Introduction

PAX8 is a member of the paired-box gene family and is expressed in embryogenesis and organ development of the thyroid, Mullerian, Wolffian, and renal/upper urinary tract and is also required for tissue homeostasis in the respective adult tissues [25, 32]. PAX8 is a transcriptional regulator of thyroid-specific genes such as thyroglobulin, thyroid peroxidase, and the sodium-iodide symporter by binding to promoter regions through its 128-amino acid paired domain and is essential for thyroid follicular cell metabolism [9, 53]. In Wolffian and Mullerian duct derived tissues, PAX8 is important for mesenchymal-to-epithelial transition (EMT), regulates branching morphogenesis and nephron differentiation [28, 49], and may also modulate WT1 transcription [10].

In diagnostic pathology, PAX8 immunohistochemistry (IHC)—in combination with other markers—is often used to determine the origin of tumors that are difficult to classify by morphology alone. Detectable PAX8 expression is considered a strong argument for a tumor origin from the kidney, thyroid, or inner female genital tract [30, 43]. Numerous reports have described PAX8 expression in cancer. For many tumor types, however, the reported frequencies of PAX8 positivity vary considerably, which makes it practically impossible to derive the prevalence of PAX8 expression in a particular tumor type from the literature. For example, the reported rate of PAX8 positivity ranges from 0 to 100% in cervical squamous cell carcinoma [20, 30], anaplastic thyroid cancer [35, 48], and non-invasive and invasive urothelial cancer [33, 46], from 0 to 96% in Merkel cell carcinoma [22, 38], from 0 to 83% in cervical adenocarcinoma [43, 46], from 0 to 75% in medullary thyroid cancer [22, 29], from 31 to 100% in papillary thyroid carcinoma [20, 54], from 42 to 95% in renal oncocytoma [46, 55], and from 38 to 100% in endometrioid adenocarcinoma of the ovary [12, 18]. These conflicting data are probably caused by the use of different antibodies, immunostaining protocols, and criteria to determine PAX8 positivity in these studies.

To better understand the prevalence and diagnostic utility of PAX8 immunostaining in cancer, a comprehensive study analyzing a large number of neoplastic and non-neoplastic tissues under highly standardized conditions is desirable. Therefore, PAX8 expression was analyzed in more than 17,000 tumor tissue samples from 149 different tumor types and subtypes as well as 76 non-neoplastic tissue categories by IHC in a tissue microarray (TMA) format in this study.

Materials and methods

Tissue microarrays (TMAs)

Our normal tissue TMA was composed of 8 samples from 8 different donors for each of 76 different normal tissue types (608 samples on one slide). The cancer TMAs contained a total of 17,386 primary tumors from 149 tumor types and subtypes. Detailed histopathological and molecular data were available for cancers of the kidney (n = 1757), ovary (n = 524), endometrium (n = 259), and the bladder (n = 1663). Clinical follow-up data were accessible from 850 renal cell cancer patients with a median follow-up time of 39 months. Data on the expression of cadherin 16 (CDH16) [21], GATA3 [36], and p63 [41] were available from previous studies using subsets of the TMAs of this study. The composition of normal and cancer TMAs is described in the results section. All samples were from the archives of the Institutes of Pathology, University Hospital of Hamburg, Germany, the Institute of Pathology, Clinical Center Osnabrueck, Germany, and Department of Pathology, Academic Hospital Fuerth, Germany. Tissues were fixed in 4% buffered formalin and then embedded in paraffin. The TMA manufacturing process was described earlier in detail [8, 19]. In brief, one tissue spot (diameter 0.6 mm) was transmitted from a tumor containing donor block to an empty recipient paraffin block. The use of archived remnants of diagnostic tissues for TMA manufacturing, their analysis for research purposes, and patient data were according to local laws (HmbKHG, §12), and analysis had been approved by the local ethics committee (Ethics Commission Hamburg, WF-049/09). All work has been carried out in compliance with the Helsinki Declaration.

Immunohistochemistry

Freshly prepared TMA sections were immunostained on one day in one experiment. Slides were deparaffinized with xylol, rehydrated through a graded alcohol series, and exposed to heat-induced antigen retrieval for 5 min in an autoclave at 121 °C in pH 7.8 DakoTarget Retrieval Solution™ (Agilent, CA, USA; #S2367). Endogenous peroxidase activity was blocked with Dako Peroxidase Blocking Solution™ (Agilent, CA, USA; #S2023) for 10 min. Primary antibody specific for PAX8 (rabbit recombinant, MSVA-708R, MS Validated Antibodies, GmbH, Hamburg, Germany; #3331-708R) was applied at 37 °C for 60 min at a dilution of 1:150. For the purpose of antibody validation, the normal tissue TMA was also analyzed by the mouse monoclonal PAX8 antibody MRQ-50 (Cell Marque™—Sigma Aldrich®, CA, USA; #363 M) at a dilution of 1:15 and an otherwise identical protocol. Bound antibody was visualized using the EnVision Kit™ (Agilent, CA, USA; #K5007) according to the manufacturer’s directions. A subset of 1009 tumors (as detailed in supplementary Fig. 4) were also analyzed with both antibodies to document the impact of antibody selection on staining results. The sections were counterstained with hemalaun. For tumor tissues, the percentage of positive neoplastic cells was estimated, and the staining intensity was semiquantitatively recorded (0, 1 + , 2 + , and 3 +). For statistical analyses, the staining results were categorized into four groups. Tumors without any staining were considered negative. Tumors with 1 + staining intensity in ≤ 70% of tumor cells or 2 + intensity in ≤ 30% of tumor cells were considered weakly positive. Tumors with 1 + staining intensity in > 70% of tumor cells, 2 + intensity in 31–70%, or 3 + intensity in ≤ 30% of tumor cells were regarded as moderately positive. Tumors with 2 + intensity in > 70% or 3 + intensity in > 30% of tumor cells were considered strongly positive.

Statistics

Statistical calculations were performed with JMP® 16 software (SAS Institute Inc., NC, USA). Contingency tables and the chi2-test were performed to search for associations between PAX8 and tumor phenotype. Survival curves were calculated according to Kaplan–Meier. The Log-Rank test was applied to detect significant differences between groups. Sensitivity and specificity were calculated using the formulas TP/(TP + FN) and TN/(TN + FP), respectively, where TP is the number of true positive, TN is the number of true negative, FP is the number of false positive, and FN is the number of false negative.

Results

Technical issues

A total of 15,223 (87.6%) of 17,386 tumor samples were interpretable in our TMA analysis. Non-interpretable samples demonstrated absence of unequivocal tumor cells or a complete lack of individual tissue spots. A sufficient number of samples of each normal tissue type was always evaluable (≥ 4).

PAX8 in normal tissues

A strong nuclear PAX8 staining, which was often accompanied by a weak cytoplasmic staining, was observed in follicular cells of the thyroid, epithelial cells of the endometrium, endocervix, and the epididymis, ciliated epithelial cells of the fallopian tube, and a subset of epithelial cells of the seminal vesicle. In the kidney, variable, weak to strong nuclear PAX8 staining of cells was seen in proximal and distal tubuli, collecting ducts and epithelial cells of the parietal membrane of the Bowman’s capsule. In some samples, a weak to moderate nuclear PAX8 staining was also seen in basal and suprabasal cell layers of the urothelium, especially in the renal pelvis. Representative images of PAX8 staining are given in Fig. 1. All these nuclear stainings were observed by both MRQ-50 and MSVA-708R. Additional nuclear staining of a subset of lymphocytes, thymic epithelial cells, pancreatic islet cells, epithelial cells of the parathyroid, and neuroendocrine cells of the gastrointestinal tract as well as a granular cytoplasmic staining in acinar cells of the pancreas and in other cells was only seen by MRQ-50 and thus considered antibody specific cross-reactivities of MRQ-50. An additional, purely cytoplasmic staining in a small subset of inflammatory cells of the intestine, gallbladder epithelium, in some samples of gastric glands, and a subset of epithelial cells of the adenohypophysis was only observed by MSVA-708R and thus considered an antibody specific cross-reactivity of MSVA-708R. Comparative images showing staining by both antibodies are given in supplementary Fig. 1. PAX8 staining was absent in intima and media of the aorta, heart (left ventricle), skeletal muscle, skeletal muscle/tongue, myometrium, muscular wall of the gastro-intestinal-tract (appendix, esophagus, stomach, ileum, and colon descendens), muscular wall of the renal pelvis and bladder, ovarian stroma, keratinocytes of the epidermis, sebaceous glands, squamous epithelium of the ectocervix, placental cells (cytotrophoblast, syncytiotrophoblast, amnion, and chorion), decidua, gastric epithelial cells, enterocytes of the small and large intestine including appendix, hepatocytes, Kupffer cells, acinic cells and ductal cells of the exocrine pancreas, mucinous and/or serous epithelium as well as ductal cells of the salivary glands (parotis, glandular submandibularis, and glandular sublingualis), Sertoli cells, Leydig cells, and germinal cells of the testis, bronchus epithelium, pneumocytes, epithelium of the paranasal sinus, glandular and ductal epithelium of the breast, cortical and medullary cells of the adrenal gland, neuronal and glial cells of the cerebrum and cerebellum, and in cells of the neurohypophysis.

Fig. 1
figure 1

Pattern of PAX8 immunostaining in normal tissues. The panels show a strong PAX8 staining in distinct cell types of the thyroid (A), kidney (B), caput epididymis (C), endocervical mucosa (D), endometrium (E), and the fallopian tube (F). PAX8 staining is absent in a parathyroid gland (G) and a lymph node (H)

Full size image

PAX8 in cancer

PAX8 immunostaining was detectable in 3,400 (22.3%) of the 15,223 analyzable tumors, including 330 (2.2%) with weak, 782 (5.1%) with moderate, and 2,288 (15.0%) with strong immunostaining. Overall, 40 (26.8%) of 149 tumor categories showed detectable PAX8 expression with 32 (21.5%) tumor categories including at least one case with strong positivity (Table 1). Representative images of PAX8-positive tumors are shown in Fig. 2. The highest rate of PAX8 positivity was found in thyroidal neoplasms of follicular origin (98.6–100%), gynecological adenocarcinomas (up to 100%), renal tumors (82.6–97.8%), and urothelial neoplasms (2.3–23.7%). A graphical representation of a ranking order of PAX8 positive and strongly positive cancers is given in Fig. 3. Clinically relevant tumors with near complete absence of PAX8 staining (< 1%) included for example all subtypes of breast cancers, gastric adenocarcinoma, prostatic adenocarcinoma, hepatocellular carcinomas, cholangiocarcinoma, gallbladder adenocarcinoma, pulmonary adenocarcinoma, neuroendocrine neoplasms, and small cell carcinomas of various sites. High PAX8 expression was associated with low tumor grade in a cohort of 365 non-invasive papillary urothelial carcinomas (p < 0.0001) but unrelated to patient outcome and/or unfavorable tumor phenotype in clear cell renal cell carcinoma, high-grade serous ovarian cancer, and endometrioid endometrium carcinoma (Table 2 and Fig. 4).

Table 1 PAX8 immunostaining in human tumors
Full size table
Fig. 2
figure 2

PAX8 immunostaining in cancer. PAX8 staining is predominantly nuclear but accompanied by a weaker cytoplasmic positivity in cases with strong positivity. The panels show PAX8 positivity in a papillary renal cell carcinoma (A), a papillary (B), and an anaplastic (C) thyroid cancer, a serous high-grade ovarian carcinoma (D), an adenocarcinoma of the cervix (E), and a non-invasive papillary urothelial carcinoma (F). PAX8 staining is absent in a neuroendocrine tumor of the pancreas (G) and in a medullary carcinoma of the thyroid containing entrapped PAX8-positive follicular epithelial cells (H)

Full size image
Fig. 3
figure 3

Ranking order of PAX8 immunostaining in tumors. Both the percentage of positive cases (blue dots) and the percentage of strongly positive cases (orange dots) are shown

Full size image
Table 2 PAX8 immunostaining and tumor phenotype
Full size table
Fig. 4
figure 4

PAX8 immunostaining and patient prognosis in clear cell renal cell carcinoma

Full size image

Comparison of PAX8 antibodies

A subset of 1,009 tumor samples from 23 different tumor types was analyzed with both MSVA-708R and MRQ-50. Both antibodies stained comparably positive in 472 ovarian cancers, including 98.1% (MSVA-708R) and 97.2% (MRQ-50) of serous, 95.7% and 100% of clear cell carcinomas, and 62.5% each of ovarian carcinosarcomas (supplementary Fig. 2). MSVA-708R had a higher sensitivity (63.6%) as compared to MRQ-50 (33.3%) in mucinous and endometrioid (90.9% vs. 78.8%) ovarian cancers. A concordantly negative result was found in 42 tumors including acinar cell carcinomas of the pancreas and neuroendocrine tumors of the lung, colorectum, and appendix. However, exclusive staining with MRQ-50 (but not with MSVA-708R) was found in 44.4% of 171 neuroendocrine neoplasias of various origin and in 87.6% of 275 lymphomas (supplementary Fig. 3). All data are summarized in supplementary Fig. 4.

Sensitivity and specificity calculations

Data on the expression of CDH16, GATA3, and p63 were available from subsets of the tumors for which PAX8 data were collected in our project. Results from a comparative analysis of PAX8 and CDH16 are shown in Fig. 5 and supplementary Table 1. These data show that PAX8/CDH16 dual positivity almost exclusively occurred in neoplasms derived from kidney, thyroid, uterus, and ovary. Various further tumor entities showed positive staining either for PAX8 or CDH16 but not for both. For the distinction between a renal cell origin and a non-renal origin of tumors (including all other tumor entities of our study), sensitivity was 88.1% and specificity 87.2% for PAX8, while sensitivity was 85.3% and specificity 95.7% for CDH16. The combination of PAX8 and CDH16 increased specificity to 96.6%. For the distinction between renal cell carcinomas and urothelial carcinomas, sensitivity was 86.7% and specificity 91.3% for PAX8, while sensitivity was 82.7% and specificity 99.8% for CDH16. The combination of PAX8 and CDH16 increased specificity to 99.9%. For comparison with established markers for the distinction between urothelial carcinomas and renal cell carcinomas, we also performed sensitivity and specificity calculation of p63 and GATA3. Sensitivity was 86.5% and specificity 100% for p63, while sensitivity was 83.7% and specificity 98.3% for GATA3. All data are summarized in Table 3.

Fig. 5
figure 5

PAX8 and CDH16 immunostaining in human tumors

Full size image
Table 3 Sensitivity and specificity of PAX8 and CDH16 to determine renal tumor origin
Full size table

Discussion

The successful analysis of 15,223 cancers provides a comprehensive overview on PAX8 expression in cancer. That PAX8 positivity was most commonly seen in neoplasms of the thyroid (follicular), the kidney, and the female genital tract was expected based on numerous earlier studies describing high PAX8 positivity rates in these entities [20, 30, 43] and because the pattern of protein expression of tumor cells typically reflects the expression of corresponding normal cells. These results also support the previously suggested utility of PAX8 for the distinction of these tumor entitles from other cancer types especially in case of metastatic disease (summarized in [39]). This is all the more true since other important tumor entities that often metastasize were almost always PAX8 negative. Tumors with close to 100% PAX8 negativity for example included carcinomas of the breast and the liver, gastric, prostatic, pancreatic, and pulmonary adenocarcinomas, as well as small cell carcinomas and neuroendocrine tumors of various sites. For most of these PAX8-negative tumor types, several other studies have described significant PAX8 positivity rates often exceeding 20% [11, 34, 40, 51] or even 50% [16, 20, 22, 43]. Antibody cross-reactivities for PAX5 and/or PAX6 which are known to occur with monoclonal [22] and polyclonal PAX8 antibodies [48] might represent a major cause for these discrepancies.

To avoid an impact of antibody cross-reactivity on our data, emphasis was placed on a thorough validation of our assay. The International Working Group for Antibody Validation (IWGAV) has suggested that either a comparison of the findings obtained by two different independent antibodies or a comparison with expression data obtained by another independent method should be performed to validate antibodies for IHC on formalin-fixed tissues [50]. As an independent approach for expression measurement, compiled RNA data from three independent RNA screening studies were used [7, 23, 24, 45]. These projects had identified PAX8 RNA only in kidney, epididymis, seminal vesicle, endometrium, cervix uteri, fallopian tube, and thyroid. The fact hat the  immunohistochemical PAX8 analysis of 76 different normal tissue categories by MSVA-708R revealed nuclear positivity in only these organs supports the validity of our assay. True expression of PAX8 in all cell types with a nuclear PAX8 positivity is further validated by the confirmation of all these stainings by our second anti-PAX8 antibody MRQ-50. The use of a very broad range of normal tissues for antibody validation increases the likelihood for detecting undesired cross-reactivities because virtually all proteins occurring in normal cells of adult humans are subjected to the validation experiment. Additional nuclear staining of lymphocytes, thymic epithelial cells, the parathyroid, and several neuroendocrine cell types that were only observed by MRQ-50 were therefore considered specific cross-reactivities of this antibody. Accordingly, a considerable fraction of neuroendocrine neoplasms and lymphomas stained positive for MRQ-50 and not for MSVA-708R, although the positivity rates of these antibodies were comparable in ovarian tumors. Cytoplasmic staining of few inflammatory cells of the intestine, a subset of cells of the adenohypophysis and of gastric glands which were not seen by MRQ-50, was identified as cross-reactivities of MSVA-708R. These, however, do not cause interpretation issues as they are non-nuclear.

The large scale of our study enabled a ranking list of human tumor entities according to the prevalence of positive PAX8 immunostaining. As it is evident from the summarized literature (Supplementary Fig. 2), this list could not have been easily compiled from the published literature because of the high diversity of published data. The use of TMAs did not only enable the scale of the study but also allowed for a very high level of standardization, which not only included traditional experimental factors such as incubation time, temperature, and antibody concentration but also other important factors such as section age and the quantity of tumor analyzed per patient. Section age of 2 weeks and higher already results in a significant reduction of staining for many antibodies [17, 27]. The fact that the tissue quantity of tumor analyzed affects the positivity rate is already seen in studies comparing one, two, or three cores per tissue block on a TMA [4, 14, 37]. Of note, the only study comparing IHC findings obtained from large sections versus TMAs containing one or several tissue cores found almost twice as many p53 positive cases on whole sections than on TMAs, but the prognostic role of p53 positivity was only found for TMA and not for whole section data [47]. While these data show that the use of larger tissue quantities also increases the risk for finding artificial or irrelevant staining, they emphasize the desirability for standardizing tissue quantities for comparative tumor tissue analysis. The quantity of tissue on a TMA spot is not much different from the amount of tumor which is often contained in small biopsies. It must be considered, however, that some false negative cases always occur in TMA studies due to preanalytical tissue damage.

Previously published PAX8 IHC data for different tumor types are shown in comparison to our results in supplementary Fig. 5 and supplementary Table 2. Based on our comparison of MSVA-708R and MRQ-50, we assume that very high positivity rates in many tumor entities that were largely PAX8 negative in our cohort were driven by similar antibody-specific cross-reactivities. These especially include B-cell lymphomas [31], neoplasms of the thymus [42], neuroendocrine tumors [22], and Merkel cell carcinomas [38]. The PAX8 positivity rates of 0–75% reported earlier for medullary carcinomas of the thyroid may to some extent be caused by PAX8 antibodies cross-reacting with PAX6 [13, 29, 48]. In addition, interspersed normal follicular cells which are commonly seen between tumor cell layers (Fig. 2H) may have contributed to the perception of PAX8 positivity in medullary cancers [13]. It is also of note that we had seen PAX8 positivity neither in 4 samples of normal parathyroid nor in 43 parathyroidal adenomas by MSVA-708R, while normal parathyroid was stained strongly by MRQ-50. Altinay et al. [1] had recently suggested that PAX8 analysis by MRQ-50 may distinguish parathyroidal adenoma (positive in 85%) from normal parathyroid (100% negative).

That PAX8 positivity was also seen in 88 cancers from 118 tumor entities that were not derived from thyroid, kidney, or the female genital tract represents a diagnostic challenge. Considering that urothelial neoplasms are among these occasionally PAX8-positive tumors, PAX8 cannot reliably be used to distinguish urothelial carcinoma from renal cancer in kidney masses [2] or from ovarian or endometrial cancer in pelvic masses as previously suggested [52]. In an earlier comparative study, our group had identified CDH16 as a suitable marker for the distinction of renal cell carcinomas from other tumor entities [21]. The combined analysis of PAX8 and CDH16 data suggests that this combination could be useful. Almost all of the 1714 (99.9%) tumors with PAX8/CDH16 dual positivity were derived from kidney/female genital tract/thyroid. Of note, CDH16 is not a recommendable marker for thyroid cancer detection, because it is often lost in thyroidal carcinomas.

Sporadic reports on a possible prognostic impact of PAX8 in different tumor types [3, 5, 6, 15, 26, 44] and the availability of a clinical database to our TMAs prompted us to search for possible associations between PAX8 expression and cancer aggressiveness. The evaluation of large cohorts of urinary bladder and renal cell carcinomas, as well as ovarian and endometrial cancers, did not suggest a major clinical and prognostic impact of PAX8 expression in these entities. The significant association between high PAX8 expression and low tumor grade in non-invasive (pTa) urinary bladder cancers fits well to the weak to moderate nuclear PAX8 staining that was occasionally seen in normal urothelium, which appears to be lost during tumor progression.

In summary, the standardized assessment of PAX8 staining in 149 different tumor types and subtypes enabled us to define a ranking order with respect to the frequency of PAX8 immunostaining in tumors. Although neoplasms derived from kidney, thyroid, and the female genital tract were most often and most strongly PAX8 positive, there were 15 additional tumor types exhibiting PAX8 positivity at least in occasional cases. A parallel analysis of CDH16 may efficiently complement PAX8 IHC to discriminate tumors from the kidney, the female genital tract, or the thyroid.