Whole blood from 24 animal species were individually trypsin-digested and analyzed by tandem mass spectrometry. Reference libraries were built for each species using SpectraST (version 4.0) which was adapted to build spectral libraries from unidentified spectra (Fig. 1). Spectra that are highly similar and therefore likely derived from identical peptides were detected by similarity clustering and merged into higher-quality consensus spectra (Supplementary Methods). Unlike conventional spectral libraries, similarity clustering is necessary for consensus spectrum building when peptide identifications are not known. Collectively, the spectral libraries contained 9,045 consensus spectra, 1,039 (11.5%) of which were detected in multiple species. The distribution and relative abundance of spectra among vertebrate species are visually represented in Fig. 2. Spectra from the blood of M. musculus, the only species for which a genome sequence database is available and peptide identifications could be made reliably, were identified by conventional sequence searching to reveal the underlying protein distribution of the spectral libraries (Fig. 3).

The performance of the clustering algorithm was validated by comparing a spectral library built by similarity clustering without knowledge of peptide identifications, as described above, and a spectral library built from sequence-search results of the same data. The latter is possible for species with comprehensive and high-quality sequence databases such as M. musculus. We implemented this strategy on a larger dataset containing 40 runs of a SDS-PAGE fractionated protein digest sample of M. musculus blood. In this dataset, 20,156 of the 38,010 acquired tandem mass spectra survived the spectrum filter and were similarity-clustered, as described above, into 3,584 clusters. At the same time, traditional sequence searching (Supplementary Methods) identified 10,324 spectra from the same dataset to a total of 1,566 distinct identifications (44% of 3,584). Therefore, less than half of all clusters were identified by sequence searching, likely due to incomplete search space (e.g. modifications not considered) and insufficient discriminating power of sequence searching. Of the clusters with at least one originating replicate identified, 1,492 entries (95%) are “pure” clusters that has only one sequence identification; 29 (2%) entries are “mixed” ones that have two or more different identifications; 35 (2%) entries are “split” clusters, for which the same identification was also found in another cluster; and 10 entries (1%) are both “mixed” and “split” (Fig. 4). This result suggests that the clustering algorithm is largely capable of merging spectra that originate from the same peptide, but not those that originate from different peptides.

The correct vertebrate species was identified as the source of the blood meal in all but one engorged larvae tested, demonstrating proof-of-principle evidence of the efficacy of this methodology (Table 1). The identification is made by finding the reference library with the highest similarity to the spectral dataset of the blood meal to be tested. A spectral dataset similarity score (SDSS) developed for this application provides a quantitative similarity measure between two spectral datasets, and hence the entire proteomes underlying the datasets. Using this algorithm, the blood in all five of the larval ticks that had fed to repletion on either of a two M. musculus mice was determined to be most similar to the M. musculus spectral reference library. These results had high bootstrap support (>99%), indicating that a large number and wide variety of spectra are available to correctly identify the blood in these ticks as M. musculus blood (Table 1). An example spectral dataset match is shown in Fig. 2. These data also demonstrate that the presence of vector proteins in the sample do not detract from the ability of this methodology to identify the source of the blood meal. Correct identifications were also achieved for all larval ticks that had fed on any of four Peromyscus leucopus mice or any of three Tamias striatus chipmunks, all with high bootstrap support (>99%) (Table 1). Interestingly, the blood in the larval ticks that had fed on two of the three Sciurus carolinensis squirrels was not identified as squirrel blood, as there was no squirrel blood spectral reference library, but instead matched most closely with the chipmunk spectral reference library (Table 1), the species that is evolutionarily most closely related to S. carolinensis among the reference libraries. Thus, this method can also identify an unknown sample to the closest-related species represented in the spectral libraries. This finding is corroborated by the evolutionary accuracy of the phylogenies built from the spectral reference libraries discussed below.

A recent publication reported the protein sequences of hemoglobin subunits of several vertebrate species by de novo sequencing and suggested that these sequences can be used to identify the source of blood meals in vectors ²⁷. To benchmark the methodology presented here, we also searched the proteomic data from our ticks against the hemoglobin sequences of M. musculus, P. leucopus and T. striatus by sequence database searching. Identification of the source of the blood meal from engorged larval ticks was also effective using methods based on detecting hemoglobins by sequence searching algorithms (Supplementary Table S1). Similar to the results by our method, larval ticks that had fed upon S. carolinensis could also be identified as ticks that had fed upon T. striatus if S. carolinensis hemoglobin sequences are not included in the database.

In 23 of the 26 cases, the spectral datasets derived from the remnants of blood in molted nymphal ticks were identified to the spectral reference libraries of the species on which the tick had fed as a larva (Table 2, Fig. 2). Of the ticks that fed on M. musculus, the M. musculus spectral library was the most similar for all of the nymphs that were held at 25°C for one month (6/6) and for three months (6/6) after blood meal, while the M. musculus spectral library was the most similar for the datasets from 4 of 6 nymphs held at 25°C for six months prior to analyses. The statistical confidence in M. musculus as the source of the previous bloodmeal was greater than 76% (bootstrap support) for each of the correctly classified nymphs. The two datasets that were not identified to the M. musculus were not misidentified to another organism; they simply had very low similarity scores with all spectral libraries, suggesting that little MS² data from the blood proteome was collected. More importantly, the spectral datasets from seven of eight nymphal ticks that had fed as larvae on natural reservoir hosts (P. leucopus (5/6) and T. striatus (2/2)) and subsequently held at 25°C for five months were successfully identified to the animal on which the ticks had previously fed, each with high bootstrap support (Table 2). These data strongly support the utility of the spectral matching methodology to identify the source of previous blood meals in nymphal ticks that are collected in nature.

Identification of the source of the blood meal from molted nymphal ticks was less effective when peptide identification relied upon de novo sequenced hemoglobin subunits (Supplementary Table S2). There were no instances in which the traditional sequence-based peptide identification methodology correctly identified the source of a blood meal that could not be identified using the spectral searching algorithm. Furthermore, the sequence-based method failed to identify the source of the blood meal for two ticks that fed on lab mouse, 4 out of 6 ticks that had fed on a white-footed mouse, and both ticks that had fed on a chipmunk. In these tick samples, no confident identification to any hemoglobin sequences were found. These results indicate that relying on a priori-selected target proteins (hemoglobin α and β) for identification was more prone to failure than the spectral matching algorithm described when the blood meal has undergone digestion over time. This is likely because the target proteins may have been degraded, whereas our proteome profiling strategy is still able to identify the source of the blood meal source based on spectra from other, non-target, proteins. Although hemoglobin is the most abundant protein in vertebrate blood, only a small fraction of the matching spectra utilized by our method can be identified to hemoglobin sequences (Fig. 2 and Fig. 3).

Although I. scapularis ticks feed only once per life stage, other arthropod vectors feed multiple times in rapid succession, which could hinder the identification of one or both sources of the blood in the vector. The ability to correctly identify the source of blood from vectors that had fed multiple times using the spectral matching methodology presented here was determined by experimentally mixing whole blood from different species at specified ratios. Whole blood from an African Lion (Panthera leo) was mixed with blood from either an Amur Tiger (Panthera tigris), an American Alligator (Alligator mississippiensis) or a Red Kangaroo (Macropus rufus) at lion:other ratios of 1:1, 1:5, or 1:10. The choice of species was determined by the evolutionary relatedness ranging from very related to distantly related. Proteins from mixed blood samples were analyzed by LC-MS² and the SDSS algorithm as described above. The spectral matching methodology can successfully identify the one of the two true species in all cases (with high bootstrap support), even if blood from both species is roughly at equal concentrations (Fig. 5). Additionally, when one source of blood is at much greater concentrations than the other (ratios of 1:5 and 1:10), the more concentrated source is identified as the top scoring hit in all cases, while the SDSS for the less abundant source varies. It is likely that blood derived from one of the multiple feedings will be predominant in a vector due to digestion between feedings or differences in the amount of blood acquired from each host. Under this scenario, this source of blood is likely to be identified by this method despite contamination with other blood sources. While the source of the blood making up the minor fraction can potentially be identified when it is at relatively high concentrations, a more sophisticated statistical method, as well as experimental samples that have fed on multiple host species, is needed to properly assess the validity of this method in identifying blood mixtures. Nevertheless, one of the true sources of the blood in a vector that has taken two blood meals in rapid succession can be identified with high confidence.

Bayesian phylogenies derived from the presence or absence of each spectrum in each spectral reference library suggest that the degree of similarity in spectral libraries derived from blood proteomes is correlated with evolutionary relatedness. Bayesian phylogenies constructed using evolutionary models designed for restriction digest data resulted in topologies that recapitulate the known relationships among the vertebrate species (Fig. 6). All nodes in the phylogeny constructed using the variable evolutionary model in MrBayes were supported by the data. The choice of evolutionary model for phylogenetic reconstruction did not affect the relationships among species although branch lengths differed and posterior support for nodes was generally lower (data not shown).

The level of biological organization (species, population, individual) that can be distinguished using the proposed methodology was investigated using spectral data derived from whole blood. To allow for separate reference and testing datasets, we used a single LC-MS² run to build the library and used a subsequent run from the same blood sample to search the library. Additionally, for two of the species (lion and tiger) we obtained whole blood from two individuals, only one of which was used to build the database. The SDSS values were calculated for each pair of samples in the test dataset and spectral libraries. The correct species most closely matched the test sample in all cases, generally with SDSS scores many fold greater than the next closest match (Table 3). Even the evolutionarily most closely-related pair of species in our dataset (lion and tiger) were accurately identified in all cases including samples from individuals that were not included in the database. Like all methods that depend on the evolutionary relatedness, we expect to reach a limit in the possible resolution. However, this resolution appears to be at a finer scale than between species as closely related as the lion and the tigers.

SpectraST (version 4.0), a tandem mass spectral library building and searching program originally developed for peptide identification ^39,49 was extended to incorporate unidentified spectra. Specifically, spectra imported into a library can be referenced by serial identification numbers without explicit peptide sequence identification. A consensus algorithm based solely on spectral similarity was applied to combine replicate MS² spectra acquired multiple times due to fragmentation of the same molecule into a consensus spectrum. This procedure is common in traditional spectral library construction and produces higher-quality spectra ^50–53. The numbers of spectra combined into each consensus spectrum, along with the source of the sample, were recorded. Spectra with limited information, which may cause rampant indiscriminate matching, were removed by spectral filtering. For algorithm details, please refer to Supplementary Methods. SpectraST is part of the Trans Proteomic Pipeline suite of software, an open-source and continuously maintained tool that supports all major computer platforms and mass spectrometers ⁵⁴.

Spectral reference libraries were built from LC-MS² analyses of whole blood from each of 24 animal species: Mus musculus (Lab Mouse), Peromyscus leucopus (White-footed Mouse), Tamias striatus (Eastern Chipmunk), Ovis aries (Domestic Sheep), Panthera leo (African Lion), Panthera tigris (Amur Tiger), Strix varia (Barred Owl), Anas platyrhynchos (Mallard), Sturnus vulgaris (Starling), Colobus guereza (Black and White Colobus), Chelodina longicollis (Snake-necked Turtle), Colubrinus loveridgei (Kenyan Sand Boa), Rousettus aegyptiacus (Egyptian Fruit Bat), Pteropus rodricensis (Rodrigues Fruit Bat), Tremarctos ornatus (Andean Bear), Equus africanus asinus (Poitou Donkey), Nanger dama mhorr (Mhorr’s Gazelle), Capra aegagrus hircus (Goat), Macropus rufus (Red Kangaroo), Corucia zebrata (Prehensile-tailed Skink), Alligator mississippiensis (American Alligator), Pituophis melanoleucus (Northern Pine Snake), Malaclemys terrapin (Northern diamondback terrapin), Rhynchocyon petersi (Giant Elephant Shrew). These species are representative of the major clades of land animals from which blood meals may be taken by hematophagous arthropods. Animal blood was generously donated by veterinarians at the Philadelphia Zoo or acquired from laboratory animals at the University of Pennsylvania following all animal care and use guidelines. Whole blood samples were trypsin-digested in- solution and the resulting peptide mixtures run on an analytical C₁₈ nanocapilary HPLC column (DIONEX, Acclaim PepMap100) into a LCQ-Deca XP Plus quadrupole ion trap mass spectrometer ^55–57. Duplicate runs were performed, and spectra acquired from each species were used to build spectral reference libraries as described above (Fig. 1) and in Supplementary Methods.

Blood fed larval ticks (I. scapularis) were acquired by allowing ticks to feed to repletion on a host animal and detach naturally¹⁶. The larval blood meal represents the first blood meal as I. scapularis ticks feed only once at each of the three life stages (larvae, nymph, and adult). We experimentally infested two M. musculus mice with larval ticks purchased from Oklahoma State University. Larvae that fed to repletion and naturally detached from mice were collected and either frozen immediately (n=5) or held at 25°C and allowed to molt to the nymphal stage. Nymphal ticks were frozen at one month (n=6), three months (n=6), or six months (n=6) post-molt for later processing and blood meal analysis. Engorged larvae that had fed upon wild P. leucopus, T. striatus, or S. carolinesis were acquired as previously described ¹⁵. Six engorged larvae from three different P. leucopus mice and two engorged larvae from two different T. straitus chipmunks were allowed to molt and held at 25°C for five months post-molt before being frozen for processing and blood meal analysis.

Whole ticks were manually crushed in 20 μl of PBS, insoluble materials removed by centrifugation at 25,000 × g at room temperature, and total protein concentrations of supernatants determined by the Bradford method (Bio-Rad). Protein extracts from engorged larvae, which are composed primarily of host blood, were run on a one dimensional NuPAGE 4–20% Bis-Tris gel (Invitrogen, Switzerland) and proteins visualized by Coomassie Blue Staining (Bio Rad). Each lane was cut into 20 equal-width bands, which were subjected to in-gel trypsin digestion and LC-MS^255–57. Total protein extracts from nymphal ticks were trypsin-digested in-solution prior to LC-MS² analysis without prior gel separation.

Data obtained from LC-MS² analyses of tick samples were searched against the spectral reference libraries by SpectraST (version 4.0) ⁴⁹, using a precursor m/z tolerance of 1.5 Da/e and search parameters identical to those used for traditional spectral searching (Fig. 1). The number of query spectra that matched each reference library spectrum, where a match is defined as dot product greater than 0.7 and dot bias less than 0.45, were counted. For each tick, the similarity between the query spectral dataset and the reference library for each species was quantified by the Spectral Dataset Similarity Score (SDSS), given by: SDSS(q,s)=∑i[mq(i)×rs(i)]∑i[mq(i)]2[rs(i)]2 where SDSS(q,s) is the measure of similarity between the query dataset q and the reference library s, i is an index that enumerates the spectra in the query dataset, m_q(i) is the number of spectra in the query dataset q that matches the i-th spectrum in the reference library s, and r_s(i) is the number of replicates that were combined into the i-th spectrum of the reference library s. With the normalization factor of the denominator, this dot product-like function ranges from 0 (no spectral matches between the query dataset and the reference library) to 1 (identical distribution of spectral counts between the query dataset and the reference library). The abundance of each spectrum is taken into account to improve the discriminatory power of this function. Several alternative dataset similarity functions were investigated and are presented in Supplementary Fig. S1.

The data were bootstrapped⁵⁸ to obtain a statistical measure of confidence in the identification of the source of the blood meal in each tick. For each query dataset, 5000 bootstrapped samples were generated by randomly selecting, with replacement, N MS² spectra from the dataset, where N is the total number of MS² spectra in the query dataset. Each bootstrapped sample was searched against each reference library and the most similar reference library (greatest SDSS score) was recorded. The confidence in the species identified as the source of the previous blood meal is the fraction of bootstrapped samples for which the most similar reference library is derived from the correct species. This strategy quantifies the uncertainty in species identification that can occur due to the stochastic selection of peptides for fragmentation in the mass spectrometer.

The spectral libraries built from the whole blood samples were used to infer phylogenetic relationships among species. The blood proteome data from each species was transformed into linear vectors representing the presence (1) or absence (0) of each library spectrum, resulting in data structures similar to restriction digest data. Bayesian phylogenies were constructed from these linear vectors in MrBayes ⁵⁹. Phylogenies were reconstructed in MrBayes using all five evolutionary models (all, noabsencesites, nopresencesites, variable, and informative) originally designed for restriction digest data.