Correlation between centromere protein-F autoantibodies and cancer analyzed by enzyme-linked immunosorbent assay

Out of 175,000 samples submitted for routine ANA screening, 42 revealed the characteristic NSp-II CENP-F pattern in IIF. These 42 samples were from 28 different individuals that were listed in groups according to their diagnoses (Table 1): invasive cancer (14 individuals, 50%), benign tumor (five individuals, 18%), and no registered neoplasias (nine individuals, 32%), mainly representing various connective tissue diseases. A representative NSp-II fluorescence pattern is illustrated in Figure 2, applying PS 26, while Ctrl 5 was used as negative control.

The reactivity of pools of anti-CENP-F-positive sera and pools of control sera to overlapping 20-mer peptides covering the CENP-F amino acid sequence 1855-2189 was analyzed by ELISA. Figure 3 illustrates the reactivity of PS pool 1 (represented by PS 1, PS 2.1, PS 9.2, PS 11, PS 15, PS 17.2, PS 18, PS 26), Ctrl pool 1 (represented by Ctrl 1-8) and ANA pos pool 4 (containing eight ANA-positive sera) to the 33 CENP-F peptides, while Additional file1: Figure S1 illustrates the reactivity of all the pools screened. As seen, CENP-F antibody reactivity was found to multiple peptides. Antibody reactivity of PS pools was found to be focused around three regions A4-A6, A12-A16 and A21-A23. No distinct reactivity was found to these regions when analyzing reactivity of control pools (healthy donor pool and Ctrl pools), although all of the healthy donor pools showed weak reactivity to peptide A6. Similarly, three out of five ANA positive pools showed weak reactivity to peptides A9 and A17. Moreover, the region A22-A24 was found to be recognized by these pools as well. However, as these sera were CENP-F negative, these patterns are believed to be related to other nuclear proteins.

Based on the collective findings, the peptides: A1, A4, A6, A9, A12-A16, A18, A21-A23, A27 and A29 were chosen for further analysis (marked with # in Figure 3). All of the available 42 samples from the 28 anti-CENP-F-positive patients and 86 Ctrl sera (56 from ANA screening (Figure 4), 20 from ELISA testing and 10 from healthy controls) were then tested individually for reactivity to the selected CENP-F peptides by ELISA. As seen in Figure 4, antibody reactivity was found to be concentrated around peptides A6, A13, A15 and A22, recognized by 23, 9, 7 and 7 anti-CENPF-positive patient sera, respectively. Of the 44 sera from 28 patients showing the NSp-II IIF pattern 7 did not react with any of the peptides. Three of these (PS3-5 were from patients, where only one sample was available, three (PS2.2-2.4) were from a patient, where only the first of four consecutive sera reacted in ELISA, and one (PS17.1) was the first of 3 consecutive sera, where the second (PS17.2) and third (PS17.3) showed reactivity with one and two peptides, respectively. The Ctrl sera (ANA-negative samples (n=30), ANA-positive samples (centromere (n=6), homogenous (n=10) and speckled (n=10) patterns) (Figure 4), CCP antibody-positive sera (n=10), DNA antibody-positive sera (n=10), healthy blood donors (n=10)) only occasionally showed weak reactivity to the CENP-F peptides (Figure 4 and results not shown). Most importantly, the sera showing the typical centromere pattern, which resembles the NSp-II pattern, were negative in ELISA.

In order to look for patterns of variation in the data set and in order to determine whether a correlation could be established between the CENP-F antibody reactivites and individual patient diagnoses, both a multivariate and a univariate analysis was used to evaluate the P-values of the contrasts between the diagnostic groups. For this, three scenarios were set: 1) all groups were regarded as different from each other; 2) the no neoplasia and the benign tumor groups were regarded as one; and 3) the benign tumor and the invasive cancer groups were regarded as one.

A principal component analysis (PCA), which describes the variation in the data set, was performed on the fused matrix and the scores of the resulting 15 PCs were analyzed in a spreadsheet in which the P-values for the contrasts between the diagnostic groups were calculated for each PC. Table 2 shows the P-values for the first six PCs. Out of all PCs only PC#1, PC#2, and PC#6 were found to describe statistically significant variations. The single significant contrast between the control and the benign tumor group in PC#4 was found to be the result of a single outlier and was therefore not taken into consideration. Figures 5A-D illustrate 2-dimensional projections of these three PCs and their respective loading plots. The diagnostic groups are represented by colors and can more or less be distinguished spatially from each other. Looking at the score plot in Figure 5A, the group with invasive cancer dominates the negative end of PC#1, while the control group forms a tilted ellipse around the center, intersected by a more vertical clustering of the no neoplasia group. The benign tumor group could be forming a cluster in the overlap zone of the other groups, but due to the limited number of samples in this group, a clear pattern cannot be established. Observed perpendicular to this in Figure 5B, the overall pattern is more diffuse except for the control group that has accumulated in the bottom right corner. Based on the results illustrated in Figure 5, the P-values displayed in Table 2 was calculated, from which the following statements are clear: PC#1 correlates significantly with the contrasts posed by the invasive cancer group, and to a lesser extent with the contrast between the control and the no neoplasia groups. PC#2 correlates exclusively and significantly with the contrasts posed by the control group, although only weakly when the tumor group is involved, and PC#6 correlates even stronger with the contrasts posed by the control group. The benign tumor group is not clearly distinguished from the other groups, partly because it is only represented by five samples, and partly because it assumes a very intermediate profile.

Knowing what is described by the different PCs, the next step is to understand how the PCs themselves are described. To do this, the loading plots in Figure 5C and D were examined. The loading plots illustrate the projections of the 15 peptides onto the PCs in such a way that the greater the distance from zero a given peptide is placed along a PC, the greater “gravity” it imposes on the samples in the score plot. Therefore, the peptides in the negative end, of e.g. PC#1, are potentially responsible for pulling the samples towards this end, and since the negative end of PC#1 is dominated by invasive cancer samples, antibodies to these peptides are potentially correlated with invasive cancer. However, this cannot be unequivocally determined by an isolated view on the PCA, as a projection of a peptide can also pose a repulsive effect and thus inhabit a peripheral position in this capacity. To clarify this, the data obtained from the univariate analysis was considered.

P-values of the contrasts between the groups were calculated for each of the 15 peptides in the fused matrix using the Student’s t-test. The results are shown in Table 3. Beginning in the negative end of PC#1, antibodies to peptide A27 have the most negative loading score. At a first glance, it seems as if these antibodies pose a strong traction upon the invasive cancer group, but according to Table 2 they actually pose a strong repulsive action upon the no neoplasia group, while they do not distinguish between the other groups. In terms of antibody profile this means that the no neoplasia group has a significantly lower expression of antibodies against peptide A27 than the other groups. CENP-F antibodies to peptides A12 and A14 appear to have the same level of impact on PC#1. Their univariate profiles, however, are very different. Antibodies to peptide A12 do not specify much contrast, antibodies to peptide A14 pose a more hierarchical pattern, being most prevalent in the invasive cancer group, less so in the benign tumor and control groups and least in the no neoplasia group. The benign tumor group was not distinguished by antibodies to peptide A14. Closer to zero of PC#1 were CENP-F antibodies to peptide A9. According to Table 3 these antibodies are highly specific for the invasive cancer group, for which the expression was significantly elevated compared to the control and no neoplasia groups. Antibodies to peptide A16 were also more frequent in the invasive cancer group when compared to the no neoplasia group, but not when compared to the control group.

Looking at PC#2 and PC#6, especially antibodies to peptides A6 and A13 were placed in the periphery. According to the univariate data analysis, antibodies to peptide A6 were interesting, as they posed significant contrasts for all pairs of groups in which the control group was included, meaning that the expression of antibodies to A6 was significantly raised in all anti-CENP-F-positive sera compared to the controls.

A PCA of the complete matrix was performed to investigate the change in antibody profile as a function of the time of blood sampling. This was performed by analyzing the six patients that were represented by more than one blood sample (PS 2, PS 6, PS 8, PS 9, PS 12 and PS 17). Additional file2: Figure S2 illustrates the 2D score plots of PC#1 and PC#2 for each of the six patients and reveals a general pattern of migration towards the negative end of PC#1 as a function of blood sampling time, suggesting a correlation between diagnosis/tumor stage and sample time.

The human CENP-F sequence (Uniprot KB ID: P49454), comprising 3210 amino acids, was used to construct 33 overlapping peptides, A1-A33, which span the amino acid sequence 1855-2189 (Figure 1). The length of each peptide was 20 amino acids, with an overlap of 10 amino acids to the next peptide. At the N-terminus, a cysteine residue was added in order to have the possibility to screen the peptides using CovaLink™ technology (see peptide ELISA). The peptides A1-A16, A18, A20-A33 were purchased from Schäfer-N (Copenhagen, Denmark), while peptides A17 and A19 were synthesized as described elsewhere[24]. The peptides, A1-A33, are listed in the supplementary table (Additional file3: Table S1).

Sera were routinely screened for the presence of antinuclear antibodies (ANA) by IIF using HEp-2 cells as described below. Out of approximately 175,000 samples, submitted for routine ANA screening over a period of approximately 8 years, 42 samples from 28 patients were positive for the NSp-II pattern, characteristic for CENP-F antibodies[25]. Most patients were represented by a single serum sample, while a few were represented by up to four samples (Table 1). Each individual anti-CENP-F-positive sample was given a unique identification code, PS, followed by a number from 1-28 referring to the patient. A decimal number from 1-4 was added to the samples from patients represented more than once, 1 referring to the oldest sample and 4 to the most recent sample. The following six patients were represented more than once: PS 2 (PS 2.1-2.4), PS 6 (PS 6.1-6.2), PS 8 (PS 8.1-8.3), PS 9 (PS 9.1-9.4), PS 12 (PS 12.1-12.4) and PS 17 (PS 17.1-17.3). In addition, 30 samples that all tested negative for ANA were used as controls. These are referred to as Ctrl 1-30. A list of the medical histories and registered cancer diagnoses among patients positive of CENP-F antibodies was acquired retrospectively from the National Board of Health (Table 1). Based on this, analyzed samples were grouped as follows: Controls (30/58), patients without neoplasia (9/58), patients with benign tumors (5/58), patients with invasive cancer (14/58). In addition 10 sera positive for CCP antibodies, 10 sera positive for DNA antibodies and 10 sera from healthy blood donors were analysed by ELISA. All serum samples were acquired from the serum bank at Statens Serum Institut.

Sera were diluted in phosphate-buffered saline (PBS) (10 mM Na₂HPO₄/NaH₂PO₄, 0.15 M NaCl, pH 7.2) (1:160) and 20 μl were applied to wells of glass slides with fixed HEp-2 cells (Immunoconcepts, Sacramento, CA, USA) and incubated at RT in a humidified chamber. Wells were rinsed with PBS, and 20 μl FITC-conjugated rabbit Igs against human IgG (DAKO, Copenhagen, Denmark), diluted in PBS (1:40), were applied to the wells and incubated 30 min. The wells were rinsed with PBS and the slides were mounted with coverslips and inspected with a fluorescence microscope (Aristoplan (Leica), BH2 (Olympus) or Eurostar (Euroimmun)). The immunofluorescence pattern and the intensity were graded by intensity of 0 (negative), +1 (weak), +2 (medium), 3+ (strong) and recorded essentially as described elsewhere[52].

In initial experiments, CovaLink plate immobilization was compared with Maxisorp plate adsorption. The covalink plates showed too high background and further experiments were done with simple adsorption in Maxisorp plates. The wells of a 96-well Maxisorp microtitre plate (Nunc, Roskilde, Denmark) were coated with 1.5 μg/well peptide dissolved in carbonate buffer (15 mM Na₂CO₃, 35 mM NaHCO₃, 0.001% phenol-red, pH 9.6) and incubated over night at 4°C. Wells were rinsed with TTN buffer and incubated with sera (with or without aluminum hydroxide pretreatment) diluted in TTN (1:200) for 1 h. The wells were rinsed with TTN (3 × 5 min) and incubated 1 h with alkaline phosphatase-conjugated goat IgG against human IgG (Sigma Aldrich, Steinheim, Germany) followed by three washes with TTN buffer (5 min each). Bound antibodies were quantified using para-nitrophenylphosphate (1 mg/mL) (Sigma Aldrich) dissolved in alkaline phosphatase substrate buffer (1 M diethanolamine, 0.5 mM MgCl₂, pH 9.8). After sufficient color reaction, the absorbance was measured at 405 nm on a Thermomax microtitre plate reader (Molecular Devices, Menlo Park, CA, USA). All samples were analyzed in duplicate.

CCP antibodies were determined with the CCP2 ELISA kit (Eurodiagnostica, Malmö, Sweden) and DNA antibodies were determined by ELISA with the VarElisa kit (Thermo Scientific, Uppsala, Sweden) following the instructions of the manufacturers.

All sera-pool screenings were normalized relative to internal reference sera and to each other.

Multivariate data analysis was performed using the principal component analysis (PCA) software, LatentiX (Latent5, downloaded from latentix.com). Prior to this, the normalized data for the individual screenings was mounted in two different data matrices. One matrix, referred to as the complete matrix, contained data from all the individual screenings, while the other, referred to as the fused matrix, contained the data from the patients represented by multiple samples fused into one average data unit per patient. This was done to allow equal impact of each patient. Both matrices were transformed logarithmically prior to analysis in order to approximate a Gaussian distribution. In LatentiX, the data was transformed using the autoscale function and the PCA models were calculated without validation.