ACN: Acetonitrile

ALP: Alkaline phosphatase

AST: Aspartate aminotransferase

ATP: Adenosine triphosphate

AUC: Area under curve

BCA: Bicinchoninic acid assay

CARF: Central Analytical Facility

CCKR: Cholecystokinin receptors

CK: Creatine Kinase

DIA: Data independent acquisition

DTT: Dithiothreitol

EBI: European Bioinformatics Institute

EGF: Epidermal growth factor

EMBL: European Molecular Biology Laboratory

FA: Formic acid

FGF: Fibroblast growth factor

GGT: Gamma-glutamyl transferase

GO: Gene ontology

IAM: Iodoacetamide

ID: Identification

LC: Liquid chromatography

MS/MS: tandem mass spectrometry

MS: Mass spectrometry

nanoLC-nanoESI-MS/MS: nano liquid chromatography nano electrospray ionisation tandem mass spectrometry

PANTHER: Protein ANalysis THrough Evolutionary Relationships

PCA: Principal component analysis

PRIDE: Proteomics identifications

PSL: Peptide spectral library

QUT: Queensland University of Technology

RT: Room temperature

SWATH: Sequential acquisition of all theoretical fragrant mass spectra

TEMED: Tetramethylethylenediamine

TFA: Trifluoroacetic acid

TOF: Time-of-flight

TP: Total protein

UQ: The University of Queensland

Until now, the use of sequential window acquisition of all theoretical fragment mass spectra (SWATH-MS) analysis ¹ to complement traditional shotgun MS-based proteomics ² on samples from species of veterinary importance has been uncommon. Only a limited number of published reports have applied this approach on animal experimental subjects, for example in the quantification of hepatic proteins of chicken exposed to heat stress, ³ analysis of seminal plasma protein of pigs, ⁴ identification of proteins involved in nutritional stress in goats, ⁵ quantification of protein alterations in plasma of acutely endotoxaemic sheep, ⁶ and detecting proteins in peptide spectral libraries of bovids. ⁷ Otherwise, much of the work in this area has been centred upon in-vitro studies on cell lysates ^{8 – 11} and body fluids obtained from mice. ¹² Despite its promising nature in protein analysis, SWATH-MS analysis remains largely untested on actual clinical samples from veterinary patients.

In this report, SWATH-MS was used to analyse serum samples obtained from sheep suffering from a range of naturally-acquired illnesses and compared with samples from healthy sheep, by interrogating a novel peptide spectral library (PSL) constructed from the ovine circulating acellular proteome. ^{6 , 13} The sick sheep serum samples were of actual clinical cases submitted to The University of Queensland (UQ) (Gatton, Australia). Serum samples from twenty healthy sheep were obtained from a commercial source (Serum Australis Pty Ltd.) and normalised to represent an analytical normal proteome background. Tryptic peptides were purified by StageTip technique, ^{14 , 15} prior to nano-liquid chromatography nano-electrospray ionisation tandem mass spectrometry (nanoLC-nanoESI-MS/MS) in a cyclic data-independent acquisition (DIA) mode, using a generic SWATH-MS acquisition method (SWATH™, SCIEX). The acquired data were processed in PeakView® software with SWATH™ Acquisition MicroApp 2.0 (SCIEX) pipeline to generate protein lists for downstream gene ontology annotation and pathway analysis of identified proteins.

The use of SWATH-MS analysis enabled protein profiles of individual sick sheep to be differentiated as compared to the normalised healthy sheep serum aliquots. This approach detected some established clinical biochemical analytes and possibly other candidate disease markers. These observations suggest that it is potentially feasible to use SWATH-MS in conjunction with the novel PSL to determine relative protein alterations in clinical samples from veterinary patients in future.

The experiments for generating this report utilised SWATH-MS technology to primarily analyse proteins in ex-diagnostic serum samples of sheep suffering from a range of different ailments in comparison with serum from healthy sheep to detect any differences in their protein profiles. The strategy adopted involved testing and validating the potential application of the nascent peptide spectral library built from circulating acellular proteome of sheep. ⁶ The overall intention here was to target protein analytes in significant laboratory findings of the submitted sheep serum samples based on the accompanying clinical data from UQ’s School of Veterinary Science laboratory. The relevant analytes for proteomics studies for this report are found in Additional file 1. ⁴⁸ Among the analytes of interest were gamma-glutamyl transferase (GGT), total protein (TP), serum albumin (Alb), aspartate aminotransferase (AST), alkaline phosphatase (ALP), creatine kinase (CK), fibrinogen and some other analytes that were listed in the reference ranges, for example globulin and serum amyloid. Another important analyte – at least in ruminants – that was not listed in the laboratory panel is haptoglobin. ²⁵ These analytes may have appeared as isoforms of their known protein moieties. In some cases however, the genes coding for these proteins were not well defined, possibly due to the consideration that the sheep genome still remains to be fully annotated. ^{26 , 27} Nevertheless, it is feasible and practical to use draft genome sequences for the interpretation of MS/MS proteomics data without the need for tedious genome annotation. ²⁸ This does not negate the point that a well-annotated genome is necessary in order to name genes and proteins unambiguously. ^{29 – 32}

Differentially expressed serum proteins between healthy and sick sheep were either known or considered to be potential biomarker candidates for diagnosing and monitoring disease states, and this could also provide leads for further research endeavours. There are two important layers to determining the significance of the differences between sick and healthy sheep serum proteomes observed in this work: a) the degree of confidence in the observed difference from an analytical standpoint and, b) the biological plausibility that such differences truly correlate with a disease state. The discussion that follows looks at these preceding two aspects.

A problem of working with the present approach of identifying proteins is that when protein isoforms are not explicitly named and identified, it introduces ambiguity that makes it challenging to identify even well-known proteins as in the instances presented in this report. For example, in the set of samples utilised in this study, Hb was identified as A0A0F6YFJ0 (Beta-C globin). There were also two protein IDs representing GGT which comprised of W5QCX2 (Transglutaminase 1) and W5PB04 (Transglutaminase 3). Serum albumin was identified as W5PWE9 which is yet to be characterised for sheep in UniProtKB. It should be noted that one isoform of albumin, P14639 (Albumin), that was not identified in this set of experiments, has already been characterised in UniProtKB. The ID feature for Aspartate aminotransferase (AST or GOT) was W5PS88. There was no protein ID directly matching ALP in this dataset. However, there were three other phosphatases including W5P195 (Dual specificity phosphatase 14), W5P3B0 (Phosphatidylinositol-3,4,5-trisphosphate 5-phosphatase) and W5QE45 (Serine/threonine-protein phosphatase) that were identified. Meanwhile, CK was represented by two IDs: W5PJ69 (Creatine kinase) and W5NQ67 (Creatine kinase). Fibrinogen was represented by four proteins including W5Q5H8 (Fibrinogen alpha chain), W5NQ46 (Fibrinogen beta chain), W5Q5A6 (Fibrinogen gamma chain) and W5PH03 (Fibrinogen like 1). Serum amyloid was represented by W5PJG0 (Serum amyloid A protein), P42819 (Serum amyloid A protein) and W5PJR0 (Serum amyloid A protein). Haptoglobin was identified as W5P0Q4 (Haptoglobin).

In veterinary clinical pathology, the determination of the quantity of protein in plasma or serum has always been traditionally based on the amount of albumin and globulin fractions, ^{33 , 34} and yet there is more to the two broad protein groups when it comes to deep proteomic analysis. For example, in the present set of protein identifications, the globulin fraction was represented by at least 12 protein families or groups. The globulin protein fraction comprised of W5P812 (Protein AMBP), W5PSC1 (Ig-like domain-containing protein), W5Q0R1 (Sex hormone binding globulin), W5NQW4 (Alpha-1-macroglobulin-like), W5PGT9 (Ig mu chain C region), W5QI15 (Ig-like domain-containing protein), W5PSK4 (Ig-like domain-containing protein), A4ZVY9 (Beta-2-microglobulin) W5NSA6 (Alpha-2-macroglobulin), W5PPQ8 (Joining chain of multimeric IgA and IgM), P49920 (Corticosteroid-binding globulin) and the various VH region immunoglobulins such as A2P2G1 and A2P2I1. Being able to identify more proteins in the different fractions of serum proteins gives this approach an edge over traditional clinicopathological protein assays.

The experiments in this report utilised samples from single sheep case reports as proof-of-concept that SWATH-MS technology can be applied to identify vast numbers of proteins and their alterations in clinical samples of veterinary patients. The comparison of clinical cases (sick sheep serum), versus normalised serum from a large number of healthy sheep is a classical approach typical of biomarker discovery or detection studies. ³⁵ With this foundation in place, it is possible to establish a standard for routine proteomic analysis of serum samples submitted to laboratories in future, once a specific baseline serum proteome of sheep has been optimised using far much larger numbers of samples. This method also has the potential to be used for identifying different protein species that show differences between samples. The pooling of samples from healthy sheep was conducted on the premise that the pooled sample would provide a representative proteome of normal sheep serum. Similarly, the rationale of using a pooled sample from sick sheep followed the same premise that that this could provide a representative picture of all the proteins present in the sick sheep, particularly proteins that might only be abundant in specific disease states that could not be detectable in serum samples from healthy individuals. The inherent downside of pooling samples is that this strategy disregards the biological variation of the individuals the samples are drawn from, ³⁶ by capturing for example, the ‘average’ proteome profile across a population. And also in the present results, 26 protein IDs were not detected when samples from sick sheep were pooled, as opposed to them being analysed individually ( Figure 2B). The reason behind this observation is not immediately clear, but batch effects during sample analysis could be advanced as a contributing factor, ³⁷ or possibly unknown effects of pooling samples for downstream analysis. It also follows that the use of pooled serum, at least from normal individuals to act as a control, is an accepted scientific practice for normalising samples that has also been widely used in some human studies. ^{35 , 36}

The evaluation of chromatographic features of peptides from samples – TICs in this case – was a practical, inexpensive and straightforward visual way of comparing ion intensities between small numbers of samples as previously reported elsewhere. ³⁸ Here, only two sample TICs were loaded per displayed panel on PeakView®: one from a sick sheep versus a normalised serum of normal sheep (Additional file 2 ⁴⁸ ). The TICs represented a measure of relative abundance of the peptides detected. ³⁹ It follows therefore that during interpretation of the TICs, it should be considered that they represented summed signals from the sample as well as background noise. ⁴⁰ The use of TICs in addition to SWATH-MS data extraction has a high depth of analysis whilst factoring in the wide dynamic range in analytes, is a useful strategy especially when considerable differences are expected between samples ³⁸ as in the present study. There were distinct differences between TIC profiles of sick sheep serum samples compared to serum profiles from normal sheep; this difference was more marked in the sample from the sheep with scabby mouth lesions (Additional file 2 ⁴⁸ ). In the majority of the analysed samples, the TIC intensities of sick sheep serum samples were generally higher than those of normal sheep serum, except for the ill-thrift and the ill lamb cases. A possible explanation for this observation could be malnutrition/starvation in the chronically ill sheep, since decreased food intake depletes protein generally and this was even more crucial in the neonatal sheep (ill lamb) that was likely still reliant on colostrum. ^{41 – 43} Proteins in circulation are either derived from synthesis or from degradation, but considering that the small intestine is the most important organ for synthesis and absorption of proteins in ruminants, ⁴⁴ the detected proteins in starving sheep are therefore most likely to have been derived from tissue degradation. As a result, there was probably a comparatively very little protein reserve to elicit the higher protein intensities compared to the other relatively acute disease case samples. As for the analysis of the pooled sample of sick sheep (SCi511), the TIC profile was generally above that of pooled normal serum up to approximately 38 min when it switched below the pooled normal serum TIC, before peaking again at the hydrophobic end of the chromatogram. An explanation for this observation remains to be established and is open to further interpretation.

The differences in protein species between individual serum samples from sick sheep could have been due to the different aetiological and pathological factors of the presenting condition in the different sheep that could have stimulated the production of different proteins or their alterations. Without follow-up clinical data on the cases or a definitive diagnosis, it is not possible to fully determine and to explain these differences. Also, clinical cases were from diverse populations which may have contributed to different serum proteomic profiles as proteomes are known to be dynamic. ⁴⁵

Different numbers of proteins were identified from the different samples and to some extent, different protein species. The sample from the sheep with scabby foot and mouth lesions had the highest number of protein IDs ( Figure 2A). A substantial number of protein IDs were common between sick and the normalised sample from healthy sheep as represented by the ‘pooled healthy control and sick overlap’ legend item in Figure 2A. The protein yield of 210 IDs from the pooled sick sheep sample (SCi511) was much lower than expected as compared to the IDs from all other individual samples considered together. It is not immediately evident as to why this was the case. This relatively low numbers of protein IDs could potentially have been due to sub-optimal tryptic digestion of this particular pooled sample, loss of peptides during sample preparation or probably due to unknown factors that influence the pooling of samples.

There were differences in protein intensities between the samples as evident in the ten most abundant proteins in each analytical case ( Figure 3). The comparison of protein intensities of the pooled serum from healthy sheep (SCi512) versus pooled serum from sick sheep (SCi511) revealed considerable differences for some proteins, with each of the points on the graph representing a protein ( Figure 4A). Fewer proteins (210 IDs) were identified in sick sheep sample pool, as compared to 335 IDs in healthy sheep serum pool. It should have also been possible to have a relative quantitation of the proteins centred on their intensities benchmarked on the calibration of the intensities from healthy sheep serum based on the principle of area under curve of the TICs ^{46 , 47} (Additional file 2 ⁴⁸ ). Using this approach, could have demonstrated the feasibility of determining what proteins were either upregulated or downregulated in the clinical serum samples from sick sheep.

The GO term classification provided an overall picture of the identified proteins in both sick and healthy sheep serum ( Figure 5). An important observation from the protein pathway analysis is that it provided 11 clearer protein pathways that peaked in sick sheep serum ( Figure 6). The roles of some of these protein pathways in sheep remain to be determined, but they have been studied considerably in humans and other model species such as mice. It is not unreasonable therefore to suggest that with homology, there is a translational potential in the observations made in this study from sick sheep that can be learnt from.

There were limitations in this experimental study. The first one was that there was only one biological sample for each clinical case study — which is not unusual for case studies. Since each clinical case was unique, to be able to adequately test current observations, three or more technical replicates (or MS injections for that matter) out of each biological sample would have improved the meaningfulness of the data score plot. The second limitation was that it was not possible to re-analyse samples drawn from the same sheep after they recovered (i.e. to act as their own controls), and/or at different stages on the disease to determine if chronicity had an effect on protein species or their alteration – considering that sick sheep samples were opportunistically utilised from laboratory archives at UQ.

The use of SWATH-MS analysis successfully identified proteins and enabled protein profiles from serum samples of sick and healthy sheep to be distinguished. It was possible to detect some established clinical biochemical analytes, for example aspartate aminotransferase, creatinine kinase, haemoglobin, serum amyloid A, fibrinogen, haptoglogin, members of gamma-glutamyl transferase group, serum albumin and various proteins of the globulin fraction. Protein pathway analysis provided useful information on the expression profiles of protein groups between sick and healthy sheep by giving them a biological meaning. There was a downside to this approach, for example the detection of alkaline phosphatase – an important analyte that was not obviously identifiable in this dataset. Some detected proteins of sheep such haemoglobin, haptoglobin and some isoforms of serum amyloid did not have distinct gene names as compared to homologous counterparts in related species. Nevertheless, this proof-of-concept study has demonstrated that with a known benchmark or standard samples from healthy individuals, it is feasible to determine relative protein alterations in clinical samples. This approach could have a place in veterinary clinical pathology in future.

The author originated the concept of the study, conducted the experiments with assistance from staff of The University of Queensland and Queensland University of Technology, and wrote the final manuscript.