Patient cohort

Combinatorial glycoarray

A TLC autosampler was used to apply glycolipid onto PVDF membranes as previously described [26]. Glycolipids and their sources, and the resulting GSC antigens created, are listed in Table 1. LM1 and sulfated glucuronyl paragloboside (SGPG) were extracted and purified as described [27,28]. Complexes (1:1 by weight) were created by admixing equal volumes of the component glycolipids. Arrays were blocked and then incubated with sera diluted 1 in 100. After washing, rabbit anti-human IgG horseradish peroxidase conjugated secondary antibody (Dako) was applied at 1:30,000 dilution. The IgG response alone was assessed as extensive previous observations have indicated that IgG is the critical immunoglobulin class in GBS. Binding was detected by enhanced chemiluminescence (ECL+, Amersham/GE Healthcare). Radiographs were digitised by flatbed scanning and spot intensity calculated using TotalLab image analysis software (Nonlinear Dynamics). The order of analysis of the sera was determined by random number generation, the sera coded, and the investigator performing image analysis blinded to the identity of the serum applied to each membrane.

For practical purposes clinical samples were screened in batches. Initially, GBS cases and healthy controls were screened using an 11x11 and then 28x10 array (Table 1). Antibody-antigen interactions statistically significantly associated with GBS, in comparison with healthy controls, were then re-evaluated with the inclusion of all 87 disease control sera.

Statistical analysis

Initial analysis was performed using the Chi-squared test on binary data (antigen binding present or absent) to investigate overall antibody-antigen-disease associations. Intensity values were used to produce heatmaps, which underwent iterative clustering to reveal patterns of binding. Differences in binding intensity between antigen groups (single gangliosides, glycolipids and glycolipid complexes) were assessed by Kruskal-Wallis with post-hoc analysis using the Mann-Whitney test. To compare the binding intensities achieved by GBS and control sera to the multiple different individual antigens on the array, raw intensity values were first logarithmically transformed. A Mixed Model Analysis of Variance (ANOVA) was then used to make multiple comparisons (in which zero results were considered as the lowest value for a continuum of binding intensities). This approach was also used to assess for associations between clinical features and antigen binding intensity, and allows for the fact that intensity values of some antigens are likely to be inter-related. The problems associated with analysing data with multiple zero values were addressed by considering and comparing results obtained by the Chi-squared and ANOVA and Mann-Whitney approaches. The latter two methods were used after replacing zero intensity values with a small positive value. Both Bonferroni and Benjamini and Hochberg step-up correction techniques were used to correct for multiple comparisons. The former is the most conservative correction for multiple comparisons, but therefore comes at the cost of a higher risk of false negatives. As such, the less stringent and Benjamini and Hochberg step-up technique was also used in certain exploratory situations.

Patient cohort

Sufficient amounts of serum were available from 181 of 225 patients. Clinical data were available for 180 of these 181 patients. 103 were male (57.2%) and 77 female (42.8%). The mean age of patients was 52.6 years (range 7 to 89). The mean/median age was 44.2/44 for healthy controls, 55.7/56 for OND controls, and 50.4/49 (range 25-83) for controls overall. Electrophysiological data were available for 172 out of 181 patients. A single set of nerve conduction studies (NCS) was performed at a median of 17 days (inter-quartile range 11-21 days) following the onset of weakness. According to the Hadden criteria [24], the majority were classified either as AIDP (85 cases, 49.4%) or equivocal (73 cases, 42.4%). Equivocal patients had a definite clinical diagnosis of GBS, but in these instances NCS were unable to distinguish demyelinating or axonal subtypes following the single timepoint testing that characterised the study design. Only 5 patients had electrophysiology indicative of AMAN (Acute Motor Axonal Neuropathy, 5.6% of characterised cases). Four cases had normal neurophysiology and in 5 the nerves were unexcitable. There were 5 deaths (2.8%) and 41 patients (22.8%) required mechanical ventilation at some point.

Glycoarray Assay Performance

The inter- and intra-assay coefficients of variation (CV) for a 10x10 glycoarray had been previously calculated at 4.1% and 8.6% respectively, using an anti-GM1 monoclonal antibody [17]. One positive GBS serum identified in the first screen of the 28x10 assay was retested on 12 occasions over a 3 month time period during the assessment of GBS and healthy control samples. The inter-assay CV for this sample over this period, involving multiple freeze-thaw cycles and different batches of reagents, was 13.4%.

Initial Screening

Following initial screening using 11x11 and 28x10 arrays, IgG antibodies in the sera of 113 GBS patients (62.4%) versus 11 controls (14.9%) bound to one or more antigens (p<0.001, Chi-squared). GBS-associated sera displayed many different binding patterns towards glycolipids and complexes (Figure 1A,B,D). No single antigen binding pattern was dominant. Overall, GBS sera bound a wider range of antigens at higher intensity compared to healthy controls (Figure 1C,D). Reactivity towards 176 different single antigens and complexes was seen with sera from GBS patients, whereas healthy control sera only bound 31 of the 181 different antigens (p<0.0001, Binomial). Figure S1 provides a detailed, zoomable version of this heatmap showing binding patterns for each individual antigen and serum. GBS patients testing positive were an average of 2.5 years older than those testing negative, whereas healthy controls testing positive were an average of 2.2 years younger than those testing negative. However, neither of these differences reached statistical significance (p=0.375 and p=0.589, independent samples T-test).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Diverse patterns of glycolipid binding in GBS sera.

(A,B) Representative 28x10 arrays. The antigen spotted at each grid location is revealed by combining the row and column headings given in the key below. Thus, locations (9,7) and (7,9) contain GA1:GM3 complexes. The line of ‘x’ across the centre of the membrane represents the negative control spots (methanol only spotted). The left most column and upper most row contain single antigens only.

(A) An array probed with GBS serum showing both complex attenuated and enhanced binding patterns. GD1b binding (14) is completely attenuated in complex with GD2 (location 14,6). In contrast, despite no demonstrable binding to GA1, GM3 or GD1a in isolation, an intense spot is seen with both GA1:GM3 (location 9,7) and GA1:GD1a (location 9,13) complexes. This is complex enhanced/complex dependent binding.

(B) Highly specific complex-dependent GBS sera. In this example, only the duplicate spots of GM1:phosphatidylserine complex (location 2,12) are bound.

(C) Boxplot of the distribution of maximum binding intensity for GBS versus healthy control (HC) sera assessed for single gangliosides, single glycolipids, and glycolipid complexes. Maximum binding intensity is significantly higher for GBS sera versus healthy control (HC) sera for each antigen group (*p<0.05, **p<0.0001). GBS sera maximum binding intensity for complexes is significantly higher than for either single glycolipids or gangliosides (†=p<0.001), but HC sera maximum binding intensity is insignificantly different between the antigen groups (not marked). (all Kruskal-Wallis/Mann-Whitney).

(D) Overview heatmap of raw, unclustered binding intensities for GBS (above horizontal white line) and HC (below horizontal white line) sera. Antigens to the right of the vertical white line are glycolipid complexes. The intensity of binding is given by the bar above the heatmap, ranging from black (negative) to red (most intense). A more detailed, zoomable version of this heatmap, allowing resolution of results for each antigen and each serum, is available online (Figure S1).

Key: (1)Sphingomyelin, (2)Phosphatidylserine, (3)Globoside, (4)Cereamide trihexosides (CTH), (5) Sulfated glucuronyl paragloboside (SGPG), (6)GM2, (7)GM3, (8)GD2, (9)Asialo-GM1, (10)Galactocerebroside (GalC), (11)LM1, (12)GM1, (13)GD1a, (14)GD1b, (15)GD3, (16)GQ1b, (17)GT1b, (18)Sulfatide.

https://doi.org/10.1371/journal.pone.0082337.g001

Glycolipid complexes

Overall, complexes are bound at a significantly higher median intensity compared to either single gangliosides or glycolipids (Figure 1C, p<0.001, Kruskal-Wallis/Mann-Witney). Furthermore, if binding to the 11 single gangliosides alone were assessed by glycoarray, the positivity rate would be 80% for AMAN cases, but only 11.8% for those with demyelinating electrophysiology. In contrast, glycolipid complexes are bound by IgG antibodies in 61.2% of sera from demyelinating cases. Overall, the use of glycolipid complex binding intensity values alone provides the best trade-off between sensitivity and specificity for discriminating GBS from healthy control sera, significantly outperforming single ganglioside and/or single glycolipid results, as assessed by receiver operator curve (ROC) analysis (p<0.0001 and p<0.001 respectively) (Figure 2A). Only the assessment of glycolipid complexes (AUC 0.758) or the inclusion of all 181 antigens (AUC 0.756) resulted in a test with an area under the curve (AUC) value in excess of 0.75, the accepted cut-off for a clinically useful test [29]. The performance of these two antigen groups was insignificantly different (p=0.33, ROC analysis).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Complex enhanced binding.

(A) Receiver operator curves for single ganglioside antigens, single glycolipids, glycolipid complexes, and for all 181 antigens. Complex antigens (dashed green line) afford the best trade-off between sensitivity and specificity.

(B) Individual value plots for GBS sera binding intensity to single sulfatide (sul), single CTH, the sum of these single intensities (sulfatide+CTH), and CTH:sulfatide complexes.

(C) Comparison between the sum of single sulfatide and CTH binding intensities (sul+CTH) and binding intensity to CTH:sulfatide complexes in individual sera. Each line links the respective values in a single serum. Lines sloping upwards from left to right indicate complex enhanced binding with a single serum.

(D) Overview heatmap of corrected complex binding intensities. The intensity value for each complex has been corrected by subtracting the binding intensities of each component glycolipid. Any residual intensity indicates complex enhanced binding. As before, GBS sera lie above and healthy control (HC) sera below the horizontal white line. The same colour scale as Figure 1 has been used. A more detailed, zoomable version of this heatmap, allowing resolution of results for each complex antigen and each serum, is available online (Figure S3).

https://doi.org/10.1371/journal.pone.0082337.g002

In many cases the complex binding intensities produced by GBS sera were greatly in excess of those seen towards the component glycolipids of the complex assessed individually – a phenomenon referred to as complex enhanced binding [18] (example in Figure 2B,C). This pattern was much more frequent (59.7% versus 13.5%, p<0.0001, Chi-squared) and pronounced (median complex enhancement 5300 versus 0 arbitrary units, p<0.0001, Mann-Whitney) with GBS versus healthy control sera, as shown by the second heatmap (Figure 2D). This heatmap displays binding intensities for glycolipid complexes following subtraction of single glycolipid binding intensities (high resolution, zoomable version showing the pattern of enhancement for each complex antigen and serum available as Figure S3). Complex enhancement was exclusively seen with GBS sera for 134 of the 161 complexes tested.

Binding to a glycolipid complex without any detectable binding whatsoever to the component glycolipids presented individually has been termed ‘complex dependent’. Overall, 79 (43.6%) GBS sera showed this pattern, compared to only 3 (4.1%) of the healthy control sera (p<0.0001, Fisher’s Exact Test). Examples were seen involving all of the glycolipids assayed (Figure S2C).

Individual antigens with significantly increased binding intensities

Logarithmic transformation of the intensity values, and analysis by mixed effects model ANOVA with Bonferroni correction, revealed 9 antigens with significantly increased binding intensities in GBS as compared with healthy controls. Seven of these were glycolipid complexes. Using the less stringent Benjamini and Hochberg step-up correction for multiple comparisons, a further 31 significant antigens (28 complexes) are returned (Table 1). The same technique was used to identify specific antibody-antigen/intensity associations for disease subtype (by electrophysiological criteria), disease severity, identified preceding infection, cranial nerve involvement, and the requirement for mechanical ventilation (Table 2).

Antibody binding patterns in other neurological disease control sera

The 40 antigens identified by intensity analysis as significantly associated with GBS were then subjected to testing with the 87 OND controls, alongside a randomly selected subset of 20 anti-GSC antibody positive GBS sera. Comparing the results of the retested GBS sera with those from the original 28x10 assay, there were no significant differences in positivity rates (using Fisher’s Exact test) or average signal intensities (by Mann-Whitney) for 35 of the 40 antigens. All five of the discrepant antigens (sulfatide:GT1a, LM1:GT1a, GM1:GT1a, GalC:GT1a, and SGPG:GalC) were excluded from further analysis.

Hierarchical clustering using a Pearson’s Correlation metric was used to uncover patterns of binding in GBS associated sera, which was then compared to that seen in controls. This revealed two major GBS binding clusters – one containing sulfatide complexes and the other GM1 complexes (boxed in Figure 3). Positivity in control cases mainly mapped to the sulfatide cluster. Four sulfatide complexes cluster separately, three of these (GM1:sulfatide, GD1b:sulfatide and GA1:sulfatide, small arrows) are infrequently represented in controls, Sulfatide:GalC (large arrow) binding is present in OND controls at a frequency similar to other antigens in the sulfatide cluster. Eight antigens displayed a significantly increased binding intensity with GBS as compared to OND control sera (Table 3), including GM1, 3 GM1 complexes, and 4 sulfatide complexes. The ROC/assay performance of this subset of 8 antigens is insignificantly different to that of all 35 antigens together (p=0.32, Figure 4A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Heatmap of glycolipid complex binding in GBS and control sera.

Clustered heatmap for the 35 antigens significantly associated with GBS (arranged on the x-axis), comparing the 181 GBS sera and 161 control sera (on the y-axis). Each row represents a single serum. The log transformed arbitrary binding intensity for each antigen is represented by the colour scale, from red (log transformed arbitrary intensity of 18, most intense) to the green background (log transformed arbitrary intensity of 0, no binding), as shown at the top of the figure. Each heatmap panel represents a separate group, as labelled. (GBS, Guillain-Barré syndrome; MS, Multiple Sclerosis; ONND, Other Non-Inflammatory Neurological Diseases; HC, Healthy Controls). Blue boxes delineate clusters of GM1 and sulfatide-complex antigens. Arrows mark sulfatide complexes which cluster separately.

https://doi.org/10.1371/journal.pone.0082337.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Differing patterns of sulfatide complex enhanced binding in disease versus control sera.

(A) The receiver operator curve for the 8 antigen intensities significantly associated with GBS versus OND controls is insignificantly different to that produced by including all 35 antigens.

(B) Individual value plots of binding intensity for GBS, healthy control (HC), and other neurological disease (OND) sera to sulfatide, CTH:sulfatifde complex, and GA1:sulfatide complex.

(C) Mean change in complex compared to summed single antigen binding intensity for CTH:sulfatide and GA1:sulfatide. On average, GBS sera show complex enhancement, whereas HC and OND sera show complex attenuation.

https://doi.org/10.1371/journal.pone.0082337.g004

Sulfatide and sulfatide series complex antibodies

Sulfatide-complex antibodies form the largest group of antibodies found to be statistically associated with GBS. Sulfatide as a single glycolipid is bound at relatively low intensity by IgG in GBS sera, there is some overlap with healthy control binding intensity, and no significant difference in binding as compared to OND control sera. Once again however, disease binding intensity and separation from control cases is substantially improved with certain sulfatide complexes, such asialo-GM1:sulfatide and CTH:sulfatide (Figure 4B). Furthermore, there is a differential relative response to single sulfatide and sulfatide-complex antigens between GBS and control sera. Overall, GBS sera show a more frequent and more pronounced complex-enhanced response to these sulfatide-containing complexes than do OND or healthy control sera (Kruskal-Wallis test for degree of complex enhancement, CTH:sul p=0.025, GA1:sul p=0.037, both p<0.001 when corrected for ties, Chi-squared comparison of frequency of complex enhancement, p<0.0001 for both antigens) (Figure 4C).

GM1 and GM1 series complex antibodies

In contrast to sulfatide, binding to single GM1 is significantly increased in GBS compared to OND control sera (Table 3). Even so, GM1 complexes have greater sensitivity and increased binding intensity over GM1 alone, without a significant loss of specificity (Figure 5A). Analysis of relative binding intensities in single patients reveals that both PS:GM1 and sulfatide :GM1 complexes are often bound by sera from patients entirely negative for GM1 reactivity (Figure 5B,C). The reverse also occurs, but is less often seen. Furthermore, investigating the association between PS:GM1 and sulfatide:GM1 binding intensity in individual sera uncovers at least three different populations of GM1-complex antibodies – those which preferentially (or indeed exclusively) bind one complex, and those which bind both relatively equally (Figure 5D).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. GM1 complex antigens.

(A) Individual value plots for selected antigens.

(B) Association between sulfatide:GM1 and GM1 binding intensities in individual sera. Points on the y-axis are GM1 negative, sulfatide:GM1 positive. Sera close to the diagonal blue line (joining points of identical binding intensity for both antigens) bind both GM1 and sulfatide:GM1 complex with similar intensity.

(C) Association between phosphatidylserine:GM1 and GM1 binding intensities in individual sera. Points on the y-axis are GM1 negative, PS:GM1 positive. Sera on the x-axis are GM1 positive, PS:GM1 negative.

(D) Association between phosphatidylserine:GM1 and sulfatide:GM1 binding intensities in individual sera.

https://doi.org/10.1371/journal.pone.0082337.g005

Comparison with ELISA

Just 39 of the 180 (21.7%) GBS sera had been positive when tested contemporaneously using a conventional GM1, GD1a, and GQ1b ELISA [25]. For these three ganglioside antigens alone, the array detected binding in only 27 cases, and was negative for 15 of the sera which had been positive on ELISA. Only 9 of 85 cases (10.6%) with demyelinating pattern electrophysiology were positive by ELISA (versus 62.4% for binding to all antigens on the array), compared to 4 of the 5 AMAN patients (80%, identical to the array). The remainder of ELISA positive cases had either equivocal (24 patients) or unavailable (2 patients) electrophysiology results. Only one GBS serum which was positive by ELISA was entirely negative by glycoarray.