degradation can differ across bacterial taxa (Krehenwinkel et al. 2018). It is also possible that structural differences in bacterial cell walls (i.e., gram‐positive vs. gram‐negative), or elements such as capsules or spores, could affect our ability to detect bacteria after prolonged storage (Philippot et al. 2021), similar to the enhanced detectability of gram‐positive bacteria after freezing (Bahl et al. 2012). For example, several Gram‐negative bacterial phyla, Chlamydiae, Planctomycetes, Spirochetes and Verrucomicrobia, each had relative abundances that were reduced after preservation in multiple host species. However, the effects of preservation on microbial taxa were not always consistent across host species. Additionally, attributes of the host animal may impact the susceptibility of the microbial communities to preservation‐induced changes. For example, reptile skin is resistant to permeability by fluids while amphibian skin is highly permeable (Shoemaker and Nagy 1977). Although we injected animal body cavities with formalin, it is possible different amounts of storage preservative (ethanol) could permeate the gut across species if intestinal permeability is similar to skin permeability. While storing microbiome samples in high concentrations of ethanol (≥ 95%) can minimise changes to the community, lower concentrations like those we used to preserve whole animals (70%) can lead to larger changes in community structure (Song et al. 2016). Differential internal exposure to ethanol across species could alter the rates of bacterial degradation.

Regardless of preservation‐induced changes, when comparing samples from freshly collected animals to those preserved for 1 year, host taxonomy was the strongest factor shaping the observed microbial communities. Host taxonomic class and species exhibited larger impacts on microbial community composition for almost all beta diversity metrics. This effect was supported by comparisons of average pairwise distances across groups of samples and the high correlation between distance matrices of different sample conditions that compare microbial communities across host species. These findings are consistent with other studies that show that host identity, at both the individual and species level, tends to be a stronger factor shaping the gut microbiome than sample storage (Song et al. 2016; Carroll et al. 2012; Blekhman et al. 2016).

In addition to showing differences in composition of the microbiome across species in general, we observed a phylosymbiotic pattern in which more closely related species had more similar microbial communities than more distantly related species, regardless of sample condition (Brooks et al. 2016). Phylosymbiosis has been observed across a variety of host animals from corals to mammals (Brooks et al. 2016; Pollock et al. 2018; Kohl, Dearing, et al. 2018; Kohl, Varner, et al. 2018), and can be attributed to various factors, including co‐diversification of hosts and microbes, selection of microbes by host physiological traits, or dispersal of microbes from similar habitats (Kohl 2020). In our dataset, the three Plethodontid salamanders ( P. cinereus , P. glutinosus and D. ochrophaeus ) formed visible clusters on PCoA plots with the two species from the same genus harbouring microbiomes that were the most similar to, yet still distinct from, one another. Additionally, there were significant positive relationships between host phylogenetic distance and microbiome distance between species, which were consistent for both freshly collected samples and those from preserved specimens. These results suggest that museum specimens may offer promise for studying phylosymbiosis in closely related clades of species despite preservation‐induced changes to the microbial community.

We were also able to uncover potential trends in microbiome composition based on host habitat. For example, there was close clustering of samples from fishes ( L. macrochirus and Perca flavescens ), newts ( Notophthalmus viridescens ) and turtles ( C. picta ) on PCoA plots. All of these species are aquatic, and differences in host‐associated microbial community composition between aquatic and terrestrial habitats have been shown within (Fontaine et al. 2021) and across host species (Kueneman et al. 2014). Furthermore, in the salamander ( P. cinereus ) and frog ( L. clamitans ) species in which samples were collected across two geographic sites, the impact of site on the microbiome for both species was similar for freshly collected samples and those from 1‐year‐old preserved specimens. Geographic site is known to influence microbiome composition within host species; however, species identity tends to be more important than location in shaping the microbiome across host species (Kueneman et al. 2014). Thus, even when differences in microbiomes across geographic sites are relatively small, those differences may still be identified a year after fixation and fluid preservation.

4.2. Differences in the Gut Microbiome Between Historical and Contemporary Specimens

In amphibians, we found larger differences between the microbial communities of freshly collected samples and samples from historical museum specimens than we had between fresh samples and those from specimens preserved for 1 year. Specifically, we found that the effect of sample condition was stronger than the effect of host species on microbial community composition and correlations were not always observed between distance matrices of different sample conditions depicting microbial community similarity across host species. Additionally, historical specimens yielded the lowest sample alpha diversity, and more bacterial phyla and genera showed relative abundances that differed from fresh samples.

Several hypotheses may explain why microbial communities of historical specimens differ more from fresh specimens than those preserved more recently. First, the historical specimens were collected from the wild over three decades before fresh specimens of the other groups were collected. Long‐term monitoring studies show gut microbial communities of populations can shift substantially through time (Degnan et al. 2012). Thus, differences observed in the gut microbiomes of historical specimens could represent real changes in live‐animal microbial communities that have occurred over time. Alternatively, these specimens have been stored in preservative for significantly longer time periods than the other group, which may have allowed for the effects of preservation to compound. When gut microbial communities of freshly collected birds were compared to birds preserved in alcohol for 2 weeks or 2 months, more bacterial taxa were lost at the 2‐month time point than the 2‐week time point (Bodawatta et al. 2020), suggesting a temporal component to these storage‐induced changes. To better understand the kinetics of microbiome changes over storage time, it will be necessary to collect specimens from the same population at the same time and repeatedly inventory communities over long (multi‐year) storage time periods. Specimens from this study are still preserved in the laboratory and could be used for such purposes.

4.3. Implications for Incorporating Museum‐Preserved Specimens Into Gut Microbiome Research

Host‐associated microbiome research has surged in recent years, but studies focused on the ecological and evolutionary implications of microbial communities for wild hosts have lagged behind human and model‐animal laboratory studies (Pascoe et al. 2017; Woodhams et al. 2020; Colston and Jackson 2016; Hird 2017). Sampling from the broad diversity of hosts contained in natural history museum collections could increase representation of understudied taxa, and some studies have already begun utilising these valuable collections for microbiome studies (Hird et al. 2014). Our study sheds light on some of the research questions museum specimens may be most helpful to address. For example, the study of phylosymbiosis may benefit from the use of museum collections. While phylosymbiosis has been demonstrated across a variety of host organisms (Kohl 2020), its ability to predict microbiome patterns in some groups, like lizards (Ren et al. 2016) and birds (Trevelline et al. 2020), has been debated. Large‐scale comparative studies including many closely related hosts are likely needed to fully understand these relationships (Williams and Fontaine 2024). Here, we demonstrate that species‐specific differences in microbial community composition and the ability to detect phylosymbiotic patterns are robust to the effects of specimen preservation, indicating that museum specimens could be used to increase the feasibility of sample collection for such studies. Additionally, our work demonstrates the ability to detect the effects of even small‐scale (~150 miles) geographic distance on microbiome composition within species. Large‐scale, often global, studies of the microbiome across geography reveal important patterns about the factors that drive microbial community composition (Kueneman et al. 2019; Rosas‐Plaza et al. 2022). Use of museum specimens for these studies could allow researchers to sample animals from around the world without leaving the museum, reducing time and travel.

Our work will increase the utility of natural history museum collections as unique resources that can inform global change biology about how anthropogenic impacts, such as land use and climate change, have impacted wildlife through time (Meineke et al. 2019). Examples where similar goals have been accomplished include the utilisation of herbarium specimens to identify changes in plant phenology due to climate change (Willis et al. 2017), and the use of dental calculus from museum bear specimens to quantify antimicrobial resistance within oral microbiota to inform understanding of the impacts of human antibiotic use (Brealey et al. 2021). Similarly, museum specimens may be used to track changes in microbial communities over time. However, our results do suggest that the impacts of preservation on microbial communities could compound over time, making similar studies in gut microbiomes more challenging. Specifically, the gut microbiomes of historical museum specimens were more different from fresh specimens than samples preserved for only 1 year. At this time, we are unable to determine if these differences are due to storage alone or real differences in microbial communities because the animal specimen collection date differed between fresh and historical samples. To understand the feasibility of temporal studies of historical microbiomes, we need more data from specimens preserved at the same time but maintained in collections for different amounts of time before sampling. If the effects of preservation on microbiota do compound consistently over time, there may still be ways to circumvent these changes. For example, models using large amounts of metadata may be able to tease apart consistent changes in the microbiome over time due to storage vs. those induced by a relevant variable of interest.

Regardless of the specific research question, our work provides important insights that can be used to shape study methodology. We observed significant changes in recovered microbial communities between freshly collected samples and those from preserved specimens. Thus, when designing studies, we recommend only comparing samples from preserved specimens to others preserved in a similar manner, rather than comparing to freshly collected samples. This advice applies generally to microbiome studies, in which consistency in sample treatment, rather than the specific treatments applied, appears to be the most important factor in obtaining useful data (Goodrich et al. 2014; Nearing et al. 2021). Furthermore, if researchers are preserving specimens themselves, consideration should be given to the time between animal collection, euthanasia and preservation, as factors such as handling stress (Suhr et al. 2023), fasting (Kohl, Amaya, et al. 2014) and time after death (Heimesaat et al. 2012) can influence microbial communities.

Next, when considering sample processing—formalin‐fixation is known to complicate the extraction of high‐quality DNA (Miething et al. 2006). Thus, we compared three commercially available DNA extraction kits to identify the optimal method for samples from preserved specimens. The CL method performed best, returning the greatest number of samples with read counts sufficient for analysis, and microbial community inventories most similar to our standard. This method combined two procedures: (1) DNA extraction following a modified protocol using QIAGEN's ‘QIAamp DNA FFPE Tissue Kit’ and (2) DNA cleaning with QIAGEN's ‘DNeasy PowerClean CleanUp Kit’. The FFPE Tissue Kit, typically used for extracting DNA from medical biopsy samples, contains steps designed to remove formalin‐induced DNA crosslinking, including lysis by proteinase K and an extended high‐heat incubation. The CleanUp Kit is used to clean and purify DNA, which is a step typically included in kits designed for faecal samples to remove common inhibitors (Li et al. 2003). Thus, our method likely produced optimal results because it combines steps to reduce DNA extraction inhibition from both formalin‐fixation and faecal material. Although more complex methods of DNA extraction could be designed specifically for microbiome samples from museum specimens, our approach using readily available kits balances approachability, affordability and consistency across studies.

Additionally, our results highlight important microbiome analysis considerations. We consistently observed that using Jaccard distance as a measure of microbiome beta diversity resulted in the most congruency in sample composition between freshly collected samples, those preserved for 1 year, and those from historical specimens (e.g., see Figures S5, S6 and S12). Metrics such as Jaccard which consider only presence or absence of particular microbes (Jaccard 1901) may perform better than other metrics which consider both presence and absence of microbes and their relative abundances, such as Bray‐Curtis (Bray and Curtis 1957), due to the impacts of preservation on the relative abundance of many bacterial taxa. These presence/absence metrics can also be more robust to PCR amplification biases (Hu et al. 2021).

Lastly, additional improvements to our techniques could be made to further increase the feasibility of using museum specimens for microbiome research. We profiled microbial communities using bacterial 16S rRNA amplicon sequencing, as it is the most widely used method in wildlife microbiome studies (Combrink et al. 2023), and our goal was to be relevant to as many researchers as possible. However, shotgun metagenomic sequencing, which is commonly used to profile microbial communities in ancient DNA samples (Brealey et al. 2021; Ottoni et al. 2021), may be more suitable to reduce amplification biases that can occur with fragmented DNA and skew taxonomic profiles (Ziesemer et al. 2015). Metagenomics also allows for deeper sequencing of microbial communities, and consequently, the taxonomic identification of more bacterial species (Laudadio et al. 2018). This depth may be particularly helpful for understanding temporal changes in microbial communities (Shade and Gilbert 2015). An additional consideration is the increased use of control samples. Best practices for the use of controls in microbiome research have evolved over time and methods have varied across studies (Hornung et al. 2019). Generally, negative controls are most important in low‐biomass samples; however, in gut microbiome studies, these ‘contaminants’ are often actual inhabitants of the gut (Glassing et al. 2016). Therefore, we utilised an algorithm to identify likely contaminants to be removed from our dataset (Davis et al. 2018) rather than removing all sequences identified in controls. Our methods could be improved by increasing the number of negative control samples and utilising additional control sample types such as positive controls and sample collection controls. However, our results are likely robust to the effects of contamination due to the strong and consistent differences we observed in the microbiome across groups.

5. Conclusion

We have presented an in‐depth analysis of how fluid‐preservation of whole animals using formalin‐fixation and storage in ethanol influences the observed gut microbiome of species from multiple vertebrate classes. Although preservation changes the microbial community, we show that signatures of the original samples' community composition, including those of host taxonomic identity and sampling location, are robust to storage. Although the use of preserved museum specimens may not be appropriate for every research project and question, we feel they can be used wisely to make important contributions to the microbiome field, which is currently lacking an ecological and evolutionary perspective across a wide variety of hosts.

Ethics Statement

All animal research was conducted in accordance with approved IACUC protocols from the University of Pittsburgh (protocol 18041349) and animals were collected with permission from the Pennsylvania Fish and Boat Commission (permit 2018‐01‐0109) and Pennsylvania Game Commission (permit 43,113).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1.

Data S2.

Data S3.

Data Availability Statement

Raw sequences and metadata used for all analyses are available from GenBank under BioProject ID PRJNA1027848 (Fontaine et al. 2025).