Indicators of Nonresponsive Celiac Disease
To identify candidate antibody biomarkers of NRCD, we applied a quantitative, specificity-based screening method to identify peptide mimotopes from random peptide libraries that capture disease-specific antibodies. Pools of two to four NRCD patients' sera, a pool of three recovered (Marsh 0) CD patients' sera and pools of four to five non-CD control patients' sera were used for library screening. Non-CD controls included healthy individuals without symptoms and negative TG2 serology as well as symptomatic disease controls with normal small bowel mucosal biopsies. To disfavour nonspecific peptide binding to antibodies in non-CD individuals, four rounds of subtractive FACS were performed by sorting library peptides that did not bind to control patients' sera (n = 33). A bacterial display peptide library comprised of three distinct pooled libraries of the form X15, X12CX3 and X4CX7CX4 (where X is any amino acid and C represents a site restricted to cysteine) with 1.2 × 10 members was used. Random peptide libraries of 15 amino acids are capable of mimicking diverse linear and discontinuous epitopes. Typical linear B-cell epitopes contain six to nine amino acids. Although discontinuous epitopes can span 9–22 amino acids, the majority of conformational epitopes include at least one linear stretch of four to seven residues. A library of 10 random 15-mer peptides contains, in theory, each of the 1.3 × 10 possible unique 7-mers with greater than 95% confidence. Cells from the final rounds of screening were plated and peptide sequences from individual colonies were determined (Table S1). Peptides obtained by screening exhibited a consensus of Qxxx(A/S/P)FP(E/D), and the frequency of peptides with this motif increased substantially during subsequent cycles of FACS (Figure 1a). To reveal other potentially important flanking residues, the amino acid sequences of peptides from the final enriched library population were aligned (Figure 1b). The position of the motif sequence within the peptide was highly conserved, which indicated the importance of the Gln at the N-terminus. This Gln was part of the N-terminal linker sequence (GQSGQ) upstream of the randomised 15-mer of each peptide library. The consensus motif was Q(P/V/A)(V/E)(A/D/Q)(A/P)FP(E/D)(A/R/Q), which shared eight identical residues with an immunodominant B-cell epitope (DGP3) in patients with active CD (Figure 1c). These mimotopes were similar to previously described B-cell epitopes of γ- and α-gliadin (Figure 1c) and T-cell epitopes from ω-gliadin, secalin and hordein (not shown).
(Enlarge Image)
Figure 1.
Consensus B-cell epitope identified using bacterial display random peptide libraries. (a) The enrichment of library members containing the FPE motif during the final five rounds of sorting. Alternating subtraction (s) and enrichment (e) steps were used with unique pools of three to five patients' sera. (b) Web logo plot of the frequency of an amino acid appearing in peptides displayed by 17 single colonies of E. coli from the final round of FACS. Larger font size indicates a higher probability of appearance. (c) Comparison of six library-isolated peptide sequences to an expanded epitope (DGP3) and B-cell epitopes from γ- and α-gliadin.
Down Selection and Characterisation of Library-Isolated Peptides. To identify clones exhibiting cross-reactivity and specificity for antibodies in NRCD sera, individual unique library clones were assayed for binding to serum IgA and IgG from six NRCD patients and four responsive CD patients (Figure S1). The clones that reacted above background levels with at least four of the six NRCD patients were then assayed with the original cohort of 11 NRCD and 25 responsive CD patients. Two of the best performing clones (C83 and C139) from the final library population, along with previously optimised clone DGP3, were further assayed using only IgG secondary antibodies with all 15 NRCD and 45 responsive CD patients (Figure 2a). Sera from NRCD and responsive CD exhibited a significant difference in reactivity with each clone (Wilcoxon rank-sum test P < 0.0001, 0.0003, 0.0001 for DGP3, C83 and C139 respectively). Among responsive patients with serology available at diagnosis, 23/24 showed a reduction in response to DGP3 after 1 year of GFD as expected (Figure 2b). Assays for IgA reactivity revealed a similar trend, but with more overlap between NRCD and responsive CD patient titres, because the mean signal intensities were decreased overall (Figure S2). Thus, random peptide library screening revealed an immunodominant dGP epitope recognised by serum IgG from patients with NRCD, but not by those with responsive CD.
(Enlarge Image)
Figure 2.
Reactivity of peptides with nonresponsive coeliac disease (NRCD) and responsive CD patients measured by flow cytometry. Clones were from the random peptide library screened with NRCD patient sera. (a) Response of serum IgG from NRCD (n = 15) and responsive CD patients (n = 45) to peptides DGP3, C83 and C139. Patients that reacted at or below background levels are represented here as 0.1% of the maximum signal, so that they appear on the logarithmic plot. Horizontal lines represent the mean. (b) Antibody reactivity with DGP3 after 1 year of gluten-free diet (GFD) for responsive patients that had serum taken at diagnosis and after 1 year of GFD (n = 24).
IgG Antibodies to dGP Are an Effective Marker of CD Recovery. The sequence similarity of the best performing mimotopes to known deamidated gliadin B-cell epitopes (Figure 1c) coupled with the responsiveness of antibody titres to DGP3 to a GFD (Figure 2b) indicated that the discovered peptides mimicked dGP. We confirmed this hypothesis with a commercially available anti-dGP IgG antibody ELISA (Figure 3a). The mean antibody titre of NRCD patients was seven-fold greater than that of responsive CD patients, and the differences in ELISA response between the two groups were significant (P < 0.0001). A positive threshold value of 12 units yielded 87% sensitivity and 89% specificity for NRCD based solely on dGP serology. There was no statistical difference between titres for RCD patients (current and latent RCD, n = 9) and the remaining NRCD patients (n = 6) (P = 0.95). Furthermore, anti-dGP IgG titres correlated with the severity of mucosal damage represented by Marsh classifications (Figure 3b). The ELISA signal differences between Marsh 0 and Marsh 1 patients or Marsh 2 and Marsh 3a/3b patients were nonsignificant, but each of the other Marsh binary comparisons was significant (Table S2). To further verify that dGP was the antigen mimicked by bacterial display peptide mimotopes, the antibody reactivity from the individual patient assays with DGP3 and C139 was compared to dGP ELISA values. Serum IgG-class dGP ELISA values correlated with cytometry measurements (Spearman ρ = 0.52 and P < 0.0001 for C139; ρ = 0.46 and P = 0.0002 for DGP3) (Figure 3c).
(Enlarge Image)
Figure 3.
Confirmation of deamidated gliadin as the primary antigen mimicked by library-isolated peptides and a marker of coeliac disease (CD) patient recovery. (a) Anti-deamidated gliadin peptide (dGP) IgG antibody ELISA discriminates nonresponsive CD (NRCD) (n = 15) and responsive CD patients (n = 45) (P < 0.0001). An adjusted ELISA cut-off of 12 units yielded 87% sensitivity and 89% specificity. (b) Anti-dGP titre increases with the severity of mucosal damage as described by Marsh staging (see Supplementary Table S2 for binary significance comparisons). (c) Correlation between flow cytometry assays with bacterial display peptides and anti-dGP IgG ELISA (P < 0.0001 for C139 and P = 0.0002 for DGP3). Each point represents an individual patient's reactivity with C139 (▲) and DGP3 (●). (d) Receiver-operating characteristic (ROC) curve for the anti-dGP IgG antibody ELISA test. The area under the curve (AUC) was 0.94 (95% confidence interval, 0.88–0.99).
The true-positive rate (sensitivity) was plotted against the false-positive rate (1-specificity) in an ROC curve as a function of varying ELISA unit cut-offs (Figure 3d). The positive threshold of 12 units yielded 87% sensitivity and 89% specificity, and a cut-off of 10 units yielded 93% sensitivity and 84% specificity. Because only one NRCD patient was positive for TG2 and four others were weakly positive ( Table 1 ), the serum IgG dGP ELISA substantially outperforms the IgA anti-TG2 ELISA with an ROC area under the curve (AUC) of 0.94 vs. 0.61 respectively (Figure S3). An optimal serum IgA-class TG2 ELISA positive threshold of 5 U/mL yielded 33% sensitivity and 100% specificity, misclassifying 10 of 15 NRCD patients as GFD responders. Assays with DGP3 and C139 (AUC = 0.88 and 0.87 respectively) displayed on bacteria using flow cytometry also outperformed TG2 (Figure S3), but did not exceed the diagnostic accuracy of the dGP ELISA (Figure 3d). On the basis of our novel finding of the persistence of elevated IgG dGP antibodies, we have suggested a revised diagnostic algorithm using non-invasive serological tests for the monitoring of CD patient recovery during a GFD (Figure 4).
(Enlarge Image)
Figure 4.
Proposed diagnostic algorithm for clinical follow-up of coeliac disease (CD) patients on a gluten-free diet (GFD).
Results
Screening for Antibody Biomarkers of NRCD Using Bacterial Display Peptide Libraries
To identify candidate antibody biomarkers of NRCD, we applied a quantitative, specificity-based screening method to identify peptide mimotopes from random peptide libraries that capture disease-specific antibodies. Pools of two to four NRCD patients' sera, a pool of three recovered (Marsh 0) CD patients' sera and pools of four to five non-CD control patients' sera were used for library screening. Non-CD controls included healthy individuals without symptoms and negative TG2 serology as well as symptomatic disease controls with normal small bowel mucosal biopsies. To disfavour nonspecific peptide binding to antibodies in non-CD individuals, four rounds of subtractive FACS were performed by sorting library peptides that did not bind to control patients' sera (n = 33). A bacterial display peptide library comprised of three distinct pooled libraries of the form X15, X12CX3 and X4CX7CX4 (where X is any amino acid and C represents a site restricted to cysteine) with 1.2 × 10 members was used. Random peptide libraries of 15 amino acids are capable of mimicking diverse linear and discontinuous epitopes. Typical linear B-cell epitopes contain six to nine amino acids. Although discontinuous epitopes can span 9–22 amino acids, the majority of conformational epitopes include at least one linear stretch of four to seven residues. A library of 10 random 15-mer peptides contains, in theory, each of the 1.3 × 10 possible unique 7-mers with greater than 95% confidence. Cells from the final rounds of screening were plated and peptide sequences from individual colonies were determined (Table S1). Peptides obtained by screening exhibited a consensus of Qxxx(A/S/P)FP(E/D), and the frequency of peptides with this motif increased substantially during subsequent cycles of FACS (Figure 1a). To reveal other potentially important flanking residues, the amino acid sequences of peptides from the final enriched library population were aligned (Figure 1b). The position of the motif sequence within the peptide was highly conserved, which indicated the importance of the Gln at the N-terminus. This Gln was part of the N-terminal linker sequence (GQSGQ) upstream of the randomised 15-mer of each peptide library. The consensus motif was Q(P/V/A)(V/E)(A/D/Q)(A/P)FP(E/D)(A/R/Q), which shared eight identical residues with an immunodominant B-cell epitope (DGP3) in patients with active CD (Figure 1c). These mimotopes were similar to previously described B-cell epitopes of γ- and α-gliadin (Figure 1c) and T-cell epitopes from ω-gliadin, secalin and hordein (not shown).
(Enlarge Image)
Figure 1.
Consensus B-cell epitope identified using bacterial display random peptide libraries. (a) The enrichment of library members containing the FPE motif during the final five rounds of sorting. Alternating subtraction (s) and enrichment (e) steps were used with unique pools of three to five patients' sera. (b) Web logo plot of the frequency of an amino acid appearing in peptides displayed by 17 single colonies of E. coli from the final round of FACS. Larger font size indicates a higher probability of appearance. (c) Comparison of six library-isolated peptide sequences to an expanded epitope (DGP3) and B-cell epitopes from γ- and α-gliadin.
Down Selection and Characterisation of Library-Isolated Peptides. To identify clones exhibiting cross-reactivity and specificity for antibodies in NRCD sera, individual unique library clones were assayed for binding to serum IgA and IgG from six NRCD patients and four responsive CD patients (Figure S1). The clones that reacted above background levels with at least four of the six NRCD patients were then assayed with the original cohort of 11 NRCD and 25 responsive CD patients. Two of the best performing clones (C83 and C139) from the final library population, along with previously optimised clone DGP3, were further assayed using only IgG secondary antibodies with all 15 NRCD and 45 responsive CD patients (Figure 2a). Sera from NRCD and responsive CD exhibited a significant difference in reactivity with each clone (Wilcoxon rank-sum test P < 0.0001, 0.0003, 0.0001 for DGP3, C83 and C139 respectively). Among responsive patients with serology available at diagnosis, 23/24 showed a reduction in response to DGP3 after 1 year of GFD as expected (Figure 2b). Assays for IgA reactivity revealed a similar trend, but with more overlap between NRCD and responsive CD patient titres, because the mean signal intensities were decreased overall (Figure S2). Thus, random peptide library screening revealed an immunodominant dGP epitope recognised by serum IgG from patients with NRCD, but not by those with responsive CD.
(Enlarge Image)
Figure 2.
Reactivity of peptides with nonresponsive coeliac disease (NRCD) and responsive CD patients measured by flow cytometry. Clones were from the random peptide library screened with NRCD patient sera. (a) Response of serum IgG from NRCD (n = 15) and responsive CD patients (n = 45) to peptides DGP3, C83 and C139. Patients that reacted at or below background levels are represented here as 0.1% of the maximum signal, so that they appear on the logarithmic plot. Horizontal lines represent the mean. (b) Antibody reactivity with DGP3 after 1 year of gluten-free diet (GFD) for responsive patients that had serum taken at diagnosis and after 1 year of GFD (n = 24).
IgG Antibodies to dGP Are an Effective Marker of CD Recovery. The sequence similarity of the best performing mimotopes to known deamidated gliadin B-cell epitopes (Figure 1c) coupled with the responsiveness of antibody titres to DGP3 to a GFD (Figure 2b) indicated that the discovered peptides mimicked dGP. We confirmed this hypothesis with a commercially available anti-dGP IgG antibody ELISA (Figure 3a). The mean antibody titre of NRCD patients was seven-fold greater than that of responsive CD patients, and the differences in ELISA response between the two groups were significant (P < 0.0001). A positive threshold value of 12 units yielded 87% sensitivity and 89% specificity for NRCD based solely on dGP serology. There was no statistical difference between titres for RCD patients (current and latent RCD, n = 9) and the remaining NRCD patients (n = 6) (P = 0.95). Furthermore, anti-dGP IgG titres correlated with the severity of mucosal damage represented by Marsh classifications (Figure 3b). The ELISA signal differences between Marsh 0 and Marsh 1 patients or Marsh 2 and Marsh 3a/3b patients were nonsignificant, but each of the other Marsh binary comparisons was significant (Table S2). To further verify that dGP was the antigen mimicked by bacterial display peptide mimotopes, the antibody reactivity from the individual patient assays with DGP3 and C139 was compared to dGP ELISA values. Serum IgG-class dGP ELISA values correlated with cytometry measurements (Spearman ρ = 0.52 and P < 0.0001 for C139; ρ = 0.46 and P = 0.0002 for DGP3) (Figure 3c).
(Enlarge Image)
Figure 3.
Confirmation of deamidated gliadin as the primary antigen mimicked by library-isolated peptides and a marker of coeliac disease (CD) patient recovery. (a) Anti-deamidated gliadin peptide (dGP) IgG antibody ELISA discriminates nonresponsive CD (NRCD) (n = 15) and responsive CD patients (n = 45) (P < 0.0001). An adjusted ELISA cut-off of 12 units yielded 87% sensitivity and 89% specificity. (b) Anti-dGP titre increases with the severity of mucosal damage as described by Marsh staging (see Supplementary Table S2 for binary significance comparisons). (c) Correlation between flow cytometry assays with bacterial display peptides and anti-dGP IgG ELISA (P < 0.0001 for C139 and P = 0.0002 for DGP3). Each point represents an individual patient's reactivity with C139 (▲) and DGP3 (●). (d) Receiver-operating characteristic (ROC) curve for the anti-dGP IgG antibody ELISA test. The area under the curve (AUC) was 0.94 (95% confidence interval, 0.88–0.99).
The true-positive rate (sensitivity) was plotted against the false-positive rate (1-specificity) in an ROC curve as a function of varying ELISA unit cut-offs (Figure 3d). The positive threshold of 12 units yielded 87% sensitivity and 89% specificity, and a cut-off of 10 units yielded 93% sensitivity and 84% specificity. Because only one NRCD patient was positive for TG2 and four others were weakly positive ( Table 1 ), the serum IgG dGP ELISA substantially outperforms the IgA anti-TG2 ELISA with an ROC area under the curve (AUC) of 0.94 vs. 0.61 respectively (Figure S3). An optimal serum IgA-class TG2 ELISA positive threshold of 5 U/mL yielded 33% sensitivity and 100% specificity, misclassifying 10 of 15 NRCD patients as GFD responders. Assays with DGP3 and C139 (AUC = 0.88 and 0.87 respectively) displayed on bacteria using flow cytometry also outperformed TG2 (Figure S3), but did not exceed the diagnostic accuracy of the dGP ELISA (Figure 3d). On the basis of our novel finding of the persistence of elevated IgG dGP antibodies, we have suggested a revised diagnostic algorithm using non-invasive serological tests for the monitoring of CD patient recovery during a GFD (Figure 4).
(Enlarge Image)
Figure 4.
Proposed diagnostic algorithm for clinical follow-up of coeliac disease (CD) patients on a gluten-free diet (GFD).
Source...