The expanding role of BAP1 in clear cell renal cell carcinoma

Abstract

Mutations drive renal cell carcinoma (RCC) biology and tumor growth. The BRCA1-associated protein-1 (BAP1) gene is frequently mutated in clear cell renal cell carcinoma (ccRCC) and has emerged as a prognostic and putative predictive biomarker. In this review, we discuss the role of BAP1 as a signature event of a subtype of ccRCC marked by aggressiveness, inflammation, and possibly a heightened response to immunotherapy.

Introduction

Clear cell renal cell carcinoma (ccRCC), the most common renal cell carcinoma (RCC) subtype, is a riveting tumor. Genes frequently mutated in other tumors such as lung, breast, colon or prostate cancer to name a few, are infrequently mutated in ccRCC. Instead, ccRCC is characterized by the inactivation of the gene von Hippel-Lindau (VHL) which is rarely mutated in other cancers. VHL is a two-hit tumor suppressor gene, located on chromosome 3p25. Deletions of this region, observed in approximately 90% of ccRCC^1-3, generally involve a large segment of chromosome 3p resulting in loss of one allele of VHL. 3p loss typically involves a chromothripsis event that is accompanied by 5q amplification. The second allele of VHL is inactivated either by mutation or methylation⁴. VHL inactivation results in accumulation of the alpha subunit of HIF (hypoxia-inducible factor) and constitutive expression of target genes⁵ promoting angiogenesis through upregulation of the vascular endothelial growth factor (VEGF)^6,7. VHL inactivation has been found in preneoplastic renal cysts and has been speculated to occur several decades before patients are diagnosed with ccRCC. In genetically engineered mice, VHL inactivation alone does not result in ccRCC^8-10. Thus, while VHL inactivation is the critical founding step, additional cooperating driver mutations are needed for ccRCC genesis.

Interestingly, chromosome 3p contains 3 other tumor suppressor genes frequently mutated in ccRCC (PBRM1, SETD2 and BAP1). PBRM1 (Polybromo-1), is mutated in ~40% of ccRCCs^11-14. The SETD2 (Set domain-containing protein 2) gene encodes a nonredundant histone 3 lysine 36 trimethyltransferase and is mutated in ~15% of ccRCC¹⁵. BAP1 (BRCA1 associated protein 1) is mutated in ~12% of ccRCC^12,13,16. Notably, all three genes are involved in epigenetic regulation. Other genes have been implicated in ccRCC pathogenesis including other chromatin-modifying genes such as KDM5C, KDM6A, and MLL2; mTOR pathway genes; and cell cycle regulators such as CDKN2A^13,17,18.

The BAP1 tumor suppressor protein

BAP1 has emerged as a critical tumor suppressor gene across multiple cancer types including melanoma, mesothelioma and RCC. BAP1 is mutated both somatically, as well as in the germline. The BAP1 gene is located on chromosome 3p21, has 17 exons, and encodes a 729 amino acid protein of 90 kDa. The BAP1 protein is a deubiquitinating enzyme with an N-terminal ubiquitin carboxy-terminal hydrolase (UCH) domain, a host cell factor 1 binding motif (HBM), and a C-terminal domain containing a coiled-coil motif that interacts with ASXL1/2. BAP1 localizes to the nucleus and contains a nuclear localization signal (NLS) in the C-terminus^19,20. BAP1 binds to chromatin and deubiquitinates histone H2A thereby opposing the silencing function of polycomb repressive complexes which ubiquitinate H2A²¹.

Mutations in BAP1 are distributed throughout the open reading frame. Independently of their destabilizing effects, truncating mutations often interfere with the C-terminal NLS, which is believed to be necessary for BAP1 function²². Missense mutations cluster in the catalytic domain, which is required for tumor suppressor function^12,14. Furthermore, BAP1 undergoes self-deubiquitylation which influences subcellular localization²³.

Discovered in 1998, BAP1 was found to bind the RING finger domain of the BRCA1 protein²⁴. However, this may not be true in all mammalian cells under normal conditions²⁵. BAP1 regulates processes including DNA damage repair, cellular differentiation, and cell cycle control²⁶.

The familial BAP1 syndrome

Inherited in an autosomal dominant pattern, BAP1 mutations confer a high risk of malignancy. The BAP1 syndrome is characterized by uveal and cutaneous melanoma, mesothelioma, and ccRCC²⁷. It is also associated with an increased risk of benign and malignant tumors including melanocytic tumors/atypical spitz tumors, basal cell carcinoma, hepatocellular carcinoma, cholangiocarcinoma, meningioma and others^28-32. RCCs have been reported in 10% of probands, occur at a younger age, and are more aggressive than their sporadic counterparts³³. As germline BAP1 mutation predisposes to ccRCCs that are indistinguishable from sporadic ccRCC, BAP1 loss may initiate ccRCC development in some settings.

Advancing our understanding of BAP1 with an immunohistochemistry assay

Using a combination of whole genome and exome sequencing, followed by targeted sequencing, we identified BAP1 mutations in 14% of primary ccRCCs¹². Most mutations were predicted to abrogate protein expression and a surrogate immunohistochemistry (IHC) test was developed¹². We developed an IHC assay with an antibody raised against the C-terminus of BAP1 and utilized genetically characterized tumor samples as controls¹². We applied this assay to a cohort of 175 tumors with BAP1 sequencing data. Positive nuclear staining was observed in 150 cases of which 148 were wild-type. The 2 discordant cases had missense mutations, which likely abrogated protein function, but did not affect the generation of a protein product¹². Twenty-five cases were negative by IHC including 22 cases with BAP1 mutations. Three samples lacking a mutation did not express BAP1 protein on western blot and therefore were likely to be BAP1-deficient. This suggested that there may be other mechanism of inactivation besides intragenic mutations. Overall, compared to gene sequencing, the IHC test had a greater sensitivity for BAP1 loss although slightly lower specificity. The positive and negative predictive values of the IHC test were 92% (for detecting BAP1 loss) and 98%, respectively. Moreover, IHC allowed the study of protein expression at the single cell level. This enabled, for example, resolving samples that were seemingly double mutant for BAP1 and PBRM1 based on mutation data, which turned out to be samples with different regions mutated for BAP1 and PBRM1. Another advantage of IHC is its routine, straightforward application and the utilization of formalin fixed paraffin embedded (FFPE) specimens, which are readily available.

Wild-type BAP1 results in nuclear immunostaining in ccRCC tumor cells. BAP1 IHC is not a quantitative assay but nearly all mutations are biallelic (intragenic accompanied by 3p loss) and abrogate nuclear staining. Mutations interfering with expression of the NLS may result in cytoplasmic staining and tumor cells where only cytoplasmic staining is noted are likely BAP1-deficient. In the absence of tumor nuclear staining, it is also important to assess the positive staining in the stromal and inflammatory cells, which offers a positive internal control. The BAP1 IHC assay has been utilized extensively (in more than 5,000 ccRCCs) at the Clinical Laboratory Improvement Amendments (CLIA)–certified University of Texas Southwestern Medical Center (UTSW) laboratory and has enabled significant discoveries^12,34-39.

BAP1 anticorrelation with PBRM1 sets foundation for ccRCC molecular classification

Interestingly, we discovered that simultaneous BAP1 and PBRM1 mutations co-occurred far less frequently than one might expect based on their individual frequencies alone¹². This mutual anticorrelation was confirmed in meta-analyses as well as in multiple studies including sequencing studies utilizing multiregional sampling^17,39. In a meta-analysis, the odds of having mutations in BAP1 were reduced in PBRM1-mutated tumors⁴⁰. In a cohort involving ~1,300 patients the odds ratio for simultaneous loss of BAP1 and PBRM1 was 0.18 (95% CI, 0.11-0.28; p<0.00001)³⁹. This is in contrast to SETD2 mutations where the odds are increased by two fold in PBRM1-mutated tumors, which is suggestive of cooperativity⁴⁰. These data show a strong negative selection for simultaneous BAP1 and PBRM1 mutations. Though there is precedent for tumors with loss of BAP1 and PBRM1 in different cell populations of the same tumor⁴¹, tumors with simultaneous loss of BAP1 and PBRM1 in the same cell populations are infrequent.

Mutation exclusivity is often observed amongst genes in the same pathway. However, as discussed below, BAP1- and PBRM1-mutant tumors exhibit different histological features, biology, and outcomes^{12,34,36,39,40}. When compared to ccRCC with PBRM1 mutations, BAP1-mutant tumors have characteristic, but independent, gene expression signatures³⁶. We analyzed 308 ccRCC from the TCGA that had RNA-Seq data available and found that when compared to the rest, the BAP1- and PBRM1-mutant signatures did not overlap beyond what is expected by chance alone, indicating that they regulate different processes¹². PBRM1 encodes BAF180 (herein referred to as PBRM1), a defining component of a switch/sucrose nonfermentable (SWI/SNF) nucleosome remodeling complex known as PBAF. PBRM1 recruits this complex to nucleosomes with a specific acetylation pattern and regulates DNA accessibility^42,43.

Gene mutations drive RCC growth and, similar to other cancers, define different subtypes. Therefore, given the mutual exclusivity, divergent gene expression and biologic behavior (discussed below), these mutations lay the foundation for a molecular genetic classification of sporadic ccRCC that has clinical implications.

BAP1 loss drives tumor aggressiveness

PBRM1-mutant tumors (especially in the absence of additional mutations, i.e. SETD2 mutations) tend to be of low grade. In contrast, BAP1-mutated tumors are of high grade (Figure 2) and are associated with aggressive features such as tumor necrosis, sarcomatoid differentiation and mTORC1 activation^12,36,44. An association of aggressive histology with advanced stage was observed in multiple cohorts including the TCGA cohort^{13,14,36,44,45}. Nucleolar size is the main determinant of grade and is controlled by mTORC1, and we found an association between BAP1 loss with mTORC1 activation¹².

Figure 2.

Representative images of clear cell renal cell carcinoma with loss of (A) PBRM1; (B) BAP1; (C) Both BAP1 and PBRM1. Tumors with BAP1 loss are frequently high grade in contrast to those with PBRM1 loss. Loss of BAP1 and PBRM1 anticorrelate and in the rare event when they co occur the tumors have most aggressive features (400x magnification).

The strongest evidence implicating BAP1 (and PBRM1) in promoting ccRCC tumorigenesis and regulating aggressiveness is provided by genetically engineered mouse models (GEMM). Using a Cre line driven by promoter sequences of Pax8 (and Six2) we showed that Vhl targeting alone fails to generate ccRCC. Instead, homozygous deletion of Vhl together with either Bap1 or Pbrm1 successfully caused neoplasms in the mouse kidney^9,10. Bap1- and Pbrm1-deficient tumors mirrored the morphology, grade and biologic behavior of human ccRCC¹⁰. Bap1-deficient tumors were of high grade, had mTORC1 activation, and increased proliferation rates (by ki-67 staining). In contrast, Pbrm1-deficient tumor exhibited small nest architecture, delicate interconnected vascular networks, low-grade nuclei, and longer latency than Bap1-deficient tumors. Interestingly, in Pbrm1-deficient kidneys, disrupting Tsc1 accelerated tumorigenesis, resulting in higher grade tumors with mTORC1 activation⁴¹. These data suggest that mTORC1 activation and BAP1 loss drive tumor grade and aggressiveness in ccRCC. Studies in mice also suggested that the cell of origin for ccRCC may be in the Bowman’s capsule. Overall, these data unequivocally implicate BAP1 and PBRM1 not only as drivers of ccRCC, but also of tumor grade.

Given the association between BAP1 and PBRM1 with nuclear grade^12,36, it is not surprising that BAP1- and PBRM1-deficient tumors are associated with different outcomes in patients. In patients with localized ccRCC, median overall survival (OS) for patients with BAP1-mutant tumors was 4.6 years (95% CI, 2.1-7.2), compared to 10.6 years (95% CI, 9.8-11.5) for patients with PBRM1-mutant tumors, corresponding to ~3x higher risk for BAP1 patients of dying from their disease (hazard ratio [HR], 2.7; 95% CI, 0.99-7.6; p=0.044) in an institutional cohort³⁶. These results were validated in a second cohort from TCGA^13,36. In addition, similar results were observed using the Mayo registry (a cohort of ~1,400 patients with non-metastatic ccRCC), where loss of BAP1 by IHC was associated with a markedly reduced rate of RCC-specific survival (HR, 3.06; 95% CI, 2.28 - 4.10; p = 6·10⁻¹⁴)^34,39 and from multi-institutional tissue microarray that included a cohort of 559 non-metastatic ccRCCs in which BAP1 negative tumors were associated with significantly worse disease-free survival (HR, 2.9; 95% CI, 1.8-4.7; p<0.0001) and OS (HR, 2.0; 95% CI, 1.3-3.1; p=0.001) than patients with BAP1 positive tumors³⁵. These results have been reproduced by multiple groups^{13,14,17,44,45}. One study suggests that the particular type of mutation may affect outcomes. Specifically, investigators found that truncating mutations may be associated with worse cancer-specific survival⁴⁴, but results have not been reproducible. Although other genes other genes have been associated with an unfavorable prognosis in ccRCC^44,46,47, BAP1 is the only one that has been consistently reported to correlate with poor prognostic factors including tumor grade, stage, size, necrosis, metastasis at presentation^44,46, and sarcomatoid histology^18,48.

Furthermore, we found that patients with loss of both BAP1 and PBRM1, though rare, were associated with even worse outcomes (median RCC-specific survival at 10 years of 42% vs. 87% for patients with wild type tumors for both BAP1 and PBRM1)^36,39,49.

BAP1 and PBRM1 (as well as SETD2) loss has also been observed in other RCCs, particularly papillary RCC, though at much lower frequency. The association of PBRM1 loss with survival has been variable, though BAP1 loss has consistently been associated with poor prognosis across RCC subtypes^13,38. In papillary RCC, inactivation of PBRM1, BAP1 and SETD2 occurs primarily through mutation and loss of chromosome 3p is rarely observed⁵⁰. Whether these mutations are driver mutations or whether PBRM1, BAP1, and SETD2 are haploinsufficient in papillary RCC remains to be determined.

The molecular evolution of BAP1-deficient tumors

Sato et al. determined that the mutant allele ratio of BAP1 is similar to VHL in 70-80% of ccRCCs suggesting that BAP1 mutations are subclonal in 20-30%¹⁴. Using BAP1 IHC on a single representative section per tumor we identified focal loss of BAP1 in ~3% of ccRCCs^34,37. More recently, to ascertain tumor phylogeny and intratumoral heterogeneity, the TRACERx program sequenced ~1200 samples from 105 ccRCC tumors of 100 patients (rare tumors with TCEB1 mutation as well as clear and papillary histology were included). TRACERx confirmed many of the prior studies. They showed mutual exclusivity of BAP1 with PBRM1/SETD2 mutations at a clonal level. BAP1 mutations were associated with low intratumoral heterogeneity (ITH), higher tumor grade and proliferation, suggesting that BAP1 loss confers sufficient fitness advantage to outcompete other mechanisms. In addition, they found that BAP1 had a propensity for being a lone driver as majority of these tumors had no additional driver mutations. Based on the conserved patterns of sequential mutations, they constructed an evolutionary classification that could be applied to ~60% of the tumors (63 of the 105 RCC). Amongst the 20 ccRCCs with BAP1 mutations, 17 could be classified: 12 were categorized as BAP1-driven and 5 as multiple clonal drivers (3 with an additional PBRM1 mutation and one with a SETD2 mutation). The mean time to progression after nephrectomy for these two evolutionary subtypes with BAP1 mutations was half that for subtypes driven by PBRM1 mutations (5.9, 4.7 respectively vs. 11.7 months).

BAP1 loss predicts worse outcome in small renal masses

Loss of BAP1 generally correlates with poor risk features, even in low stage tumors. We found that in small renal masses (SRMs) BAP1 predicted outcomes independently of known prognostic factors (including UCLA integrated staging system [UISS] nomogram) and in patients with low SSIGN (stage, size, grade, and necrosis) score³⁴. This is particularly important as most patients diagnosed with renal cancer today present with SRMs (≤ 4 cm). Given the low risk of metastasis in these patients, many can be followed with active surveillance protocols. However, a small subset behaves aggressively. In a recent study of ~2000 small SRM, BAP1 loss was significantly associated with metastases in multivariable analyses⁵¹. The finding that BAP1 is independently prognostic in SRM has practical applications since BAP1 status on biopsy is particularly accurate⁵². Recent work that included 178 patients with paired core biopsies and subsequent resection of the renal mass found that histology accuracy on the biopsy was excellent, biopsies underestimated grade, necrosis, and sarcomatoid/ rhabdoid features⁵². Amongst all prognostic variables evaluated in the ccRCC subset, BAP1 status had the highest agreement between biopsies and resection (93.5%; kappa [w] 0.78; 95% CI 0.59-0.96)⁵². Thus, BAP1 status on needle biopsy may better stratify patients with SRMs.

Deep learning algorithms exploit morphological characteristics of BAP1 deficient tumors

BAP1 has been reported to have a role in cytoskeleton remodeling²⁶. BAP1 loss in cutaneous tumors is associated with characteristic histologic features including epithelioid morphology⁵³. Similarly, ccRCC with BAP1 loss exhibit a characteristic architecture and cytologic features⁵⁴. These tumors have a tubulopapillary or expanded nested/trabecular architecture with cells characterized by moderate to abundant clear or granular eosinophilic cytoplasm, cytoplasmic globules and well-defined cell borders^35,54. Most distinctive are the nuclear features consisting of central round uniform nuclei, smooth open chromatin, large and prominent eosinophilic nucleoli, and subtle perinucleolar clearing (unpublished data and reference⁵⁴). These tumors frequently have tumor necrosis and a prominent intratumoral lymphocytic infiltrate. These findings set the foundation for evaluating deep learning algorithms to predict BAP1 status from hematoxylin and eosin stained ccRCC images. We have shown that deep learning algorithms can reliability predict BAP1 status using H&E slides alone (AUC=0.87 [range in different hold out cohorts=0.82-0.91]). The discriminative power went beyond grade and enabled the detection of subclonal loss⁵⁵. The algorithms were also accurate when applied to PDX models⁵⁵, where the human stroma is replaced by mouse stroma, showing that the morphology-BAP1 relationship is intrinsic to tumor cell characteristics.

BAP1 and inflammation - implications for immunotherapy

The frequency of BAP1 loss is higher in patients with metastasis (17-25%)^18,56-58. Similar to localized disease, BAP1 loss has been shown to have poor prognosis in advanced disease.

BAP1 studies have recently shifted focus towards immunological phenotypes^59,60. Tumors that respond to immune checkpoint inhibitors (ICI) are commonly associated with increased tumor-infiltrating immune cells, PDL-1 expression, and an inflammatory TME⁶¹. Furthermore, there is an association between BAP1 loss and PD-L1 expression; BAP1 negative ccRCCs are more likely to be PD-L1 positive than PBRM1 negative tumors (17.2% versus 8.3%, p=0.01)³⁹.

Interestingly, in the GEMM models, the Bap1-deficient tumors were observed to be associated with lymphocytes (CD4 and CD8 T cell) while Pbrm1-deficient tumors were not, suggesting a causal relationship. To investigate this further we undertook a novel empirical approach. We dissected the gene expression of RCC tumor cells by leveraging PDXs (where only the tumor cells expand)^62,63 and generated, by subtraction from the corresponding patient tumor, an empirically derived gene expression signature of the tumor microenvironment (eTME)⁶⁰. Based on the eTME, two major RCC tumor subtypes were identified: an inflamed and an angiogenesis-enriched/non-inflamed subtype. Inflamed tumors had abundant infiltration of T_regs, NK cells, T_h1 cells, neutrophils, macrophages, B cells, CD8⁺ T cells, and were enriched for BAP1 mutations (p=7.7x10⁻⁵). Additionally, an IFN-γ gene signature and PD1 expression strongly correlated with the inflamed subtype. In contrast, non-inflamed tumors were enriched for PBRM1 mutations, an angiogenesis signature, dendritic cells, and mast cells. Provocatively, patients with the inflamed subtype had indicators of systemic inflammation, such as thrombocytosis, and anemia, which are well validated prognostic factors in the IMDC (International Metastatic RCC Database Consortium) risk model^64,65. Overall, these analyses show that the molecular makeup of tumor cells could modulate the immune response in the TME.

Thus, mirroring the antithetical relationship observed between PBRM1 and BAP1, are signatures of angiogenesis and inflammation that also tend not to overlap in ccRCC (Figure 3) and have therapeutic implications^60,66,67. PBRM1-mutated tumors have been reported to have inferior outcomes to ICI-only regimens compared with those with an angiogenesis-targeting agent⁶⁶. Others have demonstrated that expression of endogenous retroviruses is enriched in BAP1 mutated tumors and correlates with IO-response^68,69. Furthermore, BAP1 mutated tumors demonstrate a more inflamed immune microenvironment suggesting that some form of immune-targeting strategy may benefit these patients^59,71. In patients enrolled in two phase III trials (COMPARZ [first line sunitinib vs pazopanib] and RECORD-3 [first line sunitinib vs everolimus] presence of a BAP1 mutation was associated with worse outcomes^57,70. Mutations in PBRM1 correlated with statistically longer time to treatment failure with VEGF-targeted therapy. However, a correlation between BAP1 loss and sensitivity to everolimus was not found. These data resulted in an improved genomically-annotated MSKCC risk model that incorporated presence of any mutation in BAP1 or TP53 or both and absence of any mutations in PBRM1 (C-index: original MSKCC model, 0.595 [95% CI 0.557-0.634] vs new model, 0.637 [0.595-0.679])⁷⁰.

Figure 3.

Representative images of clear cell renal cell carcinoma with loss of BAP1. Tumors with BAP1 loss are frequently high grade inflamed tumors that express PDL1 (200x magnification).

RNA-seq approaches have been used to predict response by identification of inflamed or angiogenic subtypes. One of the largest gene expression profiling studies to predict treatment response was undertaken for the ~800 metastatic ccRCC patients enrolled in the randomized phase 3 IMmotion151 trial evaluating the combination of atezolizumab (anti-PD-L1) and bevacizumab (anti-VEGF) vs. sunitinib (VEGFR2 inhibitor)^71,72. Unsupervised analyses of the most variably expressed genes identified 7 clusters that broadly included an angiogenic and a proliferative group. One of the proliferative clusters (Cluster 4) with the highest BAP1 mutation rate (40%) had the highest PD-L1 protein expression rate (80%). This cluster was characterized by a high immune signature, and showed preferential responsiveness to ICI-containing regimen.

Current treatments for metastatic ccRCC involve ICI, angiogenesis inhibitors and their combinations. Representing two extremes of the angiogenic to proliferative/inflamed tumor spectrum are ccRCC with pancreatic metastases versus those with sarcomatoid differentiation respectively. ccRCC with metastasis to the pancreas (independently of other organs) is associated with remarkably indolent clinical course that is independent of both metastatic tumor burden and IMDC risk scores. These tumors are characterized by low grade, PBRM1 loss, BAP1 retention, and angiogenesis^73-75. Consistent with these findings, they exhibit preferential responsiveness to antiangiogenic drugs and are resistant to nivolumab^73-75. In contrast, sarcomatoid ccRCCs are inflamed and devoid of vasculature, have high frequency of BAP1 loss (40%) and are resistant to antiangiogenic drugs, but responsive to ICI^18,71,76. As immune and vascular traits are readily accessible through analyses of histological slides,⁷³ histology is well poised to advance the biomarker field.

Conclusions

In summary, the BAP1 two-hit tumor suppressor gene cooperates with VHL loss and drives ccRCC with characteristic morphological features including hallmarks of aggressive tumors such as high grade. BAP1 loss is associated with mTORC1 activation and an inflamed, poorly vascularized, tumor microenvironment. Its mutual exclusivity with PBRM1 (which drives low grade, angiogenic tumors) set the foundation for a molecular classification with prognostic and potentially therapeutic implications (Figure 4). BAP1 prognostic role is particularly significant in SRMs, where it functions as an independent prognostic factor. Future studies will dissect the mechanism of tumor suppression and inflammation in the hopes of identifying therapeutic targets that may boost, in particular, immunotherapy.

Figure 4.

Diagram illustrating that BAP1 and PBRM1 are mutually exclusive, cooperate with VHL loss and drives ccRCC with characteristic morphological features. BAP1 loss is associated with an inflamed, poorly vascularized, high grade ccRCC unlike PBRM1 that drives low grade and angiogenic ccRCC.

Figure 1.

Representative section of a clear cell renal cell carcinoma with truncating mutation in BAP1 gene. (A)Hematoxylin and eosin-stained section reveals a high-grade tumor with spaced out vasculature and tumor cells with abundant cytoplasm, and high-grade nuclei. (B) Immunohistochemical stain for BAP1 (arrow) shows lack of nuclear immunoreactivity in the tumor nuclei. Background endothelial and inflammatory cells retain BAP1 and serve as internal control (100x magnification).