Identification of anti-COX2 antibody in IAA patients
Protoarray protein microarray analysis was performed to identify potential autoantibodies in patients with IAA (n = 7), large granular lymphocyte (LGL) leukemia (n = 12), rheumatoid arthritis (RA) (n = 10) and healthy controls (n = 5). Figure 1A presents all protein targets of autoantibodies that were present in at least two of the tested IAA patients. COX-2 was the only protein with IAA restricted autoantibody levels with > 20-fold difference to healthy controls in all positive cases.
Next, we developed an aCOX-2 IgG DELFIA immunoassay to confirm the microarray results. The DELFIA immunoassay results were in accordance with the protein microarray data, and aCOX-2 antibodies were confirmed in all three index cases but in none of the negative control patients (Supplementary Fig. 3).
To validate the findings in a larger patient cohort, we collected an international IAA cohort including a total of 405 patients (US n = 259, Japan n = 108, and Nordic countries n = 38). Sample positivity threshold for aCOX-2 Ab was set to correspond the turning point in ROC curve maximizing test sensitivity (0.36) and specificity (Supplementary Fig. 1). The highest aCOX-2 Ab positivity (61%) was found in the Nordic cohort, while in the US and Japanese IAA cohorts there were 26% and 37% aCOX-2 Ab positive patients, respectively (Fig. 1B). aCOX-2 Ab positive patients were mostly adults (Fig. 1C) and when the IAA patient cohort was split in 10-year age intervals, a clear age dependent increase in aCOX-2 Ab positivity was observed especially in patients over 40 years of age (Fig. 1D). Interestingly, the age distribution in different international cohorts followed the aCOX-2 Ab positivity percentages as patients in the Nordic cohort were the oldest, and patients in the USA cohort were the youngest (Fig. 1E). As only sporadic aCOX-2 Ab positive cases were observed in the pediatric patients, only adult (> 18 years old) patients (n = 334) were included in further analyses. The overall aCOX-2 Ab positivity in adult IAA patients was 37%. Descriptive statistics for the clinical parameters for adult IAA patients (n = 334) are presented in the Table 1, while the entire cohort (n = 405) including the pediatric patients is presented in the Supplementary Table 3.
Table 1
Study cohort demographics (adult IAA patients n = 334).
|
|
aCOX-2 Ab negative
|
aCOX-2 Ab positive
|
Total
|
p-value
|
|
|
(n = 209)
|
(n = 125)
|
(n = 334)
|
|
Age at diagnosis
|
Mean (SD)
|
39.8 (17.7)
|
61.8 (13.2)
|
48.1 (19.4)
|
< 0.001
|
(years)
|
Range
|
18.0–86.0
|
20.0–94.0
|
18.0–94.0
|
|
Gender
|
Male
|
49% (n = 102)
|
42% (n = 52)
|
46% (n = 154)
|
0.2137 (f)
|
|
Female
|
51% (n = 107)
|
58% (n = 73)
|
54% (n = 180)
|
|
PNH clone
|
Present
|
49% (n = 85)
|
63% (n = 69)
|
54% (n = 154)
|
0.0279 (f)
|
|
Absent
|
51% (n = 88)
|
37% (n = 41)
|
46% (n = 129)
|
|
Severity
|
Moderate
|
32% (n = 56)
|
29% (n = 30)
|
31% (n = 86)
|
0.6181 (c)
|
|
Severe
|
61% (n = 106)
|
61% (n = 63)
|
61% (n = 169
|
|
|
Very severe
|
7% (n = 13)
|
11% (n = 11)
|
9% (n = 24)
|
|
Hb at diagnosis
|
Mean (SD)
|
9.03 (2.17)
|
9.51 (9.68)
|
9.21 (6.22)
|
0.514
|
(g/dl)
|
Range
|
2.60–16.30
|
2.80–111.00
|
2.60–111.00
|
|
WBC at diagnosis
|
Mean (SD)
|
2.49 (1.55)
|
2.51 (1.23)
|
2.50 (1.44)
|
0.940
|
(x 10^9 /l)
|
Range
|
0.04–9.18
|
0.08–8.33
|
0.04–9.18
|
|
Plt at diagnosis
|
Mean (SD)
|
43.38 (48.58)
|
20.97 (16.54)
|
34.82 (40.97)
|
< 0.001
|
(x 10^9 /l)
|
Range
|
2–305
|
1–67
|
1–305
|
|
ANC at diagnosis
|
Mean (SD)
|
1.00 (0.97)
|
0.84 (0.72)
|
0.94 (0.88)
|
0.137
|
(x 10^9 /l)
|
Range
|
0.00–5.39
|
0.01–3.90
|
0.00–5.39
|
|
ALC at diagnosis
|
Mean (SD)
|
1.25 (0.79)
|
1.41 (0.71)
|
1.31 (0.76)
|
0.123
|
(x 10^9 /l)
|
Range
|
0.03–4.51
|
0.28–3.64
|
0.03–4.51
|
|
ARC at diagnosis
|
Mean (SD)
|
36.44 (38.91)
|
31.55 (21.88)
|
34.45 (33.05)
|
0.316
|
(x 10^9 /l)
|
Range
|
0.003–374.00
|
2.00–116.00
|
0.003–374.00
|
|
Hb at sampling
|
Mean (SD)
|
9.57 (2.07)
|
9.69 (1.83)
|
9.61 (1.99)
|
0.643
|
(g/dl)
|
Range
|
3.40–15.10
|
3.50–15.50
|
3.40–15.50
|
|
WBC at sampling
|
Mean (SD)
|
2.46 (1.61)
|
2.91 (3.61)
|
2.62 (2.50)
|
0.161
|
(x 10^9 /l)
|
Range
|
0.03–10.35
|
0.060–30.30
|
0.03–30.30
|
|
Plt at sampling
|
Mean (SD)
|
51.41 (62.07)
|
32.28 (41.54)
|
44.69 (56.39)
|
0.009
|
(x 10^9 /l)
|
Range
|
2–511
|
1–277
|
1–511
|
|
ANC at sampling
|
Mean (SD)
|
1.15 (1.18)
|
1.24 (1.80)
|
1.18 (1.42)
|
0.608
|
(x 10^9 /l)
|
Range
|
0.00–8.84
|
0.00–10.76
|
0.00–10.76
|
|
ALC at sampling
|
Mean (SD)
|
1.17 (0.75)
|
1.31 (0.85)
|
1.22 (0.79)
|
0.162
|
(x 10^9 /l)
|
Range
|
0.01–4.30
|
0.01–5.15
|
0.01–5.15
|
|
ARC at sampling
|
Mean (SD)
|
33.63 (32.46)
|
29.03 (21.61)
|
32.05 (29.17)
|
0.411
|
(x 10^9 /l)
|
Range
|
0.003–139.00
|
0.004–80.16
|
0.003–139.00
|
|
Abbreviations: SD, standard deviation; PNH, paroxysmal nocturnal hemoglobinuria; Hb, hemoglobin; WBC, white blood cells; Plt, platelets; ANC, absolute neutrophil count; ALC, absolute lymphocyte count; ARC, absolute reticulocyte count. p-values are calculated using the linear ANOVA model except for gender, PNH clone and severity which have been calculated using either Fisher’s exact (f) or Chi-square test (c).
Anti-cox-2 Autoantibodies Are Rarely Observed In Other Patient Cohorts
Control samples were obtained from collaborating clinical centers and from biobanks (detailed sample information in Supplementary Tables 1–2). All healthy controls (n = 74) and non-hematological patients without autoimmune conditions (n = 154) were tested negative for aCOX-2 Ab (Fig. 2A). Similarly, all tested patients with RA (n = 51), graft versus host disease (GVHD) (n = 56), and patients with other autoimmune diseases (n = 30) were aCOX-2 Ab negative, while sporadic aCOX-2 Ab positive cases were identified among multiple sclerosis (MS, n = 2/98, 2%) and type 1 diabetes (DM1, n = 2/44, 5%).
Clinically, IAA shares characteristics with many related hematological disorders such as hypoplastic myelodysplastic syndrome (MDS), LGLL, and pure red cell aplasia (PRCA). All LGLL (n = 68) and PRCA (n = 12) patients were aCOX-2 Ab negative. Some aCOX-2 Ab positive cases were identified among patients with MDS (n = 2/80, 3%) and idiopathic thrombocytopenia (ITP, n = 4/105, 4%). Nineteen percent (n = 3/16) of PNH patients were tested aCOX-2 seropositive.
Anti-COX-2 Ab is associated with older age, the HLA-DRB1*15:01 genotype, and lower platelet counts at diagnosis
For logistic regression analysis, we selected adult patients with less than 30% missing data values and with HLA genotype and treatment information available (n = 276). Missing data points were imputed and the multivariate analysis revealed a significant association between HLA-DRB1*15:01 genotype and aCOX-2 Ab positivity with odds ratio (OR) reaching 14.96 (CI 6.40-34.98, p < 0.001) (Table 2, Fig. 2B). Adult aCOX-2 Ab positive IAA patients were also confirmed to be older (OR 1.34, CI 1.16–1.55, p < 0.001) than the autoantibody negative IAA patients. The square of age had a statistically significant OR (p = 0.004) over one, which suggests that the probability of being aCOX-2 Ab positive increased non-linearly. There was also a statistically significant association between aCOX-2 Ab positivity and platelet count at diagnosis (OR 1.34 CI 1.16–1.55, p < 0.001), aCOX-2 Ab positive IAA patients displaying lower platelet counts than the aCOX-2 Ab negative IAA patients (Fig. 2C). No differences were detected between the aCOX-2 Ab positive and negative patients in disease severity or in treatment responses to IST (Fig. 2D) or eltrombopag (Fig. 2E).
Table 2
Logistic regression analysis of the clinical parameters (adult IAA patients n = 276)
|
|
aCOX-2 Ab negative
|
aCOX-2 Ab positive
|
OR (univariable)
|
OR (multivariable)
|
OR (multivariable. MI)
|
HLA-DRB1*15:01
|
Absent
|
115 (71.4)
|
18 (18.0)
|
-
|
-
|
-
|
|
Present
|
46 (28.6)
|
82 (82.0)
|
11.39 (6.29–21.56. p < 0.001)
|
11.34 (3.55–43.47. p < 0.001)
|
14.96 (6.40-34.98. p < 0.001)
|
Gender
|
Female
|
82 (48.5)
|
65 (60.7)
|
-
|
-
|
-
|
|
Male
|
87 (51.5)
|
42 (39.3)
|
0.61 (0.37–0.99. p = 0.048)
|
1.23 (0.33–4.80. p = 0.760)
|
0.77 (0.34–1.72. p = 0.516)
|
Age at dg (years)
|
Mean (SD)
|
38.9 (17.6)
|
61.4 (13.7)
|
1.08 (1.06–1.10. p < 0.001)
|
1.52 (1.22–1.98. p = 0.001)
|
1.34 (1.16–1.55. p < 0.001)
|
Age at dg^2
|
Mean (SD)
|
1821.9 (1636.7)
|
3956.6 (1612.1)
|
1.00 (1.00–1.00. p < 0.001)
|
1.00 (0.99-1.00. p = 0.006)
|
1.00 (1.00–1.00. p = 0.004)
|
PNH clone
|
Absent
|
78 (53.4)
|
41 (42.3)
|
-
|
-
|
-
|
|
Present
|
68 (46.6)
|
56 (57.7)
|
1.57 (0.94–2.64. p = 0.089)
|
1.02 (0.26–3.84. p = 0.981)
|
1.18 (0.51–2.73. p = 0.698)
|
Severity
|
Moderate
|
41 (26.5)
|
28 (28.0)
|
-
|
-
|
-
|
|
Severe
|
101 (65.2)
|
61 (61.0)
|
0.88 (0.50–1.58. p = 0.676)
|
1.92 (0.27–15.57. p = 0.523)
|
1.17 (0.37–3.68. p = 0.782)
|
|
Very severe
|
13 (8.4)
|
11 (11.0)
|
1.24 (0.48–3.17. p = 0.654)
|
2.09 (0.11–43.76. p = 0.624)
|
1.49 (0.26–8.58. p = 0.657)
|
Hb at dg (g/dl)
|
Mean (SD)
|
8.9 (2.1)
|
9.5 (10.1)
|
1.02 (0.98–1.09. p = 0.454)
|
1.36 (0.93–2.05. p = 0.122)
|
1.08 (0.90–1.29. p = 0.412)
|
WBC at dg (109/l)
|
Mean (SD)
|
2.4 (1.5)
|
2.5 (1.3)
|
1.05 (0.89–1.25. p = 0.536)
|
2.46 (0.83–7.64. p = 0.103)
|
1.18 (0.59–2.38. p = 0.637)
|
Plt at dg (109/l)
|
Mean (SD)
|
40.8 (46.4)
|
21.6 (16.8)
|
0.97 (0.96–0.98. p < 0.001)
|
0.98 (0.94–1.01. p = 0.219)
|
1.34 (1.16–1.55. p < 0.001)
|
ANC at dg (109/l)
|
Mean (SD)
|
0.9 (1.0)
|
0.9 (0.7)
|
0.90 (0.66–1.20. p = 0.499)
|
0.28 (0.06–1.13. p = 0.077)
|
0.64 (0.26–1.59. p = 0.336)
|
ALC at dg (109/l)
|
Mean (SD)
|
1.2 (0.8)
|
1.4 (0.7)
|
1.37 (0.94-2.00. p = 0.102)
|
1.08 (0.31–3.82. p = 0.901)
|
0.99 (0.38–2.55. p = 0.980)
|
Logistic regression analysis for adult (>18 years old) IAA patients. For categorical variables numbers of patients together with the proportion of patients from aCOX-2 Ab negative and positive cohorts are reported. For continuous variables mean values together with standard deviations (SD) are reported. Abbreviations: Dg, diagnosis; PNH, paroxysmal nocturnal hemoglobinuria; Hb, hemoglobin; WBC, white blood cells; Plt, platelets; ANC, absolute neutrophil counts; ALC, absolute lymphocyte counts; SD, standard deviation; OR, odds ratio; MI, multiple imputation.
Univariate analysis of the whole cohort (n = 405) returned a modest, but statistically significant OR 2.02 (1.28–3.22, p = 0.003) for the presence of the PNH clone in the aCOX-2 Ab positive IAA patients (Supplementary Table 3). This difference, however, was not confirmed in the multivariate analyses, implying that the presence of the PNH clone is cofounded with some other variable e.g., the HLA-DRB1*15:01 genotype.
Acox-2 Ab Assay Has Excellent Specificity And Sensitivity In Older Patients With Hla-drb1*15:01 Genotype
Next, IAA patients were split into HLA-DRB1*15:01 positive and negative groups. The age-dependent increase in aCOX-2 Ab positivity was even sharper when the analysis was restricted to the HLA-DRB1*15:01 positive IAA patients (Fig. 2F). Similarly, test performance indicators were markedly improved when the analysis was restricted to HLA-DRB1*15:01 positive, adult (> 18 years or > 40 years old) IAA patients. The overall test specificity was 98%, and sensitivity reached 83% in > 40 years old HLA-DRB1*15:01 positive IAA patients (Table 3). There was also a clear improvement in in false negative rate (FNR, from 0.68 to 0.17), and in Area Under ROC Curve (AUC, from 0.65 to 0.91) only > 40 years old, HLA-DRB1*15:01 positive IAA patients were included in the analysis.
Table 3
Diagnostic test parameters
|
All IAA
|
> 18 yo IAA
|
> 40yo IAA
|
All
DRB1*15:01 pos IAA
|
> 18yo
DRB1*15:01 pos IAA
|
> 40yo
DRB1*15:01 pos IAA
|
Number of patients
|
405
|
334
|
208
|
174
|
157
|
113
|
Accuracy
|
0.74
|
-
|
-
|
-
|
-
|
-
|
Inaccuracy / Error rate
|
0.26
|
-
|
-
|
-
|
-
|
-
|
Sensitivity
|
0.32
|
0.37
|
0.55
|
0.56
|
0.63
|
0.83
|
95% CI
|
0.28–0.37
|
0.32–0.43
|
0.50–0.63
|
0.49–0.63
|
0.55–0.70
|
0.76–0.90
|
Specificity
|
0.98
|
-
|
-
|
-
|
-
|
-
|
95% CI
|
0.97–0.99
|
-
|
-
|
-
|
-
|
-
|
Youden's Index
|
0.30
|
-
|
-
|
-
|
-
|
-
|
False Positive Rate (FPR)
|
0.02
|
-
|
-
|
-
|
-
|
-
|
False Negative Rate (FNR)
|
0.68
|
0.63
|
0.45
|
0.44
|
0.37
|
0.17
|
Positive Likelihood Ratio (LR+)
|
16.9
|
-
|
-
|
-
|
-
|
-
|
Negative Likelihood Ratio (LR-)
|
0.69
|
-
|
-
|
-
|
-
|
-
|
Positive Predictive Value (PPV)
|
0.91
|
-
|
-
|
-
|
-
|
-
|
Negative Predictive Value (NPV)
|
0.71
|
-
|
-
|
-
|
-
|
-
|
Predictive Summary Index (PSI)
|
0.62
|
-
|
-
|
-
|
-
|
-
|
Diagnostic Odds Ratio (DOR)
|
24.4
|
-
|
-
|
-
|
-
|
-
|
Area Under ROC Curve (AUC)
|
0.65
|
0.68
|
0.77
|
0.77
|
0.81
|
0.91
|
Standard Error
|
0.02
|
-
|
-
|
-
|
-
|
-
|
95% CI
|
0.61–0.69
|
-
|
-
|
-
|
-
|
-
|
Descriptive parameters to define the usefulness of aCOX-2 Ab test as a novel diagnostic marker. Parameters were calculated for all IAA patients, to adult patients >18 years old (yo) and to patients > 40 years of age. All age groups were analyzed both with and without restriction to the HLA-DRB1*15:01 genotype. The age and genotype information were not available for all controls, thus, many of the values are only presented for the whole cohort. Mathematical equations used for the calculations are given in Supplementary Table 4.
To confirm the persistence of aCOX-1 autoantibodies in follow-up, we measured the autoantibody levels in 21 patients who received IST as their first line treatment (Fig. 2G). In most IAA patients, the autoantibody levels remained stable during a 24–36-month follow-up. In two patients, a transient drop in the aCOX-2 Ab levels was detected at 6 months together with a partial or complete treatment response to IST. The antibody levels returned above aCOX-2 Ab assay positivity threshold by 12 months. In a 3rd patient, a similar decrease was observed at 6 months and no follow-up samples were available. In a separate cohort of 9 patients, from whom follow-up samples were available at random timepoints after the initial diagnosis, aCOX-2 Ab levels were stable in all patients during the follow-up (Supplementary Fig. 4).
Acox-2 Autoantibodies Bind The C-terminal Part Of Cox-2
Denaturing SDS-PAGE electrophoresis followed by western blotting was performed with patients’ plasma samples, and it confirmed that in 9 out of 10 cases, aCOX-2 antibodies were able to bind denatured full-length recombinant COX-2 (Supplementary Fig. 5). To reveal the antigenic epitope in COX-2, a microarray-based linear peptide mapping platform was utilized. We identified an almost identical, DIN amino acid (D590-N592) signature containing C-terminal epitope in five of the ten tested aCOX-2 Ab positive patient plasma samples and additionally, other closely mapping C-terminal epitopes in two patients (Fig. 3A). The linear mapping method could not identify an antigenic epitope for three of the aCOX-2 Ab positive patients, and thus, they were subjected for conformational peptide screen (Fig. 3B; Supplementary Fig. 6). Conformational peptide mapping with cyclic peptides revealed two additional C-terminal COX-2 epitopes. All the identified epitopes are summarized in Fig. 3C.
The expression of PTGS2 mRNA
To understand in which cells PTGS2 (gene coding for COX-2) is expressed in human bone marrow, we reanalyzed a recently published single-cell RNA sequencing (scRNAseq) data set (33) including CD34 + cells from bone marrow samples from untreated AA patients (n = 15) and healthy controls (n = 2). Of the identified cell phenotypes (Fig. 4A), The highest PTGS2 expression levels were found in the granulocyte-monocyte/common monocyte precursor (GMP/CMP) and multipotent-progenitor /hematopoietic stem cell (MPP/HSC) clusters (Fig. 4B). Interestingly, the PTGS2 expression levels were significantly elevated in AA patients’ GMP/CMP cluster (Fig. 4C).