Metaanalysis of the significance of matrix metalloproteinases for lymph node disease in patients with head and neck squamous cell carcinoma

Authors


Abstract

BACKGROUND

The objective of the current metaanalysis was to evaluate the expression patterns of matrix metalloproteinases (MMPs) in patients with head and neck squamous cell carcinoma (HNSCC), to evaluate reported series, and to determine whether there is an expressed value to quantitate the risk of metastasis.

METHODS

A review of the published literature was conducted according to defined selection criteria. Fixed and random effects models were applied for estimation of the summarized odds ratio and 95% confidence intervals, including a test for homogeneity of the odds ratios of the studies. Finally, forest plots were created to allow for visual comparison of the results and an estimation of heterogeneity.

RESULTS

The heterogeneity of data collection and statistical methods did not allow final judgments on the significance of immunohistochemical MMP expression analysis in patients with HNSCC or the impact of MMPs in predicting metastatic behavior. Fourteen studies with 710 patients for 5 different MMPs (MMP-1, MMP-2, MMP-3, MMP-9, and MMP-14) were included in the current metaanalysis. The results indicated that MMP-2, MMP-3, MMP-14 possibly played a role in the metastatic behavior of HNSCC tumors.

CONCLUSIONS

The authors recommended the standardization of staining procedures and evaluation protocols as a necessary step to allow for valid comparisons of the multitude of results published by different study groups. Cancer 2005. © 2005 American Cancer Society.

The poor prognosis of many patients who suffer from head and neck squamous cell carcinomas (HNSCC) is associated with the high lymphogenic metastatic spread of this tumor entity. Distant metastases are seen rarely without prior lymphogenic metastatic spread.1 Although many different histologic and molecular markers have been evaluated regarding their impact on lymphogenic metastatic spread,2 to date, none of these markers has appeared useful in clinical diagnosis. Since their first description, matrix metalloproteinases (MMPs) have been discussed widely concerning their possible involvement in tumor progression and metastatic behavior.3 Due to the heterogeneity of data collection and statistical methods, no final judgment can be drawn on the value of specific MMP family members in HNSCC tumor progression. The objective of the current study was to evaluate MMP expression patterns in patients with HNSCC in a metaanalysis in an effort to determine a single, exact value for the quantitative prediction of risk.

MATERIALS AND METHODS

Literature Review

A literature search was performed utilizing the PubMed data base (http://www.ncbi.nlm.nih.gov/pubmed) searching for the terms “matrixmetalloproteinases”, “squamous cell carcinoma,” and “head and neck cancer.” Selection criteria were as follows: 1) male and female patients with HNSCC; 2) clinically or histologically lymph node negative (N0) or lymph node positive carcinoma of the neck, and 3) immunohistochemical analysis of > 1 MMP(s). Exclusion criteria were 1) missing absolute numbers on lymph node status, 2) exclusive statistical data on the correlation between MMP analysis and lymph node disease, and 3) analysis of MMPs in HNSCC without negative controls for immunohistochemical staining.

Statistical Analysis

The statistical procedures described by Whitehead and Whitehead4 were applied in this metaanalysis. The odds ratio with 95% confidence interval was calculated for each individual study. Odds ratios were checked by means of the “test for homogeneity of odds ratio” according to Zelen.5 The P value was evaluated as a descriptive measure, and small P values were accompanied by higher heterogeneity. Finally, a common odds ratio was calculated for all publications for each MMP according to Mehta et al.6 In a “fixed-effects model,” the only source of variability assumed is the coincidence variation within the study. In addition, the calculation of the odds ratio was performed based on the “random effects model.”4 This procedure allows the calculation of a summarized odds ratio in case of inhomogeneous studies. Finally, forest plots were created to allow a visual comparison of the results and an estimation of the heterogeneity. The analyses were performed using the statistical software packages StatXact 5 (Cytel Software Corporation, Cambridge, MA) and SAS (SAS Institute Inc., Cary, NC).

RESULTS

The literature search identified 29 publications on immunohistochemical analysis of MMP expression and metastatic behavior in patients with HNSCC. Due to a lack of original data, 14 publications were excluded from this metaanalysis, and 1 publication did not allow for evaluation of the odds ratio. Thus, 14 studies with a total of 710 patients could be included into the metaanalysis (Table 1). Among these, 196 patients (27.6%) had SCC of the larynx, 60 patients (8.4%) had SCC of the pharynx, 450 patients (63.4%) had SCC of the oral cavity and neighboring structures, and 4 patients (0.6%) had SCC of the nasal cavity.

Table 1. Studies Cited in the Current Review Presented in Alphabetical Order
StudyNo. of patientsTNM determinationTumor statusLymph node statusStageExamined MMPs
T1/2T3/4N0N1–N3I/IIIII/IV
  1. MMP: matrix metalloproteinase; —: no data; c: clinically determined lymph node status; p: pathologically determined lymph node status.

Franchi et al., 20021743c + p26172419MMP-9
Hong et al., 20009442519MMP-2, MMP-9
Horikawa et al., 2000183c121212MMP-9
Imanishi et al., 20001070c30313332MMP-2, MMP-14
Krecicki et al., 20011150c + p22283515MMP-2
Kurahara et al., 1999796c + p58385838MMP-1–MMP-3, MMP-9, MMP-14
Kusukawa et al., 19931246p20261212MMP-2
Kusukawa et al., 19951565c + p50156565MMP-3
Kusukawa et al., 19961665c + p471838273434MMP-3
Repassy et al., 19981315c411114MMP-2
Riedel et al., 2000195212401438MMP-9
Sawatsubashi et al., 199888367167211MMP-1
Yoshizaki et al., 1997202711161611MMP-14
Yoshizaki et al., 20011451p42929222427MMP-2, MMP-14

Positive immunohistochemical signals were found for MMP-1 in 124 patients, MMP-2 in 212 patients, MMP-3 in 152 patients, MMP-9 in 150 patients, and MMP-14 in 129 patients. Notably, some studies analyzed the expression of several different MMPs.

Results for MMP-1

The literature research identified 6 publications that examined the expression of MMP-1. Due to a lack of original data, 66.7% of the publications were excluded, whereas 2 studies (33.3%) that involved a total of 179 patients were included in the current metaanalysis.7, 8 Among these, 124 patients (69.2%) had expression of MMP-1, and a P value of 0.0014 confirmed heterogeneity. Results of the fixed and random effects models are summarized in Table 2. Figure 1 illustrates forest plots of both studies, including the common odds ratios for both models. The breadth and length of the confidence intervals show the value of the published studies and the calculated common odds ratios.

Table 2. Results of the “Fixed Effects Model” and the “Random Effects Model” on the Relation between Matrix Metalloproteinase Expression and Lymph Node Status
MMP/test procedure95%CI (lower limit)OR95%CI (upper limit)
  1. MMP: matrix metalloproteinase; 95%CI: 95% confidence interval; OR: odds ratio.

MMP-1   
 Fixed effects model0.822.075.76
 Random effects model0.051.6652.68
MMP-2   
 Fixed effects model2.744.637.99
 Random effects model2.163.967.28
MMP-3   
 Fixed effects model1.954.269.94
 Random effects model3.705.096.98
MMP-9   
 Fixed effects model1.372.635.19
 Random effects model1.092.294.84
MMP-14   
 Fixed effects model2.174.7010.83
 Random effects model
Figure 1.

Forest plot (odds ratio and 95% confidence interval) of the results from studies that analyzed matrix metalloproteinase 1 in relation to lymph node status, including the common odds radio calculated according to both models.

Kurahara et al.7 verified the clinical lymph node (cN) status and the pathologic lymph node (pN) status in their report. However, they did not correlate the MMP expression levels with either tumor status or with cN status or pN status. Sawatsubashi et al.8 could not find any correlation with tumor status (odds ratio, 1.44; confidence interval, 0.41–5.37) and did not mention how they determined of lymph node status (Table 1). However, it is noteworthy that there was a significant difference (P = 0.0003; Fisher exact test) in MMP-1 expression between tumors from different locations (oral cavity vs. larynx).

Results for MMP-2

Although a total of 17 publications were found, 52.9% of those publications had to be excluded due to a lack of original data, and 1 publication (5.9%) did not allow evaluation of the odds ratio. In the remaining 7 studies (41.2%), with a total of 367 HNSCC tissue specimens, the relation between MMP-2 expression and lymph node disease was examined.7, 9–14 The P value of 0.1766 indicated heterogeneity. The common odds ratio is summarized in Table 2. The forest plot (Fig. 2) shows that results differ.

Figure 2.

Forest plot (odds ratio and 95% confidence interval) of the results from the studies that analyzed matrix metalloproteinase 2 in relation to lymph node status, including the common odds ratio calculated according to both models.

Two studies7, 11 considered cN status and pN status, two other studies10, 13 considered only cN status, and two other studies12, 14 considered only pN status, whereas one study9 did not consider lymph node status at all. None of those studies correlated MMP expression with cN status or pN status. Four of seven studies10, 11, 13, 14 correlated MMP expression and tumor status. The P value of 0.6053 indicated homogeneity. The common odds ratio (Fig. 3) is summarized in Table 3. MMP expression levels appeared to differ significantly (P = 0.04; Fisher exact test), depending on the localization of the tumor (oral cavity, hypopharnyx, larynx).

Figure 3.

Forest plot (odds ratio and 95% confidence interval) of the results from studies that analyzed matrix metalloproteinase 2 in relation to tumor status, including the common odds ratio calculated according to both models.

Table 3. Results of the “Fixed Effects Model” and the “Random Effects Model” on the Relation between Matrix Metalloproteinase Expression and Tumor Status
MMP/test procedure95%CI (lower limit)OR95% CI (upper limit)
  1. MMP: matrix metalloproteinase; 95% CI: 95% confidence interval; OR: odds ratio.

MMP-2   
 Fixed effects model0.671.372.84
 Random effects model1.011.231.50
MMP-9   
 Fixed effects model0.451.868.48
 Random effects model
MMP-14   
 Fixed effects model0.220.600.75
 Random effects model0.220.752.51

Results for MMP-3

A review of the published literature led to 6 publications, 3 of which (50%) had to be excluded due to lack of original data. The remaining publications included a total of 226 tissue specimens,7, 15, 16 which showed MMP-3 expression in 152 specimens (67.3%). The odds ratios for the studies ranged between 3.00 and 6.75 (Fig. 4) and revealed a P value of 0.5967 in the test for inhomogeneity. The common odds ratios for both test procedures are summarized in Table 2. The results of the fixed-effects model appear to be more adequate due to a lack of inhomogeneity.

Figure 4.

Forest plot (odds ratio and 95% confidence interval) of the results from studies that analyzed matrix metalloproteinase 3 in relation to lymph node status, including the common odds ratio calculated according to both models.

All publications7, 15, 16 demonstrated a clinically and pathologically verified lymph node status but did not correlate cN status or pN status with MMP expression. In their study, Kusukawa et al.15 only considered T1 and T2 tumors, thereby not allowing for comparison with higher staged (T3/T4) tumors. In another study, Kusukawa et al.16 correlated MMP expression with tumor status; however, they did not find any significant correlation (odds ratio, 1.76; confidence interval, 0.50–6.68). It is important to note that these publications7, 15, 16 only considered tumors of the oral cavity and neighboring structures.

Results for MMP-9

Fourteen publications investigated MMP-9 expression. Due to lack of data, however, nine articles had to be excluded from the current metaanalysis. The remaining 5 publications with a total of 238 patients7, 9, 17–19 demonstrated MMP-9 expression in 150 tissue specimens. The common odds ratio is summarized in Table 2. The odds ratios (Fig. 5) correspond along with the breadth and location of the 95% confidence interval (Table 2). The lower confidence limits are > 1.0, indicating statistical significance. However, statistical significance may fail if the analysis were adjusted for multiple testing of several MMPs.

Figure 5.

Forest plot (odds ratio and 95% confidence interval) of the results from studies that analyzed matrix metalloproteinase 9 in relation to lymph node status, including the common odds ratio calculated according to both models.

cN status and pN status were considered in two publications,7, 17 whereas one publication considered only pN status,18 and two other publications9, 19 did not provide any information about lymph node status. None of the studies mentioned compared levels of MMP-expression with cN status or pN status. Two studies18, 19 compared the level of MMP expression with tumor status. A heterogeneity check led to a P value of 1.0, which confirmed homogeneity. The common odds ratios (Fig. 6) are summarized in Table 3. It is noteworthy that MMP expression was dependent on the tumor location (oral cavity, oropharynx, hypopharyx, nasopharynx, larynx; P = 0.001; Fisher exact test).

Figure 6.

Forest plot (odds ratio and 95% confidence interval) of the results from studies that analyzed matrix metalloproteinase 9 in relation to tumor status, including the common odds ratio calculated according to the fixed-effects model.

Results for MMP-14

Six studies that investigated the role of MMP-14 in HNSCC were found during the literature search. Two articles had to be excluded due to a lack of original data. The 4 remaining articles were included in the current metaanalysis, and MMP-14 expression was found in 129 of 198 patients over those 4 studies.7, 10, 14, 20 The odds ratios of the studies were between 1.43 and 10.91 (Fig. 7). The heterogeneity check showed a P value of 0.5525, which confirmed homogeneity. For the fixed-effects model, a common odds ratio of 4.70 was found (Table 2).

Figure 7.

Forest plot (odds ratio and 95% confidence interval) of the results from studies that analyzed matrix metalloproteinase 14 in relation to lymph node status, including the common odds ratio calculated according to the fixed-effects model.

Clinical lymph node status (cN) and pathologic lymph node status (pN) were assessed in the study by Kurahara et al.,7 whereas Imanishi et al.10 and Yoshizaki et al.14 verified cN status and or pN status, respectively. Another study by Yoshizaki et al.20 did not comment on lymph node status at all. However, none of those studies compared MMP expression levels with cN status or pN status. Three of four studies10, 14, 20 compared MMP expression levels with tumor status. The P value of 0.288 indicated heterogeneity. The common odds ratio (Fig. 8) is summarized in Table 3. MMP expression differed significantly, depending on the localization of the tumor (oral cavity, hypopharynx, larynx; P = 0.007; Fisher exact test).

Figure 8.

Forest plot (odds ratio and 95% confidence interval) of the results from studies that analyzed matrix metalloproteinase 14 in relation to tumor status, including the common odds ratio calculated according to both models.

DISCUSSION

Meta-analyses are conducted when a consistent evaluation of risk factors is not possible due to several varying epidemiologic studies. Their objective is to find the single, most exact estimator for the quantitative effect of a risk factor when the result from a single study alone is not sufficient for a consistent assessment. Evaluation of risk factors often is difficult due to variability of study design, data collection, statistical analyses, and presentation of the results. This also holds true for studies regarding MMP expression in patients with malignant disease.

The objective of the current study was to use metaanalysis to investigate whether there is a correlation between MMP expression and lymphatic metastatic spread that would help in predicting the presence of lymph node metastases by evaluating the MMP expression profile in the primary tumor. However, 51,7% of the published investigations had to be excluded from the current metaanalysis due to lack of original data. Consequently, one-half of the current literature does not contribute toward making a final verdict on the significance of MMPs for lymph node disease in patients HNSCC. Thus, it is difficult to determine whether, in fact, MMPs play a key role in the very complex cascade of the metastatic process.

One further problem is related to the selection bias of positive results, which may be facilitated by the statistic method of the metaanalysis. Positive correlations may be represented excessively due to the following facts: There is a tendency to publish only positive results; conversely, correlations that could not be proven remain unpublished. If the published results are now bundled statistically, and if new objectives are deduced from these data, then an overly positive bias can be assumed. This applies especially if only limited numbers of studies have been published on a certain topic, which is true with regard to the immunohistochemical detection of MMP-1.

All studies that were included in the current metaanalysis investigated MMP expression in HNSCC tumors with different anatomic locations and compared them with lymph node status. However, none of the studies sufficiently distinguished cN status or pN status, and some studies even showed a total lack of information. Because the validity of prognostic parameters like MMPs largely depends on histopathologic confirmation, it appears difficult to draw clear conclusions from studies without information about pathologic staging. To make it even more complicated, several studies included patients who had tumors with different stages; however, only a few studies used multivariate analysis to correlate MMP expression with tumor status or lymph node status. Therefore, most studies do not allow for a clear conclusion regarding a potential correlation between tumor size, MMP expression, and lymphatic metastatic spread. Thus, it is necessary to discuss critically the current literature on MMP expression and the highly complex mechanism of lymphatic metastatic spread. This also is true for the patient who had MMP overexpression correlated with the localization of the primary tumor, which also was reported in some of the studies that were included in the current metaanalysis.

The heterogeneity of the analyzed reports expands not only to the patient cohorts but also to the different methodologies of the respective investigations. MMP levels can be determined with a multitude of different procedures. The great variety of MMP detection methods limits the comparability of studies, because it is not appropriate to evaluate results that have been gained by different detection methods. For example, MMPs and TIMPs can be described on the mRNA and protein levels. Immunohistochemical staining is of particular interest, because it can be carried out in formalin fixed tissue samples, which facilitates reproducibility in diagnostic histopathology. Furthermore, immunohistochemical staining allows for an exact location of MMPs. For these reasons, the detection of MMPs by means of immunohistochemical staining was chosen as the inclusion criterion for the current metaanalysis.

Although all of the studies that were included in this metaanalysis evaluated the expression patterns of MMPs by means of immunohistochemical staining, different antibodies often were used. Another crucial point is that these studies differed regarding their definition of positive and negative MMP expression. In some studies,9, 12, 15, 16 the limit between “MMP expression” and “no expression” was determined on the basis of a comparison between tumor specimen staining and the staining of histologically healthy epithelium, whereas other studies defined the limit between positive and negative on the basis of the percentage of immunohistologically stained cells in the specimen. For example, Imanishi and coworkers10 chose a staining of 5%, Franchi and coworkers17 chose a staining of 10%, and Riedel and coworkers19 chose a staining of 25% of the cells as the limit between “expression” and “no expression.”

Sawatsubashi et al.,8 Krecicki et al.,11 and Repassy et al.13 differentiated between “immune reaction” and “no immune reaction.” The original articles published by Horikawa et al.18 and Kurahara et al.7 did not provide exact definitions of “expression” and “no expression.” Kurahara et al.7 divided the specimens according to their staining into four groups. For the current metaanalysis, Group 0 (i.e., the group with almost negative staining of both tumor and stroma cells in the specimens) was classified with “no expression,” whereas the other groups were classified with “MMP expression.” In contrast to Horikawa et al.,18 who reported the percentage of immunohistochemically stained cells in the specimen for each patient, we classified the values 0.5% of cells stained and 2.9% of cells stained as “no expression” and 58.1% of cells stained as “MMP expression.”

Because different groups use many different parameters, obviously, it does not seem very surprising that several investigators have reported differing or even contradicting results. Therefore, it seems to be important to consider carefully the results of this metaanalysis, which potentially may assist in identifying some trends of MMP expression in patients with HNSCC.

The role of MMP-1 and MMP-9 within the metastatic process of HNSCC remains unclear. For MMP-2, it can be assumed that the risk of lymph node metastases is increased in patients who have HNSCC tumors that are positive for MMP-2. Furthermore, our current results show an increased risk for lymph node metastases in patients who have tumors with expression of MMP-3 or MMP-14.

In conclusion, based on heterogeneity of data collection, study design, statistical analyses, methodology, and evaluation protocols, meta-analyses of studies investigating immunohistochemical MMP expression in the process of HNSCC tumor invasion and metastatic spread have left the significance of MMP expression unclear. The standardization of immunohistochemical staining procedures and evaluation protocols will be required as a first step. These are the prerequisites to achieve comparable results for further evaluation and allow for thorough definition of metalloproteinase candidates for further studies on their significance in head and neck tumor progression and metastatic spread. Based on the complex composition of the extracellular matrix, multivariate analyses should be preferred over univariate analyses of MMP expression to clarify the interactions between several MMPs and TIMPs involved in the lymphogenic metastatic process of HNSCC.

Ancillary