Peptide Microarray Content Design

A total of 457 T. cruzi protein sequences annotated in the CL-Brener genome (version Jan. 2012) (23) were selected for inclusion in the microarray (according to the groups defined in Table I). Proteins in Group 1 were randomly sampled from the proteome. If a sampled protein was a putative ortholog of a previously picked protein (as defined by the OrthoMCL algorithm (24)), that protein was skipped. Hence, the final protein set for this group contained no homologous proteins. Group 2 contained T. cruzi proteins without previous serology evidence, selected based on shared features with known antigens such as evidence of expression, subcellular localization, presence of tandem repeats, disordered regions and B-cell epitope predictions, using an integrative method developed in our group (25). Homologous proteins were removed from this group, as described for Group 1. Group 3 is composed of selected members of the MASP (mucin-associated surface proteins) family. This is a multigene family composed of ∼1300 genes, with a high level of polymorphisms (26, 27), localized at the surface of infective forms of the parasite. This protein family has earlier been proposed to participate in host-parasite interactions (28). The six MASP subgroups (29) were represented in the group. Group 4 is composed of 68 proteins with previous evidence of seroreactivity in T. cruzi infected humans (these were obtained from the IEDB (5) or manually curated from the literature), more information in supplemental Table S3. Protein Group 5 is composed by 54 neo-proteins of random protein sequences. These random sequences have the same mono- and di-peptide distribution as found in the T. cruzi proteome. All unique 15-mer derived from protein sequences in these five groups were included in the array in a single copy, except for a random subset of peptides, which were replicated for technical variability assessment (6549 peptides in two or more copies, 992 in five or more copies). Peptide fields were distributed in 12 microarray sectors of ∼18,700 peptides each. Peptides from the same protein were synthesized at randomized positions within a single sector. The chip contains additional 2386 unique peptides corresponding to eight proteins not accounted within the protein groups defined above (and of which 2 proteins resulted seropositive), including members of the T. cruzi TASV protein family (30, 31) and 21 epitopes of high prevalence in healthy humans (not associated to Chagas disease, identified in the IEDB with a search for linear B-cell epitopes from any source with sero-prevalence > = 50%, assayed in at least 20 subjects). The sequences of these 532 proteins (457 T. cruziproteins, 54 neo-proteins and the additional 21 short peptidic epitopes of high prevalence in healthy humans) are available in supplemental Table S4.

Derivatization of Synthesis Slides

Synthesis slides (Nexterion P) for the photolithographic synthesis of peptide arrays were purchased from Schott AG, Germany and derivatized with a 2% w/v linear copolymer of N,N′-dimethylacrylamide and aminoethyl methacrylate (both from Sigma-Aldrich) mixed in a 20:1 w/w ratio before polymerization for 2 h at room temperature in freshly degassed 0.1 m sodium borate buffer, pH 8 containing 0.025% v/v TEMED and 0.1% w/v ammonium persulfate (18).

Peptide Arrays Synthesis

Each peptide field was composed of 2 × 2 mirrors controlled by the Digital Micro-mirror Device, and with a single-mirror border separation within fields. This setup allows high-throughput while maintaining a high field resolution. The spacer polypeptide DAPAD was added to all peptides at their C-terminal. The FLAG peptide DYKDDDDKK extended c-terminally with a linker peptide (GAPAGAP) was included in the microarray in 852 copies for peptide synthesis quality control and as corner (reference positions) markers. Synthesis of the arrays was performed as described previously (18).

Human Sera Samples

T. cruzi—infected human sera samples used in this study were obtained from the Laboratorio de Enfermedad de Chagas, Hospital de Niños “Dr. Ricardo Gutierrez” (Buenos Aires, Argentina). All procedures were approved by the institutional review board of this institution. Written informed consent was obtained from all individuals, and all samples were decoded and de-identified before they were provided for research purposes. Sera were collected from clotted blood obtained by venipuncture and analyzed for T. cruzi-specific antibodies by commercially available kits: enzyme-linked immunosorbent assay (ELISA) using total parasite homogenate (Wiener lab, Argentina) and indirect hemagglutination (Polychaco, Buenos Aires, Argentina). The negative panel was composed of 10 samples from healthy, non-infected individuals that gave negative results in the mentioned tests. Using these 10 samples, four different T. cruzi seronegative sera pools were prepared by taking five random samples (3 μl per serum) in each case, and labeled them A_neg, B_neg, C_neg and D_neg. In the case of Chagas-positive samples, we selected nine samples from patients that had no clinical symptoms, and were classified in the Chagas chronic indeterminate stage (19). Using these nine samples, four different T. cruzi seropositive sera pools were prepared by taking five random samples (3 μl per serum) in each case, and labeled them A, B, C, and D. We used the smallest pool size, which would cover most of the high prevalence antibody specificities. For this, we estimated that all epitopes of a prevalence of 50% or higher will be present at least in 1 individual serum with a probability higher than 95%. Briefly, for an epitope of prevalence x in the infected population, the probability of randomly taking n subjects without specific antibodies, assuming independence is (1 – x)ⁿ. Then, we calculated the minimum x, such that (1 – x)ⁿ < 0.05 (i.e. that the probability of not sampling an epitope of prevalence x% or higher in n subjects is less than 0.05). Performing this calculation, we find n = 5, i.e. (1 – 0.5)⁵ = 0. IgGs were purified from 15 μl sera pools using Melon Gel IgG spin purification kit (Thermo Scientific), following the manufacturer’s protocol. Purified IgG samples were checked in 12% SDS-PAGE gels. IgG concentration was estimated after staining by Coomassie Brilliant Blue, by comparison against a standard curve made by electrophoresing different quantities of purified bovine γ-globulin (IgG, 150 kDa, BioRad Laboratories, Hercules, CA).

Assays with Peptide Arrays and Data Acquisition

Microarray slides were incubated essentially as described in (16) with a few modifications to accommodate sequential incubations. Briefly, slides were incubated at room temperature overnight with 1 ml of purified pooled IgGs, diluted to 20 μg/ml in 0.15 m Tris/Acetate pH 8.0, 0.1% v/v Tween20. After washing with the incubation buffer, slides were incubated for 2 h with secondary antibody (Cy3 goat anti-human IgG, Abcam Cat. No. Ab97170) at 1 μg/ml. After a second washing step with incubation buffer, followed by air-drying of the slides in a nitrogen jet, the peptide array slides were scanned and recorded with an Innoscan 900, Cambridge, MA laser scanner (INNOPSYS, Carbonne, France) at 1 μm resolution, with an excitation wavelength of 532 nm. The images recorded with the laser scanner were analyzed using the PepArray analysis program (Schafer-N, Copenhagen Denmark). Auxiliary “marker” peptides with sequence DYKDDDKKGAPAGAP containing the FLAG epitope tag, were used for positioning of the grid and to quantify spots’ intensities. For all microarray experiments described, the same procedure was performed first with the negative sample (from healthy subjects), and then sequentially with the positive sample (from infected patients) in the same conditions. Therefore two readouts were obtained per slide: the negative sample data and the cumulative signal of negative and positive sample data. A total of 8 microarray chips were assayed, labeled A1 (Sample A_neg followed by Sample A, Replicate 1), A2 (Sample A_neg followed by Sample A, Replicate 2), A3 (Sample A_neg followed by Sample A, Replicate 3), B1 (Sample B_neg followed by Sample B, Replicate 1), B2 (Sample B_neg followed by Sample B, Replicate 2), C1 (Sample C_neg followed by Sample C, Replicate 1), C2 (Sample C_neg followed by Sample C, Replicate 2), and D1 (Sample D_neg followed by Sample D, Replicate 1). Raw intensity data sets for each chip and sample are available in public databases (See Data availability Section).

Peptide Mapping, Data Normalization, Signal Smoothing and Negative Subtraction

In previous successful applications of HD-peptide microarrays (16, 18, 32, 33), minimal data normalization was performed. However, unlike previous work, here we analyzed a complex antibody sample and compared relative intensities across different experiments. This demanded further normalization. A first normalization step was implemented, aimed at equalizing two sequential reads of a single chip (the negative and the cumulated negative+positive data) and decreasing sector (spatial) effects across the chips. For this, we defined an intensity baseline for each sector as its most frequent intensity value (mode). Each peptide was then centered by subtracting its associated baseline value. Once baseline signal was adjusted, intensity dispersion across sequential reads of a single chip was adjusted by dividing each value by the 90th percentile of the intensity values of the 24,000 random peptides in the same assay, therefore bringing the negative and negative+positive cumulated signals into a common scale. Peptide sequences and their fluorescence values were mapped back to the parental proteins sequences (only allowing a perfect sequence match of the 15 residues). The plots display intensity values for individual peptides along the protein sequence (i.e. from position 1, to protein length minus 14). We devised a simple smoothing procedure to improve the signal to noise ratio of individual measurements. Smoothing was performed on each protein profile, and consisted in a running median filter of window length 5, followed by a running mean filter of window length 7. Basically, the normalized signal of the peptide at position i in the parental protein sequence is first replaced by the median value calculated from the five peptides at positions i – 2 to i + 2. In a second run, after extreme values had been removed, the value of the peptide at position i in the parental protein sequence is now replaced by the mean value calculated from the seven peptides at positions i – 3 to i + 3. When reaching the C-terminal end, we padded the window length with zeroes to complete the scan. Epitope mapping performance (average AUC, as presented in Results), was incremented by ∼5% after signal smoothing. Smoothed negative sample signal was subtracted from smoothed cumulated negative+positive signal to obtain a smoothed Chagas specific antibody-binding signal, which was plotted as shown in Fig. 1. In many proteins, such as the Ribosomal protein L19 and antigen CA-2/B13 in Fig. 1, a repetitive signal pattern is produced by tandemly repeated (antigenic) amino acid motifs in those protein sequences. Because in each microarray experiment we obtain a single reactivity value for each distinct peptide, if the same peptide has to be mapped to different places in a protein sequence, its associated value would also be repeated, therefore producing such a regular signal pattern. In many cases some of these repeats are not strictly conserved, because of sequence polymorphisms (they are imperfect repeats, but still recognizable as repeat units). This creates an irregular or imperfect repetitive signal pattern for some proteins, usually at the borders of the tandemly repeated motifs. To compare intensity values from different chips and biological samples, an additional step was required to compensate for chip-to-chip intensity distribution disparities. In this case, quantile normalization (34) was performed to force the 3 signal distributions (pre-normalized and smoothed intensity values of negative samples, negative + positive samples and subtracted signals) to be the same across chips/samples. This was done independently for the negative, cumulative and subtracted data sets, using the R package preprocessCore (34). The three normalized data sets obtained for each chip after this transformation were used for all subsequent analysis and figures shown in this paper (distributions are shown in supplemental Fig. S3). In the subtracted signal distribution, 1 unit is approximately equivalent to the 95th percentile. The normalized and smoothed values from a total of 8 chips assayed (see above) are publicly available (refer to “Data Availability” section).

Epitope Mapping Performance Analysis

For this analysis, we compiled a validation data set of antigens with B-cell epitope mapping data, that met the following criteria: (1) had specific reactivity against chronic T. cruzi infected human sera; (2) the mapped epitopes had been reported in at least 2 independent publications; (3) were able to discriminate T. cruziseropositive and seronegative humans in ELISA format using synthetic peptides. From an initial data set of 24 proteins with mapped B-cell epitopes, only nine remained that met this criteria, after removing paralogs and redundant epitopes (peptide sequences and references are listed in supplemental Table S3, column F “High-confidence mapped B-cell Linear Epitopes”). In each antigen sequence, we mapped the exact location of the reported epitope onto the parental sequence. In the case of reported epitopes of length >5 residues, we split the original epitope into smaller “units” of length 5. This extension of the epitope definition was especially relevant in the case of repetitive antigens, where the presence of cross-reactive degenerate repeats (with minor amino acid substitutions) is frequent. Then, we classified as “positive” every 15-mer peptide containing any of these epitopes, and as ‘negative’ the remaining peptides. Receiver operating characteristic (ROC)¹ analyses were performed for each protein, to assess whether higher chip intensities were associated with epitope localizations. The area under the ROC curve (AUC) was measured for each antigen using the R package ROCR (35). Redundant peptides (identical 15-mers) were excluded from the analysis. The supplemental Fig. S1 shows ROC curves, AUC values, and sensitivities and specificities versus cut-off, for each of these 9 antigens, considering individual sera samples or their average. To assess the performance of combined data from different experiments, the normalized and negative sample-subtracted peptides values of different chip/assays were averaged and ROC analysis performed as described.

Sensitivity, Positive Predictive Value, and Cut-Off Definition

The cut-off used to define “positive” and “negative” peptide reactivity was set to maximize classification performance measures of sensitivity (true positives/(true positives + false negatives)) and positive predicted values (PPV = true positives/(true positives + false positives)). Positive predictive value is used in this paper in the context of binary classification of antigens and non-antigens, equivalently to classification precision; it should not be confused with classification of subjects based on infection status as frequently used in diagnosis. The validation set of 9 antigens with mapped linear B-cell epitopes of high confidence (Epitopes listed in supplemental Table S3, column F “High-confidence mapped B-cell Linear Epitopes”) was used as true positive (i.e. if classification is perfect, these proteins should contain highly reactive peptides and hence be classified as ‘positives’ proteins). The set of 54 neo-proteins (Group 5) was used as negative controls (i.e. if classification is perfect, these ‘proteins’ should not contain highly reactive peptides and hence be classified as negative proteins). For this task, the averaged normalized signal from the 8 chips was used. As observed in Fig. 4C, it is clear that in the cut-off range around 3 (from 2.6 to 3.4) all positive controls (validated epitopes; dark blue curve) are detected but no negative controls are (random sequences; black curve), i.e. both sensitivity/recall and PPV/precision are optimal in this interval. Therefore, a cut-off of 3 was chosen for the averaged signal. Conducting the same analysis on individual samples required a higher cut-off to maintain a PPV of 100% at the expense of sensitivity. In this case, a threshold of seven resulted in a PPV of 100% in all samples and a sensitivity of 0.83 ± 0.11, illustrating the previously described variation between different biological samples.

Definition of Antigenic Regions

For simplification, any continuous protein sequence range where all its peptide reactivities were above the cut-off (average signal > 3) was defined as an antigenic region. These are shown in supplemental Table S2. Here, we have also included a list of non-redundant antigenic regions (Sheet 2, “unique15mers”) by removing those regions that are identical, in terms of the sequence of their most reactive 15-mer peptides (Max pep sequence in supplemental Table S2), to other regions of higher ranks (according to seroreactivity). Finally, an additional filter was applied to remove related antigenic regions (Sheet 3, “unique7mers”). Here, an antigenic region was removed if its most reactive 15-mer had an internal 7-mer sub-sequence matching the most reactive 15-mer of an antigenic region of higher rank.

Analysis of Technical Reproducibility and Biological Variability

Variability between identical peptides was assessed from 992 different peptide sequences (a random sample) that were each replicated in at least 5 random positions in the microarray (and up to 25 replicates per peptide). Coefficients of variation associated to these peptides (raw values, ranging from 0 to 255) were calculated per chip, as the standard deviation of intensity divided by mean intensity. Mean intensities were corrected by adding one, to avoid divisions by zero. Because the length of most continuous B-cell epitopes is less than the length of the synthesized peptides, antibody-binding signals are expected to be shared among multiple overlapped peptides (i.e. among adjacent peptides covering a common antigenic determinant). The intensity correlation between all pairs of peptides derived from the same protein (but physically located at random microarray positions) was high for highly overlapped peptides (PCC of 0.86 ± 0.02 (S.D.), 14 residue overlap, i.e. with single residue shifts) and monotonically decreases for lower sequence overlaps (down to a PCC <0.25 for pairs of peptides with a six residue overlap). This analysis was performed with a subset of 105 reactive and nonrepetitive proteins (i.e. excluding those proteins where >20% of sequence is composed of internal tandem repeats, as detected by the TRUST repeat detection method (36)).

To assess technical variability across chips, PCC were calculated for all pairs of chips assayed with the same sample (three pairs of sample A replicates, one of sample B replicates and one of sample C replicates). In this case, the signal correlation for the same peptide across different chips, was calculated from normalized and smoothed intensity values of ∼199,000 unique peptides. Correlations of peptide intensities between chips assayed with different samples (biological replicates) were calculated in the same way. In addition, protein-wise inter-sample comparisons were performed. First, each protein was assigned a reactivity value equal to the normalized value of its most reactive peptide. Then, we calculated for each pair of samples, the fraction of proteins that passed the reactivity signal cut-off (a value of seven was previously defined for individual samples) in any of the two samples (S1 OR S2) or in the two samples simultaneously (S1 AND S2). The proportion of shared positive proteins between two samples was calculated as (S1 AND S2)/(S1 OR S2). This is illustrated in the scatterplot in Fig. 2C.

ELISA Validation of TSSA Antigen

The Gluthatione S-transferase (GST)-fusion protein bearing the central and antigenic region of CL Brener TSSA (GST-TSSA-CL) has been described (22). GST-TSSA-CL deletion variants spanning the 15-mer peptides covering the antigen in region of TSSA with offset of six residues (equivalent to peptide-to-peptide overlap of nine residues) were constructed by fill-in of partially complementary forward and reverse oligonucleotides (supplemental Table S5) containing BamHI and EcoRI sites on their 5′ ends, respectively. All of these constructs were treated with restriction enzymes and cloned using the BamHI and EcoRI sites of the pGEX1λT vector (GE Healthcare). Colony-PCR was used for the initial screening of the colonies, which were subsequently confirmed using Sanger-based sequencing. GST-fusion molecules were expressed, purified and quantified as previously described (22). Each GST-TSSA-CL deletion variant (plus a GST and GST-TSSA-CL negative and positive control, respectively) was tested against individual human serum (nine T. cruzi positive and three T. cruzi negative samples, as described in Human Sera Samples section) by ELISA as described elsewhere (37). The seroprevalence of each peptide was calculated as the proportion of individual sera samples with mean absorbance above the cut-off (n = 9). ELISA Δ O.D. (absorbance) in reactive samples, the quantity plotted in Fig. 3 – right axis was calculated for each peptide, as the mean O.D. (3 replicates average) detected in T. cruzi positive samples (only if measured OD is above ELISA cut-off), minus the average O.D. measured in negative control samples. Agreement of ELISA versus chip data was evaluated by Spearman’s rank correlation (r = 1, p = 0.017) of normalized and subtracted chip data versus average absorbance of the nine positive samples, after subtracting average absorbance of three negative sera samples + 5 standard deviations.

Data Availability

The array design and raw and processed array data have been deposited in the ArrayExpress database at the European Bioinformatics Institute under accession numbers A-MTAB-526 and E-MTAB-3008, respectively. The complete set of peptides was also submitted to the Immune Epitope Database Resource (5) under submission ID 1000642.