Malignant ACTs are rare but highly aggressive and have a poor prognosis. Their prognosis is related to the tumor stage when the diagnosis is made, both clinically and by the pathologist (1)
. The differential diagnosis of benign (ACA) and malignant (ACC) tumors of the adrenal cortex is currently based on several histological parameters according to the Weiss scoring system, in which tumors with scores equal to or below two are classified as benign and those with scores equal to or above four as malignant. With regard to tumors with score 3, the Weiss scoring system might be insufficient to achieve a definite differential diagnosis between ACCs and ACAs (4)
. Everybody is in agreement that there is a need to identify novel molecular markers that will improve the differential diagnosis among ACTs and allow earlier identification of cases with malignant potential.
To meet this need, we performed an IHC investigation using 11 molecular markers, which were used to label a collection of samples from normal and pathological adrenal glands. These molecular markers have been previously studied separately by other research groups and were demonstrated to have potential usefulness for the differential diagnosis of ACTs. However, previous reports were sometimes either contradictory or subjective, mainly because results were analyzed using merely the researcher's observations (4, 13, 14, 21, 25)
One of the main strengths of our study was the use of a computerized evaluation method that allowed us to remove the subjectivity of the observer, and which could be used in future to study other molecular markers suggested by recent genomic studies (9, 26)
, which, if appropriately confirmed, may also become useful in clinical practice.
The employment of the ROC curves in our study was important, since it allowed us to determine the diagnostic accuracy of the molecular markers; compare the diagnostic accuracy of the different markers, and also calculate the best cutoff value to be used in the differential diagnosis of ACTs.
The main limitation of our study was the limited number of samples that we had access to; however, it could be a good starting point for large-scale studies, expanding the number of molecular markers but using this objective method of quantification.
Through StAR immunostaining, we confirmed that its expression was higher in functioning ACAs than in nonfunctioning ACAs, as expected. StAR is involved in a limiting step of steroidogenesis, the delivery of the precursor of steroid hormones, cholesterol, to the inner mitochondrial membrane, for the first enzymatic step in the steroidogenic pathway (20, 27, 28)
. Also as expected, the NAG group exhibited the highest expression of StAR. In contrast, ACC samples exhibited a lower expression of StAR compared with ACAc samples probably because the ACC group included more nonfunctioning tumors and also exhibited a generally decreased expression of StAR, possibly associated with its abnormal steroidogenesis. It was technically impossible to analyze the differences in the subgroups of functioning and nonfunctioning carcinomas due to the small number of cases in these subgroups, which resulted in a lack of statistical power. According to the ROC curve analysis, the accuracy of StAR as a marker for the differential diagnosis between ACCs and ACAcs has a high discriminative power, with an AUC value of 0.85. The cutoff value was calculated to be 8.26%.
We confirmed the growth factor IGF2 to be an excellent marker for differentiating between carcinomas and ACAn. This marker had an AUC value of 1.00, corresponding to 100% specificity and sensitivity for distinguishing ACAn from carcinomas using a cutoff value for the percentage of the stained area of 27.11. On the other hand, when comparing IGF2 data for total ACA vs ACC samples, we observed a lower AUC value, reflecting a lower accuracy for differential diagnosis. Soon et al
reported a slightly higher AUC value than us (0.86 vs 0.81) by comparing ACA and ACC samples. A possible explanation for this difference is that in the study of Soon et al
a lower percentage of ACAc were included (22)
, whereas we found IGF2 to be expressed in adenomas producing Cushing's syndrome. This finding will need confirmation with further research, as it has never been reported before. In conclusion, IGF2 has been proposed by many authors as a good marker for differentiating between ACAs and ACCs (21, 22, 29)
, but at least for the time being we suggest that its use be limited to the differential diagnosis between ACAns and ACCs.
No significant differences in the expression of the cell cycle molecular markers p53 (TP53), MDM2, and p21 (CDKN1A) were found between ACA and ACC samples. p53
is a tumor suppressor gene and encodes a protein that promotes DNA repair (7, 10)
; MDM2 is a protein that inactivates p53 by binding to both the wild-type p53 and the mutated p53 protein (30, 31)
; and p21 is a cyclin-dependent kinase inhibitor (CDKi) induced by p53, which when overexpressed triggers cell cycle arrest in proliferating cells (10)
. Although no significant results were obtained for the p53 protein, we observed that some ACC samples exhibited a very high expression of this protein, which indicates the presence of p53
mutations in these cases; however, other samples exhibited low expression, and it was this heterogeneity of p53 staining in the ACC samples that resulted in the difference between ACC and ACA samples for this marker not being significant.
The expression of cyclin D1, in contrast, was significantly higher in ACC samples than in ACA samples. Cyclin D1 is a regulator of the G1 to S phase transition of the cell cycle (32)
. Using the ROC curves to analyse the detection of differences between total ACA and ACC samples, we found an AUC value of <0.80, suggesting that this molecular marker is not very useful for the diagnosis of ACCs. Previous studies, using a cutoff of ‘5% positive cells’, failed to identify positive staining for this marker (13, 14)
The expression of Ki-67 protein was significantly higher in ACC samples than in ACA samples and NAGs. The ROC curve analysis for distinguishing between ACC and total ACA samples demonstrated an AUC value of 0.96 and the value of 0.50% as the best cutoff for the differential diagnosis of ACTs. Previous studies have reported similar results, and so the utility of Ki-67 is well supported (13, 14, 21, 22)
In this study, we could not associate the abnormal expression of β-catenin with the malignant character of the tumors, since we found nuclear expression in ACC samples as well as in nonfunctioning ACAs. Tissier et al
. had already verified that abnormal expression was observed in both ACAs and ACCs and that most ACAs exhibiting abnormal β-catenin immunostaining were nonfunctioning ACTs, corroborating our results (4, 16)
E-cadherin, which is a protein of cell adhesion, generally reported to be associated with β-catenin, was not found in any of the studied samples, as has been described previously by Khorram-Manesh et al
In contrast, p27 immunostaining was the most novel positive finding of this study, since it allowed a clear distinction between ACCs and all other groups of tumors. The protein encoded by p27
(CDKN1B) is a CDKi that regulates cell cycle progression from the G1 to the S phase of the cell cycle and upregulation of the expression of p27 results in cell cycle arrest and apoptosis (33)
. The percentage area stained for p27 was significantly higher for ACC samples than for all the other groups of samples. Analysis of the area under the ROC curve suggested that p27 has an excellent diagnostic accuracy for distinguishing between ACCs and both functioning and nonfunctioning ACAs with a value of 7.23% as the best cutoff for the differential diagnosis of ACTs. A previous study has shown the presence of p27 in almost all cases of ACCs, but failed to recognize its potential as a biomarker, since p27 was also observed in a substantial percentage of ACAs (13)
. Nakazumi et al
had also already reported that the expression of p27 is increased in ACAs. However, both Nakazumi's and Stojadinovic's studies were carried out by direct observation by the researchers, rendering some level of subjectivity to the interpretation of immunostaining results, which we attempted to overcome by using an automated method of analysis. We also determined the percentage area stained, while the aforementioned studies measured the number of stained nuclei, which is a possible explanation for the differences in the results. An additional explanation for the discrepancy between the results may be the use of different primary antibodies in the studies (13, 25)
It must be pointed out that these two previous studies had reached contradictory conclusions, as in Nakazumi's study the difference in the percentage of stained nuclei between the ACA and ACC samples favored the marking of the benign tumors, and although statistically significant, this difference was less pronounced than the one observed in the more recent study carried out by Stojadinovic, which found increased p27 staining in malignant tumors, similar to that observed in our study (13, 25)
. By being far less subjective and defining more correctly the cutoff value through the ROC analysis, our method made the identification of the distinction between ACCs and ACAs very significant. Results that were similar to the results of our study have been described previously for breast tumors and melanomas (34, 35, 36)
The presence of high levels of a CDKi in ACC samples is somewhat counterintuitive. However, the positive results of several studies indicate the existence of a mechanism that allows some cancer cells to either have a tolerance phenomenon for this inhibitor of cell cycle progression or to develop the ability to repress the activity of p27 as an important step in tumor progression. This would mean that p27 could be present but would be unable to produce its usual actions to arrest the cell cycle. An alternative hypothesis is that p27
gene could have mutations, resulting in a modified p27 protein that could have a still unknown role in tumorigenesis or tumor progression. However, p27
mutations have been described as a very rare phenomenon in human cancer. Nickeleit et al
suggested an interesting intuitive hypothesis, which states that if a tumor cell does not need to mutate a tumor suppressor gene, this might mean that the resulting protein must have some sort of a tumor-promoting function, even if so far unidentified.
Through correlation studies, we could verify that there were positive correlations between the levels of the growth factor IGF2, the cell cycle regulators p27 and cyclin D1, and Ki-67, meaning that the markers IGF2, p27, and cyclin D1 may all be promoting the high proliferative drive of ACCs. In our samples, only one case of ACC was negative for p27 and positive for Ki-67, while none of the cases positive for p27 were negative for Ki-67. Combining the AUC of Ki-67 and p27 did not produce any additional improvement in the ROC curve analysis, since each of these markers separately had already attained an excellent level of discrimination.
In conclusion, of the studied molecular markers, p27 and Ki-67 were the ones that demonstrated the highest discriminative power in differentiating between ACCs and ACAs, while IGF2 only seemed to be useful in differentiating between ACCs and ACAns and StAR for the differential diagnosis between ACCs and ACAcs. The main novel demonstrations of this study were the use of an automatic method of analysis to remove subjectivity and that p27 is overexpressed in ACCs, suggesting that this CDKi should have a still unknown role in adrenocortical tumorigenesis and possibly also represent a potential treatment target for malignant ACTs.