Prognostic value and computer image analysis of p53 in mantle cell lymphoma

P53 prognostic cut-off values differ between studies of mantle cell lymphoma (MCL), and its immunohistochemistry (IHC) interpretation is still based on semiquantitative estimation, which might be inaccurate. This study aimed to investigate the optimal cut-off value for p53 in predicting prognosis of patients with MCL and the possible use of computer image analysis to identify the positive rate of p53. We calculated p53 positive rate using QuPath software and compared it with the data obtained by manual counting and semiquantitative estimation. Survival curves were generated by using the Youden index and the Kaplan–Meier method. The chi-squared (χ2) test was used to compare MIPI, Ann Arbor stage, and cell morphology with p53. Spearman rank correlation test and Bland–Altman analysis were used to compare manual counting, computer image analysis and semiquantitative estimation, as well as the consistency between different observers. The optimal cut-off value of p53 for predicting prognosis was 20% in MCL patients. Patients with p53 ≥ 20% had a significantly worse overall survival (OS) than those with p53 < 20% (P < 0.0001). MCL patients with MIPI intermediate to high risk, Ann Arbor stage III–IV, and blastoid/pleomorphic variant cell morphology had more p53 ≥ 20%. There was a strong correlation between computer image analysis and manual counting of p53 from the same areas in MCL tissues (Spearman’s rho = 0.966, P < 0.0001). The results of computer analysis are completely consistent between observers, and computer image analysis of Ki-67 can predict the prognosis of MCL patients. MCL patients with p53 ≥ 20% had a shorter OS and a tendency for MIPI intermediate to high risk, Ann Arbor stage III–IV, and blastoid/pleomorphic variant. Computer image analysis could determine the actual positive rate of p53 and Ki-67 and is a more attractive alternative than semiquantitative estimation in MCL.


Introduction
Mantle cell lymphoma (MCL) is an aggressive B-cell non-Hodgkin lymphoma (NHL) and comprises approximately 3 to 10% of NHLs [1][2][3]. The clinical prognoses of MCL are heterogeneous, and risk stratification is based on the clinical indices comprising the mantle cell lymphoma prognostic index (MIPI) and a combined biological MIPI, including Ki-67 [4][5][6]. However, risk stratification might be further improved by biological markers, such as p53. P53 protein expression might serve as an alternative marker for TP53 gene mutations [7][8][9] and has been confirmed to be of prognostic value independent of the biological MIPI score in multiple MCL studies [7,[10][11][12]. However, the prognostic cut-off values of p53 differ in different studies; thus, its cut-off value use is still controversial. Furthermore, p53 immunohistochemistry (IHC) interpretation is still based on semiquantitative estimations that are inaccurate due to significant interobserver variability. Therefore, finding practical and efficient ways to precisely interpret the positive rate of p53 with less inter-and intraobserver variability would be of great value for diagnosing MCL.
In this study, we investigated the prognostic cut-off value of p53 in MCL. Moreover, we applied computer image analysis software to calculate p53 in MCL patients and compared the data with those obtained from manual counting and semiquantitative estimation.

Patients
A total of 65 MCL specimens were obtained from the Department of Pathology, West China Hospital of Sichuan University from 2014 to 2021. All cases were newly diagnosed and reviewed by trained pathologists according to the 2016 World Health Organization (WHO) classification for tumors of hematopoietic and lymphoid tissues [13]. Anonymous data regarding sex, age, Ann Arbor stage, tumor tissue type, tumor cell morphology, and survival time were obtained retrospectively from the patients' medical records and telephone followups. All patients were followed up from the date of diagnosis to July 31, 2021. This study was approved by the Institutional Review Board of West China Hospital of Sichuan University (registration number: WCH2021-00,333). All recruited patients gave written informed consent following the Declaration of Helsinki.

IHC staining of p53 and Ki-67
The EliVision method was used for IHC staining of p53 [14]. Briefly, the paraffin blocks were cut into 4-μm sections and pretreated in ethylenediaminetetraacetic acid (EDTA) buffer at pH 9.0 for 20 min. Then, they were incubated with p53 at a dilution of 1:200 (clone DO-7; MXB Biotechnologies) or Ki-67 at a dilution of 1:150 (clone MIB1; DAKO) and 3,3′-diaminobenzidine (DAB) chromogen. Finally, they were counterstained with hematoxylin and rinsed with deionized water. The expression of p53 and Ki-67 was detected in the nuclei of tumor cells. Each slide was scanned by the Hamamatsu Digital Slide Scanner NanoZoomer 2.0-HT C9600-13.

Manual counting and computer image analysis of p53-positive cells and Ki-67-positive cells
Manual counting and computer image analysis of each image were performed using QuPath software (version 0.1.2) [15,16]. QuPath software offers the direct selection and annotation of representative tumor areas within scanned histopathological slides of MCL and provides a tool for manual counting and computer image analysis of p53-positive cells and Ki-67-positive cells within the representative area. Since the pathologist has access to digitalized histopathological slides, computer image analysis of p53-positive cells and Ki-67-positive cells only takes seconds to complete. Additionally, QuPath software can be used for image-based manual counting by the human eye and marking of positive and negative nuclei with red or blue circles, respectively (Fig. 1A). This image-based manual counting of the number and location of nuclei can provide the opportunity for direct are marked with red and blue circles. B Image-based computer image analysis. Positively stained (brown) and negative nuclei (blue) are marked with red and blue irregular circles comparison of identical images from independent observers, as well as direct comparison to computer image analysis of identical images.
Areas where lymphoma cells are densely populated were chosen, and hotspot areas were avoided. Images were extracted from the typical region, and identical areas were randomly selected for manual counting and computer image analysis. All digitalized images were manually counted by a pathologist with experience (observer 1) and a PhD student (observer 2). The randomly selected image was exported from each representative area and manually counted to obtain a minimum of 1000 cells counted in each case. To avoid selection bias, each image was counted as a whole. When using computer image analysis, two observers analyzed the same image separately.

Semiquantitative estimation of p53-positive cells and Ki-67-positive cells
Semiquantitatively estimated p53 and Ki-67 were performed by an experienced pathologist (observer 1) and a PhD student (observer 2). An experienced pathologist selected the representative areas.

Statistical analysis
Overall survival (OS) was calculated from the date of diagnosis of death or the last follow-up. The sensitivity and specificity of p53 were tested by assessing the area under the ROC curve (AUC). Furthermore, the Youden index and optimal cut-off value were calculated using the DeLong test [17]. Survival curves were generated using the Kaplan-Meier survival analysis method, and the log-rank test was used to examine differences in OS. Comparisons of cell morphology and p53 were conducted using the chisquared (χ 2 ) test. The Spearman rank correlation test was used to calculate the correlation between manual counting, computer image analysis, and semiquantitative estimation, as well as the consistency between different observers. Bland-Altman analysis was used for comparison of manual counting, computer image analysis and semiquantitative estimation [18,19]. The statistical analysis was performed with MedCalc 20.009 software (MedCalc Software Ltd), and P < 0.05 was considered statistically significant.

Patient characteristics
The cohort included 65 MCL patients. The male to female ratio was 2.61:1, and the median age was 63. Twenty (31%) patients presented with B symptoms. Thirty-eight (58%) patients were at Ann Arbor stage I-II and 27 (42%) were at stage III-IV. From these patients, 43 (66%) tumor tissue specimens were derived from lymph nodes, and others were derived from the gastrointestinal tract, tonsil, nasopharynx, bone marrow, etc. The tumor cell morphology of 36 (55%) patients was the classical type, and 29 (45%) patients had blastoid/pleomorphic variants. Follow-up data were obtained for all 65 patients, and 51 (78%) patients with MCL had died (Table 1).
Manual counting, computer image analysis, and semiquantitative estimation of p53 and their correlation with survival, MIPI, Ann Arbor stage, and cell morphology Figure 1 shows images based on manual counting (Fig. 1A) and computer image analysis (Fig. 1B) of identical images from a representative region. As shown in Fig. S1, strongly stained nuclei were defined as positive. The median p53 assessed by manual counting of at least 1000 tumor cells was 43.1%, and the mean was 45.0% (range 0-100%). The median p53 calculated by computer image analysis of the same areas by manual counting was 53.3%, with a mean of 48.2% (range 0.1-100%). The p53 acquired by semiquantitative estimation displayed a median of 40% and a mean of 39.4% (range 0-100%).
Manual counting, computer image analysis and semiquantitative estimation of p53 can predict OS. Each of these methods showed an AUC > 0.85 on ROC curves (P < 0.0001; Fig. 2, Table 2). The optimal p53 cutoff for predicting OS was determined for each method. For manual counting, the cut-off value was 16.20%, the sensitivity was 80.39%, and the specificity was 100%. For computer image analysis, the cut-off value was 17.66%, the sensitivity was 74.51%, and the specificity was 100%. For semiquantitative estimation, the cut-off value was 5.00%, the sensitivity was 76.47%, and the specificity was 92.86% ( Table 2). As manual counting and computer image analysis are more precise than semiquantitative estimation, we used the approximate value of manual counting and computer image analysis, 20%, as the cut-off value of p53 for subsequent analysis.
To assess the prognosis of these three methods, we performed survival analysis. The Kaplan-Meier survival curve analysis showed that patients with p53 < 20% had a significantly longer OS than those with p53 ≥ 20% (P < 0.0001, Fig. 3A-C). Moreover, patients with MIPI high risk had shorter OS than those with intermediate and low risk (P < 0.0001, Fig. 3D). When stratified by Ann Arbor stage, the OS of stage III-IV patients was significantly shorter than that of stage I-II patients (P < 0.0001, Fig. 3E). The OS in classical MCL patients was significantly longer than that in blastoid/pleomorphic MCL patients (P < 0.0001, Fig. 3F). Moreover, MIPI high-risk patients had less p53 < 20% than MIPI intermediate-and low-risk patients (P < 0.005, Table 3). Ann Arbor stage III-IV patients had more p53 ≥ 20% than stage I-II patients (P < 0.0001, Table 3). Patients tended to show blastoid/pleomorphic variant MCL when p53 expression was higher than 20% (P < 0.0001, Table 3).

Correlation between manual counting, computer image analysis and semiquantitative estimation
In the same areas, manual counting and computer image analysis of p53 showed a strong correlation (Spearman's rho = 0.966, P < 0.0001, Fig. 4A). There was a significant correlation between manual counting and semiquantitative estimation (Spearman's rho = 0.938, P < 0.0001, Fig. 4B), albeit weaker than the correlation between manual counting and computer image analysis. Furthermore, a significant correlation was found between semiquantitative estimation and computer image analysis (Spearman's rho = 0.898, P < 0.001, Fig. 4C).
Bland-Altman plots comparing manual counting with the computer image analysis of identical areas revealed a tendency toward a higher evaluation of p53 by computer image analysis (Fig. 4D), and comparing manual counting p53 with semiquantitative estimation revealed a tendency toward a lower evaluation of p53 by semiquantitative estimation (Fig. 4E). We also used the Bland-Altman plot to compare computer image analysis with the semiquantitative estimation of identical areas. The results revealed a tendency toward a lower evaluation of p53 by semiquantitative estimation (Fig. 4F).

Validation of computer image analysis
To validate the repeatability of computer image analysis, different observers (observer 1 and observer 2) analyzed the same image via manual counting, computer image analysis, and semiquantitative estimation to compare the bias caused by different observers. Meanwhile, to verify the accuracy of computer image analysis, we also analyzed Ki-67-positive cells with manual counting, computer image analysis and semiquantitative estimation and examined differences in OS. Manual counting revealed a very low interobserver variability (Spearman's rho = 0.9765, P < 0.0001, Fig. 5A). Of interest, when two observers analyzed the same image by computer image analysis, the results showed complete consistency (Spearman's rho = 1, Fig. 5B). When using semiquantitative estimation, there was a significant correlation between observer 1 and observer 2 (Spearman's rho = 0.8790, P < 0.0001, Fig. 5C), which was weaker than manual counting and computer image analysis.
The prognostic value of Ki-67 in MCL and its cut-off value for predicting prognosis have been well studied [5,20]. Therefore, Ki-67 was analyzed in MCL patients by manual counting, computer image analysis and semiquantitative estimation, and the cut-off value of Ki-67 was defined as 30%. The survival curve showed that the OS of patients with Ki-67 < 30% was significantly longer than that of patients with Ki-67 ≥ 30% (P < 0.0001, Fig. 5D, F).

Discussion
In this study, we demonstrated that 20% could be used as the optimal cut-off value for determining the prognosis of MCL. MCL patients with p53 ≥ 20% had a shorter OS than those with p53 < 20%. In addition, patients with MIPI intermediate to high risk, Ann Arbor stage III-IV, and blastoid/pleomorphic variants tended to have p53 ≥ 20%. We also verified the possibility of using computer image analysis software to count p53 in patients with MCL. IHC staining for p53 is a promising tool for predictive purposes, as it serves as an alternative marker for TP53 mutations and 17p deletions [7][8][9]21]. In MCL, the prognostic significance of the immunohistochemical expression of p53 has been studied ( Table 4). Rodrigues et al. also used manual counting and computer image analysis for the immunohistochemical counting of p53 and prognosis analysis [7]. They defined 30% as the prognostic cut-off value of p53, which is based on a previous publication [11] and routine clinical practices at Lund University Hospital. However, the prognostic cut-off values of p53 in MCL in other studies are controversial, ranging from 1 to 50%, and the criteria for defining these cut-off values are unclear [10][11][12]22]. We used manual counting and computer image analysis to count the positive rate of p53 and calculated the cut-off values for predicting OS with the optimal specificity and sensitivity through the Youden index. Manual counting was 16.20%, and computer image analysis was 17.66%. To simplify clinical application, we used 20% as the cut-off value of p53 for survival analysis, and the results revealed that MCL patients with p53 ≥ 20% had a worse prognosis than those with p53 < 20%. In addition, we also found that p53 ≥ 20% was associated with MIPI intermediate to high risk, Ann Arbor stage III-IV, and blastoid/pleomorphic morphology, which is consistent with previous studies [12,21,23].
Manual counting of IHC is considered to be the most accurate interpretation method at present, but it is very time consuming and probably not applicable in routine clinical tests. In routine diagnostics, semiquantitative estimation replaces this time-consuming approach. However, semiquantitative estimation is less precise due to larger inter-and intraobserver variability. Thus, we compared three methods of p53 counting to define a simple and accurate method. Computer image analysis of p53 was more accurate in predicting prognosis and had a stronger correlation with manual counting than semiquantitative estimation. Computer image analysis of p53 could count larger tumor regions than manual counting, making it more representative in assessing the percentage of p53-positive cells within the tumor. Through evaluating p53-positive cells by different observers, we found that the results of   computer analysis were completely consistent between observers. The consistency between observers is also good when using manual counting, but it is slightly worse in semiquantitative estimation. At the same time, we used Ki-67 for verification. The results showed that computer image analysis of Ki-67 can predict the prognosis of MCL patients. Additionally, computer image analysis only takes seconds. It is faster than manual counting and creates repeatable results for retrospection.
The computer image analysis revealed a tendency toward slightly higher estimates of p53 compared to manual counting. Potential reasons for the overestimation of p53 may include positively stained background cells that were miscalled tumor cells and debris that was miscalled positive staining [24]. Therefore, computer image analysis could benefit from strict quality control of the IHC staining method.
The limitation of computer image analysis is the inability to distinguish between different cell types such as normal cells and tumor cells [25]. Therefore, pathologists must carefully select representative lymphoma regions. Once pathologists choose the representative region of lymphoma, this would not be a disadvantage in MCL because the reactive cell number is low in MCL [26].
Our study concluded that MCL patients with p53 ≥ 20% had a shorter OS and a tendency for the MIPI intermediate to high risk, Ann Arbor stage III-IV, and blastoid/pleomorphic variant. Furthermore, computer image analysis is more precise than semiquantitative estimation and can be applied to the interpretation of p53 and Ki-67 IHC staining in patients with MCL.