Cup-like nuclear invaginations (NIs) in acute myeloid leukemia (AML) blasts have been associated with NPM1 mutations. Precision for enumeration of NI blasts has not been previously studied. Furthermore, the sensitivity and specificity for the morphologic prediction of NPM1 mutations have been variously reported. By using 66 AML specimens (17 with NPM1 mutations and 49 without), we found that interobserver reproducibility for enumeration of NI blasts was high (r = 0.98) and that identification of this feature was teachable (r = 0.96). No NPM1 mutation–negative case had greater than 7% NI blasts. The fraction of NI blasts was highly variable among 17 NPM1 mutation–positive cases, ranging from 0% to greater than 40%. These data indicate that an NI blast fraction of more than 10% is highly specific for NPM1 mutation–positive cases but with a sensitivity of about 30%. Therefore, although NI blasts can be reliably identified in routine smears and although they are a specific marker of NPM1 mutation–positive cases, the majority of NPM1 mutation–positive cases lack this distinctive finding.
Acute myeloid leukemia
Somatic mutations in the nucleophosmin gene (NPM1) are the most common genetic aberration known in acute myeloid leukemia (AML).1 The mutations almost always involve a 4-base-pair insertion in a limited region of exon 12.2 These insertions cause a change in the reading frame, the destruction of the nucleolar localization signal, and the creation of a nuclear export signal, which results in the NPM1 protein relocalizing to the cytoplasm.3,4NPM1 mutations are associated with a normal karyotype, an internal tandem duplication in the juxtamembrane region of the FLT3 gene (FLT3-ITD), a higher WBC count in the peripheral blood, a higher fraction of blasts in the bone marrow, lower expression of the CD34 antigen, and female sex.1,5–8 In the absence of an FLT3-ITD mutation, NPM1 mutations correlate with improved outcome.1,5–8 AML with mutated NPM1 is a provisional entity in the current World Health Organization classification of myeloid neoplasms.9
NPM1 mutations were also found to associate with distinct nuclear morphologic features with prominent nuclear invaginations (NIs) called “cup-like” or “fish-mouth.”10 These morphologic features had been noticed previously in association with a distinctive immunophenotype (CD34–/HLA-DR–)11 and also with a high proportion of cases with the FLT3-ITD mutation.12 By using the inclusion criterion of NI in 10% or more of the blast population spanning at least 25% of the nuclear diameter in Wright-Giemsa–prepared smears, Chen et al10 noted that 64% had a mutated NPM1 gene compared with 20% in a set of matched control samples with French-American-British (FAB)-M1 morphology without NI. In a second study, Chen et al13 confirmed the association of cup-like nuclei with NPM1 mutations and also confirmed the association with FLT3-ITD initially reported by Kussick et al.12 Finally Kroschinsky and coworkers14 identified a set of cases with cup-like AML and also found an association with NPM1 and FLT3-ITD mutations. All of the studies found that NIs are also associated with less expression of HLA-DR and a normal karyotype.
Two of the articles on the association of NIs with mutations suggested that the finding of cup-like morphologic features might be useful in selecting patients for mutation analysis.13,14 In this study, we evaluated interobserver concordance in identifying AML with NIs and the sensitivity and specificity of this morphologic feature in predicting the presence of mutation in the NPM1 gene.
Materials and Methods
Samples from 81 patients were identified retrospectively from the University of Rochester Medical Center Molecular Diagnostic Laboratory, Rochester, New York. These 81 consecutive studies were obtained in the course of routine clinical laboratory evaluation for AML and were requested by the attending hematologists during a 3-year period. The study was approved by the University of Rochester Medical Center Institutional Review Board. NPM1 mutations were detected by using a fluorescence-based assay as previously described.15 This assay was designed to have an analytic sensitivity sufficient to detect an NPM1 exon 12 mutation in 10% of the analyzed cells. We excluded 15 cases because of missing or inadequate slides (4 cases) or because the cases were not de novo AML (11 cases).
Analysis proceeded in 3 steps. Before determining consensus criteria for the identification of NI blasts, 2 readers (W.R.B. and J.M.B.) performed independent and blinded counts of all cases. Next, consensus criteria were defined, and cases were again reviewed. Primary end points were the interobserver correlations with and without consensus criteria. To assess whether these criteria were teachable, a pathology trainee (J.P.) subsequently repeated the study.
The characteristics of the NPM1+ specimens are listed in Table 1. Of 66 AML specimens, 17 (26%) had mutation of the NPM1 gene: 11 type A, 3 type B, 1 type D, 1 type J (using the nomenclature in reference 2), and 1 novel mutation (c.864_865delGCinsCTGGCG).15 Of 17 cases positive for the NPM1 mutation, 13 (76%) had a normal karyotype. A range of morphologic features were seen: 5 had a monocytic predominance (FAB-M4/M5); 4 were without significant maturation (FAB-M1). FLT3-ITDs were identified in 5 of 17 patients with NPM1 mutation. In 15 of 17 cases, complete remission was achieved; 6 cases have not relapsed to date (median follow-up, 18+ months). The remaining 49 cases were negative for NPM1 mutations.
To assess the need for more rigorously defined morphologic criteria for the identification of blasts with NI, these 66 cases were scored for cup-like nuclear indentations by 2 of us (J.M.B. and W.R.B.). The reviewers had no knowledge of the NPM1 status for each case or the fraction of cases known to be positive for the mutation in the entire collection. No effort was made to create a morphologic definition beyond that accessible in the literature as defined in the introduction. Typical NI blasts are shown in Image 1. The 2 reviewers identified similar fractions of NI blasts (range, 0%–54% vs 0%–62%, with a median of 2% for each). In 65 of 66 cases, the fractions of NI blasts determined by each reviewer were within the 95% confidence interval (CI) for the determination of the proportion by the other reviewer (based on a differential of 100 blasts). The disparity for the 1 case with nonoverlapping CIs in the initial review (4% [95% CI, 2%–10%] vs 17% [95% CI, 11%–26%]) was not reproducible with differential counts of more blasts, with both reviewers obtaining fractions less than 10% for this case when an additional 100 blasts were counted. Both reviewers identified the same 2 cases with significantly more than 10% NI blasts (case 4, 54% and 62%; case 12, 18% and 37%). Neither reader enumerated significantly more than 10% NI blasts in any other case.
While there was general agreement regarding the scoring for NI blasts between the 2 reviewers, with an interobserver correlation coefficient of 0.95, the statistical significance of this correlation depended exclusively on the 2 cases with more than 10% blasts. Elimination of the 2 cases with the highest fraction of NI blasts reduced the correlation coefficient to 0.2.
This blinded review without a prior consensus definition of NI blasts identified only 2 of 17 patients with NPM1 mutations (the 2 cases described above). For the remaining 15 of 17 cases shown to have an NPM1 mutation, both reviewers had scored less than 10% NI blasts. Both reviewers scored less than 10% NI blasts for 49 of 49 patients whose specimens were negative for NPM1 mutations.
A possible cause for the failure to sensitively identify NPM1-associated morphologic features could be the lack of consensus criteria that results in interobserver variability. We used the following criteria to define a blast with NIs: The indentation must appear as though a spherical object had compressed an otherwise round or slightly ovoid nucleus, and the compression must be greater than 25% of the diameter of the nucleus (eg, the left panel of Image 1). To assess the reproducibility of these consensus criteria for the identification of NI blasts, the 17 cases positive for NPM1 mutations were reviewed again. While each reviewer knew that all cases were from patients with the NPM1 mutation, the results of prior interpretation and score were unknown because each case was coded, and the reviewers worked independently.
In the specimens known to have an NPM1 mutation, the range of blasts with NIs was 0% to 50%, with an arithmetic mean of 8% and a median of 3% (Table 1). With these criteria, the linear correlation coefficient for the fraction of NI blasts between the 2 observers was 0.98 (P ≤ .0001) Figure 1. This highly significant linear correlation cannot be explained by a few outlier values: elimination of the 2 cases with the highest fractions of NI blasts still yields a correlation coefficient of 0.91 (P ≤ .0001). The interobserver correlation using the consensus criteria is greater than the intraobserver correlation comparing each reviewer’s findings before and after adoption of the consensus criteria (r = 0.9 and r = 0.85 for reviewers 1 and 2, respectively).
Interobserver reproducibility of enumeration of blasts with nuclear invagination (NI). The graph shows a comparison of the percentage of blasts with NI in the 17 patients with acute myeloid leukemia who had an NPM1 mutation.
To assess the practicality of this definition, a pathology resident (J.P.) repeated the scoring of these 17 cases with NPM1 mutations using the consensus definition of a blast with NIs. The results were highly correlated with the scoring of the 2 original reviewers (r = 0.95 for the trainee’s scoring vs the mean of scorings by the original reviewers). This high correlation suggests that this particular definition would be sufficiently reproducible for clinical practice.
Receiver operating characteristic curve for identification of the NPM1 mutations by the percentage of nuclear invagination blasts. The cutoff values resulting in each paired specificity/sensitivity are shown in the squares.
Next, given the reproducibility of this definition, we applied it to all 66 study specimens. Neither reviewer identified more than 6% NI blasts in any of the 49 cases negative for an NPM1 mutation. These data suggest that 6% cells with cup-like nuclear indentations is the upper limit for AML cases that lack the NPM1 mutation.
The sensitivity and specificity for various percentages of NI blasts for the identification of cases with NPM1 mutations are summarized in the receiver operating characteristic (ROC) curve Figure 2. The data show that a cutoff of 7% NI blasts provided excellent specificity but detected only about 30% of the NPM1 mutations in this set of patients. The ROC curve also indicates that lowering the cutoff value of NI blasts to values less than 7% results in an abrupt loss of specificity without any gain in sensitivity.
The recent discovery of a high correlation of NPM1 mutations with distinct nuclear morphologic features described as cup-like or fish-mouth suggests that morphologic features may be helpful in predicting molecular alterations that are prognostically significant.10,13,14 This prompted us to assess the reliability of observer concordance in the evaluation of this aspect of morphologic features and also to evaluate the sensitivity and specificity of the identification of such cells in predicting mutation of the NPM1 gene.
Before a morphologic observation is widely implemented, it ought to be shown that it can be reliably discerned with good concordance between trained observers. We have found that enumeration of cup-like or fish-mouth nuclear indentations is highly reproducible between experienced hematopathologists and is not difficult to learn.
However, we have also found that although there is a clear relationship between NI blasts and NPM1 mutations, this association is not strong enough to be clinically useful. NI blasts are not specific to AML with NPM1 mutations. For example, the images shown in Image 1 to depict a “classic” NI blast are from a case that lacked the NPM1 mutation. Our data suggest that scoring NI blasts has limited usefulness in the detection of cases with NPM1 mutations. The ROC analysis shows that when the number of NI blasts exceeds 10%, it is highly likely that an NPM1 mutation is present. However, the presence of 1% to 6% NI blasts was found in half of the cases negative for NPM1 mutations (Figure 2); these NI blasts were morphologically indistinguishable from NI blasts in cases positive for the NPM1 mutation (Image 1).
Spurred by the lack of sensitivity of this definition, we attempted to identify alternative morphologic features in cases that were positive for NPM1 mutations (data not shown). Two of us (J.M.B. and W.R.B.) attempted several alternative definitions of cup-like or fish-mouth nuclei, such as allowing smaller indentations or deeply cleaved nuclei to be included as NI blasts (eg, the right panel of Image 1). However, none of these alternative definitions seemed to provide a significant correlation with NPM1 mutations.
In our survey, a high fraction (>6%) of NI blasts does identify cases with NPM1 mutations. However, the majority of cases with NPM1 mutations will not have more NI blasts than cases without NPM1 mutations. Our findings are consistent with earlier reports, which found an association of NI blasts with NPM1 mutations.13,14 However, the prior studies focused on the association itself rather than investigating the usefulness of this association for diagnosis. We conclude that the association of NPM1 mutations with distinct morphologic features is valid for only the occasional case: for most cases, it cannot be substituted for genetic analyses for NPM1.
Mutations in nucleophosmin (NPM1) in acute myeloid leukemia (AML): association with other gene abnormalities and previously established gene expression signatures and their favorable prognostic significance. Blood. 2005;106:3747–3754.
John M.Bennett, JenniferPryor, Todd S.Laughlin, Paul G.Rothberg, W. RichardBurackAm J Clin Pathol(2010)134 (4):
648-652DOI: http://dx.doi.org/10.1309/AJCPULO8SLW0RKJLFirst published online: 1 October 2010 (5 pages)