Introduction

Mantle cell lymphoma (MCL) is a mature B cell lymphoma with small- to medium-sized tumor cells and CCND1 translocation [1]. In most MCLs, t(11;14)(q13:q32) translocation between IGH gene and CCND1 gene results in overexpression of cyclin D1 [1]. Usual immunophenotype of MCL is BCL2, CD5, and cyclin D1-positive and BCL6, CD10, and CD23-negative [1]. However, MCLs with aberrant phenotypes, including CD5-negative or CD10 and BCL6-positive cases, were also reported [1]. In addition, there are cyclin D1-negative MCLs that overexpress cyclin D2 or cyclin D3 [1]. The diagnosis of cyclin D1-negative MCLs is performed with SOX11 staining, which is positive for most MCLs, regardless of cyclin D1 overexpression [2].

The transcription factor SOX11 is encoded by SOX11, which is a member of the SOX gene family [3]. The nuclear expression of SOX11 was found to be specific to MCL, compared to that in other types of lymphomas [4]. However, the expression seemed to vary widely with respect to antibodies, and some studies suggested the performance of antibodies or different cut-off values across studies as the reasons of variation in the results [1, 5, 6]. In addition, MCLs with indolent clinical courses, including leukemic non-nodal MCL, small cell variant MCL, and MCLs with aggressive behavior (such as pleomorphic and blastoid variants), tended to show low expression rate of SOX11 [1, 5, 7, 8]. Despite the varied expressions, objective evaluation of the possible causes has not been performed yet.

In this meta-analysis, we aimed to evaluate the factors causing variation in reported SOX11 expression using subgroup analysis and meta-regression.

Materials and methods

Published studies and selection criteria

We searched PubMed, EMBASE, and Cochrane library through May 9, 2018, with the following key words: “SOX11” and (“lymphoma” or “lymphomas”). Reference lists of review articles were also searched. Duplicate data and articles were excluded considering the authors and their affiliations. Original articles were included if SOX11 immunohistochemistry was performed in human MCL cases. When multiple articles from an author or institution were found; the most informative article was selected for the current study. Non-English articles, article or conference abstracts without sufficient information for meta-analysis, review articles, case reports, comments, errata, articles on cell line or animals, and those concerning SOX11 studies with methods other than immunohistochemistry were excluded. The selection process is shown in Fig. 1.

Fig. 1
figure 1

Flow diagram of study selection

Data extraction

The following data from all eligible studies were extracted [5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25]: the first author’s name, year of publication, number of SOX11-positive MCLs, number of total MCLs, type of MCLs, species and clone of SOX11 antibody, proportion of MCL subtypes (cases with indolent characteristics, and aggressive variants), and cut-off values of SOX11. For cases with indolent characteristics, the following were included: leukemic non-nodal MCL, small cell variant MCL, and MCL with indolent clinical course.

Statistical analyses

All data were analyzed using R version 3.4.3, with “meta” package [26, 27]. We investigated logit-transformed proportion of SOX11-positive MCLs as the effect size of each study. Based on random effect models, statistical heterogeneity was evaluated using Higgins’ I2 statistics. For I2 value greater than 40%, the studies in meta-analysis were considered heterogeneous [28]. Subgroup analysis was performed by setting the species and clone of the antibodies as moderators. Meta-regression analyses were performed with proportion of indolent cases, proportion of aggressive variants, and cut-off values as covariates. Residual heterogeneity, which could not be explained by the covariates used in the meta-regression, was also considered present for I2 > 40%. Publication bias was examined by Egger’s test, Thompson’s test, rank test, and the test for funnel plot asymmetry based on a linear regression model [29].

Results

Selection and characteristics of the studies

Three hundred and eighty-three reports were identified in the database search. A total of 21 studies fulfilled the inclusion criteria [5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25]; all were case-control studies. Four studies used more than one antibody [5, 14, 16, 18]. Rabbit polyclonal antibodies were used for eight study populations [5,6,7,8,9, 16, 24]. Mouse monoclonal antibodies were used for 16 [5, 10,11,12, 14,15,16,17,18,19, 22, 25]. Goat polyclonal antibody was used for one [18]. Four studies did not specify the species of antibody used [13, 20, 21, 23]. Among the studies with mouse monoclonal antibodies, clone MRQ-58 was used on 10 study populations [5, 10,11,12, 14,15,16, 18, 19, 22]. MCLs with indolent traits (leukemic non-nodal MCL, small cell variant MCL, and MCL with indolent clinical course) were included in 10 study populations [7, 9, 11, 13, 16, 20, 24]. Aggressive variant MCLs (pleomorphic and blastoid variants) were included in 14 study populations [5, 8, 10,11,12,13,14, 16, 19, 20]. Cut-off value was specified in 14 study populations [6, 10,11,12, 14, 16, 19, 21, 22, 24]. The proportion of SOX11-positive cases ranged from zero to one (Table 1). For all studies, meta-analysis was performed by using a random effect model.

Table 1 Characteristics of studies reporting SOX11 immunohistochemistry in mantle cell lymphoma

The combined proportion of SOX11-positive cases was 0.80 (95% CI = [0.72, 0.87]). Substantial heterogeneity was observed between the studies, and it means that the reported SOX11-positive rates in MCLs are highly heterogeneous (I2 = 83%) (Fig. 2). To explore sources of heterogeneity, we considered subgroup analysis and meta-regression as discussed in the following sections. Before conducting the analyses, we checked whether there was an evidence of publication bias in our collection of studies. The funnel plot did not show strong evidence of publication bias (Fig. 3), and Egger’s test (p = 0.3091), Thompson’s test (p = 0.2426), and rank test (p = 0.3677) showed no significant result at 0.05 level [29].

Fig. 2
figure 2

Forest plot of all studies reporting SOX11 immunohistochemistry in mantle cell lymphoma

Fig. 3
figure 3

Funnel plot of meta-analysis for studies reporting SOX11 immunohistochemistry in mantle cell lymphoma. Individual studies are represented by small circles

Subgroup analysis

Subgroup analysis was done to see if any specific group could be the source of heterogeneity. In subgroup analysis, the studies were sectioned into subgroups, and heterogeneity of each was tested. Three categorical variables were considered in subgroup analysis. Due to the limited number of studies, a univariate approach was employed. The first categorical variable was the antibody clone or catalog number. All studies, except six without specified antibody clone or catalog number, were grouped according to their monoclonal antibody clone or polyclonal antibody catalog number [5, 7,8,9,10,11,12, 14,15,16,17,18,19, 22, 25]. HPA000536 and MRQ-58 groups (the number of studies ≥ 2) showed I2 > 65% (Fig. 4). MRQ-58 groups were expected to show low heterogeneity, but all groups showed high heterogeneity, thereby implying heterogeneity to be explored further.

Fig. 4
figure 4

Forest plot of subgroup analysis by monoclonal antibody clones or polyclonal antibody catalog numbers

The second categorical variable was the antibody clonality. All studies, except four without specified antibody species, were divided into two groups according to their antibody clonality, rabbit polyclonal, and mouse monoclonal [5,6,7,8,9,10,11,12, 14,15,16,17,18,19, 22, 24, 25]; subgroup analysis was done in an attempt to explain the source of heterogeneity. Both groups showed high heterogeneity (Fig. 5); the group of mouse monoclonal antibodies showed lower heterogeneity (I2 = 74%) than the group of rabbit polyclonal antibodies (I2 = 80%) as expected (Fig. 5).

Fig. 5
figure 5

Forest plot of subgroup analysis by clonality of antibodies

The third categorical variable was the clone of monoclonal antibodies. Studies with monoclonal antibodies were divided into two groups—those with clone MRQ-58 and the rest—and subgroup analysis was done to see if the source of heterogeneity could be explained by the monoclonal antibody clone. Both groups showed high heterogeneity. The clone MRQ-58 that was expected to show more homogeneous result actually showed lesser heterogeneity (I2 = 71%) than that of the rest (I2 = 81%) (Fig. 6).

Fig. 6
figure 6

Forest plot of subgroup analysis by clones among mouse monoclonal antibodies (“Clone2 = True” means MRQ-58)

All three subgroup analyses did not show any specific group without heterogeneity, and possible source of heterogeneity could not be found.

Meta-regression

Meta-regression was applied to test whether certain continuous variables could explain the heterogeneous result obtained by utilizing regression analysis method.

Three continuous covariates were considered in meta-regression (Table 2). Due to the limited number of studies, a univariate approach was employed. The first covariate was the proportion of cases with indolent traits. 10 study populations with specified cases of indolent traits were included in the meta-regression [7, 9, 11, 13, 16, 20, 24]. Results showed that the proportion of cases with indolent traits was a statistically significant covariate (intercept = 1.51, 95% CI = [0.71, 2.31], p = 0.0002, slope = − 2.87, 95% CI = [− 4.60, − 1.14], p = 0.0012). The residual heterogeneity was not substantial (I2 = 34.91%). This result implied statistically significant inverse relation between the proportion of indolent cases and heterogeneity of SOX11-positive rate in MCLs. In other words, the two factors, SOX11-positive rate and proportion of indolent cases, fit into statistically significant regression model. Since residual heterogeneity was low, the proportion of indolent MCL could be regarded as an important source of heterogeneity.

Table 2 Meta-regression with covariates of the proportion of indolent and aggressive cases, and of the cut-off value for SOX11

The second covariate was the proportion of aggressive cases. 15 study populations with specified aggressive cases were included in the meta-regression [5, 8, 10,11,12,13,14, 16, 19, 20]; it did not show statistically significant coefficient (intercept = 1.37, 95% CI = [0.42, 2.32], p = 0.0046, slope = 0.91, 95% CI = [− 2.35, 4.16], p = 0.58). The meta-regression showed substantial residual heterogeneity (I2 = 83.90%).

The third covariate was the cut-off value. 14 study populations with specified cut-off values were included in the meta-regression [6, 10,11,12, 14, 16, 19, 21, 22, 24]; it showed that the cut-off value was statistically significant (intercept = 2.94, 95% CI = [1.83, 4.06], p < 0.0001, slope = − 0.08, 95% CI = [− 0.14, − 0.03], p = 0.0037). The residual heterogeneity was substantial (I2 = 72.54%). It suggested the cut-off value as another statistically significant source of heterogeneity showing inverse relation. In other words, the two factors, SOX11 positive rate and cut-off value for SOX11 positivity, fit into statistically significant regression model. However, since the residual heterogeneity is still high, the cut-off value may not be the only source of heterogeneity.

Discussion

The present meta-analysis evaluated SOX11 immunohistochemistry results in MCL. The estimated overall proportion of SOX11-positive MCL was 0.80, with substantial heterogeneity. As shown by subgroup analysis, substantial heterogeneity still remained after accounting for the individual antibody, clonality of the antibodies, and a specific clone among monoclonal antibodies as moderators. Meta-regression showed that, unlike the aggressive MCLs, the indolent MCLs and the cut-off value for SOX11 immunohistochemistry explained heterogeneity.

We speculated that the possible causes of heterogeneity were antibodies with different performances, varied cut-off values in immunohistochemistry interpretation, and proportion of indolent or aggressive MCLs. In fact, different performances between rabbit polyclonal and mouse monoclonal antibodies or among different mouse monoclonal antibodies were suspected by researchers [6, 16, 18, 30]. Sensitivity and specificity of a MRQ-58 clone were regarded to be superior to that of other mouse monoclonal antibodies by some authors [5, 10, 14, 16, 18]. The group of MRQ-58 clone showed lesser heterogeneity than the groups of other monoclonal antibodies, though still noticeably high. Evaluating antibody performances was beyond the scope of this study. In addition, certain antibody might actually outperform others. However, at least, there had been no remarkable differences between antibodies with respect to the heterogeneity of SOX11-positive proportion in immunohistochemistry.

Proportion of indolent MCL cases was a statistically significant factor causing heterogeneity. Current WHO classification recognizes such indolent MCLs first, by the subtype category of leukemic non-nodal MCL and second, as a histological variant of the small cell MCL [1]. The WHO classification describes leukemic non-nodal MCL as SOX11-negative tumors [1]. Small cell variant MCLs are shown to be SOX11-negative by some researchers [7, 31]. Some classical MCLs with indolent clinical behavior, but neither leukemic non-nodal nor small cell MCL, are reported to be SOX11-negative [9, 31]. Regardless of histopathological classification, they shared small cell histology, negative SOX11 staining, and indolent clinical behavior. Since most of the MCLs with indolent traits were SOX11-negative, we expected the indolent MCLs to be an important factor contributing to the heterogeneity in the overall analysis of SOX11 immunohistochemistry, as also supported by our meta-regression results.

Proportion of aggressive MCL cases did not affect heterogeneity among the studies. Studies on SOX11 expression in MCL reported some pleomorphic and blastoid MCLs to be SOX11-negative [4, 5, 8, 19]. Therefore, there was a possibility that the proportion of aggressive MCL cases, included in the studies considered for this meta-analysis, caused heterogeneity. However, the proportion of aggressive MCLs was not shown to be a statistically significant cause of heterogeneity.

SOX11-negative aggressive MCLs might cause potential confusion with cyclin D1-positive diffuse large B cell lymphoma (DLBCL), since there is no clear definition regarding such cases with aggressive histological feature, positive cyclin D1, and negative SOX11 immunohistochemistry. Currently, the main available choice in diagnostic approaches in such cases is the CCND1 translocation test [13, 32]. However, many of cyclin D1-negative MCLs lacked CCND1 translocation, and harbored CCND2 or CCND3 translocations instead [33, 34]. In this context, SOX11 can be used as a tumor marker, and one study included in this meta-analysis actually used SOX11 as a determining factor [10]. With the emergence of SOX11 as a relatively new diagnostic marker, the definition of DLBCL, MCL, and their subtypes might need to be revisited.

The cut-off value used for interpretation of SOX11 immunohistochemistry was suspected to be a potential source of variable SOX11 immunostaining, as seen in a previous report [5] as well as suggested by our meta-regression. Since the different cut-off values adopted by each author cause heterogeneity, a single well-defined cut-off threshold for SOX11 immunohistochemical interpretation might provide consistent results for SOX11 immunostaining in MCL.

Regarding the cut-off value, SOX11 mRNA and protein expression had been correlated in several earlier studies [5, 14, 17, 18, 30, 31, 35]; most of them showed good correlation between RT-PCR and immunohistochemistry results [14, 17, 30, 31, 35]. One study had reported that SOX11 mRNA-negative cases showed complete absence of immunohistochemistry-positive cells, whereas mRNA-positive cases showed at least weak immunostaining in most cells [18]. Another study proposed that weak and variable SOX11 staining may be regarded as positive [5]. Based on these studies, the possible optimal cut-off for SOX11 immunohistochemistry should be low. However, since even normal lymph node showed a few scattered positive cells, extremely low number of positive cells might best be ignored [5].

Our current meta-analysis was limited by the relatively small number of published studies considered. Especially, a meta-regression, using the three covariates together, could not be performed since only few articles recorded all the covariates. In the near future, it will be necessary for researchers to make their studies more reproducible by providing full details on the dataset.

In conclusion, this meta-analysis of SOX11 immunohistochemistry in MCL suggested the indolent MCLs and the cut-off value as important sources of overall heterogeneity, whereas antibody and aggressive cases were not. Although SOX11 is recognized as a marker for MCL, its practical application is not as well-accepted as cyclin D1. Defining cut-off values for SOX11 immunohistochemistry in MCL is expected to facilitate robust results and wider use of the antibody in the diagnosis of lymphoma.