Carbonic anhydrase IX and pathological features as predictors of outcome in patients with metastatic clear-cell renal cell carcinoma receiving vascular endothelial growth factor-targeted therapy


Toni K. Choueiri, Lank Center for Genitourinary Oncology, Dana-Farber Cancer Institute/Brigham and Women’s Hospital, Harvard Medical School, 44 Binney Street, Boston, MA 02115, USA.
e-mail: and Sabina Signoretti, Dana-Farber Cancer Institute and Brigham and Women’s Hospital, 75 Francis Street, Mailstop: Thorn Bld, 504A, Boston, MA, 02115, USA.


Study Type – Prognosis (retrospective cohort)
Level of Evidence 2b


To investigate the utility of tumour carbonic anhydrase IX (CAIX) expression and histological features for predicting the outcome in patients with metastatic clear-cell renal cell carcinoma (mRCC) treated with vascular endothelial growth factor (VEGF)-targeted therapy.


We identified 118 patients with mRCC initiating first-line VEGF-targeted therapy, including 94 with clinical and histological data, and available tissue. The primary endpoint was to detect an interaction between sorafenib vs sunitinib treatment and CAIX status on tumour shrinkage. Other treatment outcomes were also assessed.


There was heterogeneity in tumour responsiveness to sunitinib or sorafenib according to CAIX status; the mean shrinkage was –17% vs –25% for sunitinib-treated patients with high vs low tumour CAIX expression, compared to –13% vs +9% for sorafenib-treated patients (P interaction, 0.05). A higher tumour clear-cell component was independently associated with greater tumour shrinkage (P= 0.02), response (P= 0.02) and treatment duration (P= 0.02).


Although CAIX expression had no prognostic value in patients with clear-cell mRCC treated with VEGF-targeted therapy, it might be a predictive biomarker for response to sorafenib treatment. Patients with a higher clear-cell component in their tumours are likely to have a superior clinical benefit from VEGF-targeted therapy.


metastatic RCC


vascular endothelial growth factor


(protein) von Hippel-Lindau


response rate


hypoxia-inducible factor


platelet-derived growth factor


carbonic anhydrase IX




Response Evaluation Criteria in Solid Tumors


overall survival


Metastatic RCC (mRCC) has historically been considered one of the most treatment-resistant malignancies [1]. A growing understanding of the underlying molecular biology of clear-cell RCC, the most common histological subtype in RCC, identified a central role of the von Hippel-Lindau (VHL) gene in RCC pathogenesis. A decrease or an absence in functional VHL protein (pVHL) leads to the accumulation of hypoxia-inducible factor (HIF), which in turn leads to the induction of several hypoxia-regulated proteins such as vascular endothelial growth factor (VEGF), platelet-derived growth factor (PDGF), carbonic anhydrase IX (CAIX) and many others that can promote angiogenesis, tumour growth and invasion [2].

Targeting the VHL-HIF-VEGF pathway by humanized anti-VEGF monoclonal antibody such as bevacizumab, or small-molecule tyrosine kinase inhibitors of VEGF and PDGF receptors such as sunitinib and sorafenib, has yielded considerable antitumour effects and has led to the general acceptance of these therapies in advanced RCC [3–5]. However, as new agents with alternative molecular targets emerge in RCC therapy, further investigation is warranted of patient and tumour characteristics that might allow for the selection of patients who can benefit the most from VEGF-targeted therapy.

CAIX is a cytosolic transmembrane protein implicated in regulating cell proliferation in response to hypoxia, and is expressed in the vast majority of clear-cell RCCs but absent in normal kidney tissue [6]. Low CAIX expression (present in ≤85% of the tumour cells) as determined by immunohistochemistry (IHC), was associated with a three-fold increased death rate and a less favourable response to high dose interleukin-2 [7,8]. In the new era of VEGF-targeted therapy, the association of CAIX expression with outcome is unknown. Moreover, CAIX expression is mediated by the HIF transcriptional complex and it has been shown to correlate with VHL inactivation [9,10]. Preliminary work by Choueiri et al.[11] showed that patients treated with sorafenib responded to VEGF-targeted therapy only if their tumour had inactivation of VHL (by mutation or promoter hypermethylation). Although there was no similar correlation for patients treated with sunitinib in the previous study, a separate study showed that high expression of HIF-1α and -2α in RCC tumours correlates with response to sunitinib therapy [12]. Thus, the predictive value of the expression of the VHL–HIF-regulated protein CAIX with respect to VEGF-targeted therapy is worth assessing.

Whilst there is much interest in molecular biomarkers, more conventional histological characteristics might have similar predictive value. Data from others [8,13] suggested that metastatic clear-cell RCCs either with papillary, or >50% granular or no alveolar features on histological examination, were unlikely to respond to high-dose interleukin-2. Whether there might be similar histological features that are predictive of the response to VEGF-targeted therapies has not yet been investigated rigorously.

Towards this end, we examined the clinical and pathological characteristics, CAIX expression and clinical outcome in patients with clear-cell mRCC receiving VEGF-targeted therapy, to investigate the prognostic and predictive utility of CAIX expression and pathological features in patients treated with these therapies.


All patients with clear-cell mRCC who received therapy with sunitinib, sorafenib, valatinib (PTK787), and bevacizumab at Dana-Farber Cancer Institute and Beth Israel Deaconess Medical Center were considered for evaluation. Patients were part of an institutional review board-approved protocol for data and tissue to use for clinical research. The outcomes recorded as part of a prospectively collected database including clinical and histological characteristics [14,15]. Tumour shrinkage, objective response rate (RR) using Response Evaluation Criteria in Solid Tumors (RECIST) criteria, time on active therapy or treatment duration, and overall survival (OS) were recorded. Both RR and tumour shrinkage were rigorously documented. Patients could have been treated previously with cytokines but not with a VEGF-targeted agent. All available frozen or formalin-fixed, paraffin-embedded pretreatment RCC tumour material for each patient (from the nephrectomy specimen or from a biopsy if the nephrectomy specimen was not available) was retrieved. The current investigation received a waiver from the institutional review board at both institutions.

The pathology slides were centrally reviewed by one pathologist (S.S.) who was unaware of both the previous pathology interpretation and the therapeutic response. A detailed histological review of each case reported the Fuhrman grade, the percentage of alveolar, tubular, papillary and cystic patterns, the clear-cell component percentage, and the percentage of light and dense eosinophilic cytology (Fig. 2).

Figure 2.

Examples of tumour areas predominantly constituted by cells with clear (A), light eosinophilic (B), and dense eosinophilic (C) cytoplasm.

Tissue sections for CAIX were stained using IHC methods in an autostainer system (Dako, Carpinteria, CA, USA). Briefly, sections were dewaxed, soaked in alcohol, and after microwave treatment in antigen-unmasking solution for 10 min, incubated in 3% hydrogen peroxide for 15 min to inactivate endogenous peroxidase. Sections were then incubated with the mouse monoclonal antibody MN-75 (1 : 10 000 dilution) [16] and the reaction detected using the LSAB+TM detection kit (Dako). The antibody staining was assessed semiquantitatively by one pathologist (S.S.) unaware of the clinicopathological variables. Each specimen was scored based on the staining intensity and the percentage of positive cells. As described by Bui et al.[7], specimens in which >85% of cells stained for CAIX were labelled as high-expressing tumours, whereas those in which ≤85% of cells expressed CAIX were labelled as low-expressing tumours (Fig. 1).

Figure 1.

IHC staining of RCC tissue using the MN-75 anti-CAIX antibody. A, representative of a low CAIX staining; B, representative of high CAIX staining.

The primary endpoint was to detect any difference in outcome, measured primarily by tumour shrinkage, between sorafenib and sunitinib depending on CAIX status. Patient and disease characteristics were summarized descriptively; continuous clinical variables were dichotomized using thresholds previously reported, and for convenience the percentages of each pathological characteristics were reported either as >or ≤50%, or absent vs present. Patients were also categorized according to previously described pathological criteria [13]. The associations of CAIX status with patient and disease characteristics and with centrally assessed pathological features were assessed using Fisher’s exact test and extensions for categorical variables, or Wilcoxon rank-sum tests. The association of CAIX status with tumour shrinkage and with objective response used the Wilcoxon rank-sum and Fisher’s exact tests, respectively; multivariable analysis of tumour shrinkage used linear regression generalized estimating equation and robust sems because of some non-normality in the distribution. Associations of CAIX status with treatment duration and OS were assessed using log-rank tests and distributions were estimated using the Kaplan-Meier method. Linear generalized estimating equation or Cox regression models were fitted to examine the interaction of treatment and CAIX status, to test for evidence of heterogeneity in the predictive value of CAIX for the two different treatments (sunitinib and sorafenib).


We identified 118 patients with mRCC initiating treatment between August 2001 and November 2007 with sunitinib, sorafenib, bevacizumab and valatanib as their first VEGF-targeted agent. Eighteen patients were excluded based on tissue unavailability and another six because they had papillary RCC histology. In all, 94 patients with clear-cell RCC, available tissue at baseline and follow-up information were included in this analysis. Excluded (24) and included (94) patients did not differ in patient and disease characteristics. The median (range) duration of follow-up was 21 (1–72) months.

The patients included in the analysis all had tumours of clear-cell histology, 91% had previous nephrectomy and most had a good performance status (Table 1); 40% were previously treated (mostly with cytokines or vaccines). The vast majority of patients were treated with sunitinib (39, 41%) or sorafenib (40, 43%).

Table 1. 
Patient and disease characteristics at the start of treatment
Factorn (%)
  • *

    Cytokines, vaccines, chemotherapy, thalidomide.

  • †2002 definition; of 10 unknown, five were favourable-intermediate and five were intermediate-poor.

 Sunitinib39 (41)
 Sorafenib40 (43)
 Bevacizumab13 (14)
 PTK787 2 (2)
 Male62 (66)
 Female32 (34)
Age, years 
 ≤6074 (60)
 >6049 (40)
Eastern Cooperative Oncology Group performance status 
 051 (54)
 132 (34)
 2 11 (12)
Previous systemic therapy* 
 No56 (60)
 Yes38 (40)
Previous radiotherapy 
 No65 (69)
 Yes29 (31)
Previous nephrectomy 
 No 8 (9)
 Yes86 (91)
No. sites of disease 
 ≤258 (62)
 >236 (38)
Haemoglobin level 
 <Lower limit of normal41 (44)
 ≥Lower limit of normal48 (51)
 Unknown 5 (5)
Corrected calcium, mg/dL 
 ≤1077 (82)
 >10 6 (6)
 Unknown 11 (12)
Lactate dehydrogenase 
 ≤1.5 times upper limit of normal58 (62)
 >1.5 times upper limit of normal26 (28)
 Unknown10 (11)
Memorial Sloan Kettering Cancer Center risk factors 
 Favourable 8 (9)
 Intermediate48 (51)
 Poor28 (30)
 Unknown10 (11)

High CAIX expression was found in 65 tumours (69%); CAIX expression was not associated with clinical characteristics, including risk factors (P= 0.22), number of metastatic sites (P= 0.14) or sites of metastases, except for a possible association with bony metastases (P= 0.06; data not shown). In an attempt to determine the effect on CAIX expression when a tumour metastasizes, the level of CAIX expression in the metastatic lesion and the primary tumours were compared. Twenty-three patients (24%) had resection or biopsy of a metastatic lesion. There was no association of CAIX expression with specimen site assessed (high CAIX in 17/23, 74%, metastatic specimens, and in 48/71, 68%, kidney specimens; P= 0.62).

The centrally assessed pathological features of patients’ tumours are summarized in Table 2; most tumours (71%) were Fuhrman Grade 3 or 4. Tumours were classified as good (28%), intermediate (38%), poor (31%), or unassessable (3%) according to previous pathological criteria [13]. Of the patients, 48% had ≤25% clear-cell component in their tumours, compared to 27% with 25–50% and 23% with >50% clear-cell component. Tumours from 32 patients (34%) consisted of >50% cells with light eosinophilic cytoplasm, and tumours from 24 patients (26%) consisted of >50% cells with dense eosinophilic cytoplasm. Figure 1 shows examples of tumour areas with predominant clear (A), light eosinophilic (B), and dense eosinophilic (C) cell component. Sarcomatoid features were present in only five patients (5%). Sixty-seven patients (71%) had >50% alveolar features and only four (4%) had any papillary features. High CAIX expression was associated with a higher clear-cell component in the tumour (P= 0.02), a higher percentage of alveolar pattern (P= 0.08) and a lower dense eosinophilic cell component (P= 0.06), but not with the pathological criteria [13] (P= 0.42).

Table 2. 
Centrally assessed pathological features of the RCC tumours according to CAIX status of the tumour
FeatureLow CAIXHigh CAIXAll
n (%):
Highest Fuhrman grade   
 2 1 (3)13 (20)14 (15)
 2/3 3 (10) 8 (12) 11 (12)
 3–425 (86)42 (65)67 (71)
 Not assessable 2 (3) 2 (2)
Alveolar pattern   
 ≤50%10 (34)14 (22)24 (26)
 >50%19 (66)48 (74)67 (71)
 Not assessable 3 (5) 3 (3)
Tubular pattern   
 ≤50%29 (100)59 (91)88 (94)
 >50% 0 3 (5) 3 (3)
 Not assessable 3 (5) 3 (3)
Papillary pattern   
 Absent29 (100)58 (89)87 (93)
 Present 0 4 (6) 4 (4)
 Not assessable 3 (5) 3 (3)
Cystic pattern   
 Absent29 (100)62 (95)91 (97)
 Not assessable 3 (5) 3 (3)
Solid pattern   
 Absent18 (62)49 (75)67 (71)
 Present 11 (38)13 (20)24 (26)
 Not assessable 3 (5) 3 (3)
Sarcomatoid features   
 Absent27 (93)60 (92)87 (93)
 Present 2 (7) 3 (5) 5 (5)
 Not assessable 2 (3) 2 (2)
Rabdoid features   
 Absent27 (93)61 (94)88 (94)
 Present 2 (7) 2 (3) 4 (4)
 Not assessable 2 (3) 2 (2)
Light eosinophilic cytology   
 ≤50%19 (66)41 (63)60 (64)
 >50%10 (34)22 (34)32 (34)
 Not assessable 2 (3) 2 (2)
Dense eosinophilic cytology   
 ≤50%18 (62)50 (77)68 (72)
 >50% 11 (38)13 (20)24 (26)
 Not assessable 2 (3) 2 (2)
Clear cell cytology   
 <25%19 (66)26 (40)45 (48)
 25–50% 5 (17)20 (31)25 (27)
 >50% 5 (17)17 (26)22 (23)
 Not assessable 2 (3) 2 (2)
Pathological criteria [13]   
 Good 7 (24)19 (29)26 (28)
 Intermediate10 (34)26 (40)36 (38)
 Poor12 (41)17 (26)29 (31)
 Not assessable 3 (5) 3 (3)

The median (interquartile range) tumour shrinkage for all patients was −11 (−29 to +4)%. There were responses as assessed by RECIST in 21 (22%) patients (all partial responses) and stable disease in 53 (56%). Seventy-two patients had ceased treatment, and the median treatment duration was 7.8 months. The median OS was 27 months from treatment initiation and 37 patients had died.

Patients with high vs low tumour CAIX expression had mean tumour shrinkages of −12% and −5%, respectively (P= 0.38, Wilcoxon rank-sum test). Similarly, patients with high CAIX expression did not have a significantly higher partial RR (23% vs 21%, P= 1.0), treatment duration (9.4 vs 5.5 months, P= 0.23) or OS (2.8 vs 1.7 years, P= 0.43) than those with low CAIX expression.

In multivariable modelling, a higher clear-cell component, but not CAIX status, was associated with greater tumour shrinkage (P= 0.02 and 0.9, respectively); the presence of rhabdoid features (P= 0.06) and ≤50% light eosinophilic cytology (P= 0.05) were also associated with greater tumour shrinkage. A higher clear-cell component was also associated with a partial response to treatment (P= 0.02), longer treatment duration (P= 0.02), but not with OS (P= 0.57).

We also examined whether there was a better threshold value for CAIX than the 85% proposed by Bui et al.[7], but did not find one that was a better discriminator of treatment outcome (data not shown).

To test the hypothesis that CAIX expression might have a different predictive value for different VEGF-targeted treatments, we investigated the association of CAIX status with tumour shrinkage within the subsets of patients who received either sunitinib (39) or sorafenib (40). There was heterogeneity in tumour responsiveness to sunitinib or sorafenib according to CAIX status (P= 0.055 for interaction); the mean shrinkage was −17% vs −25% (mean difference +8%, 95% CI −14% to +31%) for sunitinib-treated patients with high vs low tumour CAIX expression, compared to −13% vs +9% (mean difference −22%, 95% CI −42% to −1%) for sorafenib-treated patients. This did not translate into a differential association either with treatment duration or with overall survival, although the few events limit that analysis. Table 3 summarizes treatment outcome for sunitinib and sorafenib, based on CAIX staining status.

Table 3.  Treatment outcome for sunitinib- and sorafenib-treated patients according to tumour CAIX staining status
VariableSunitinibSorafenibP, interaction
  1. A hazard ratio of <1.0 represents a longer duration for patients with high CAIX-expressing tumours.

No. of patients  9 3015 25 
Mean tumour shrinkage, %−25−17+9−130.055
Partial response, n (%)  4/9 10 (33) 1 (7)  3 (12)
Treatment duration    0.590
 N ceased  5 1714 22 
Median, months  7.6  9.1 3.9  8.6 
Hazard ratio (95% CI)   1.16 (0.43–3.14)   0.80 (0.41–1.58) 
OS    0.310
Number died  1  7 4 14 
Median, monthsNot reached 3615.6Not reached 
Hazard ratio (95% CI)   1.11 (0.12–9.95)   0.47 (0.20–1.10) 


Clear-cell RCC histology encompasses the vast majority of patients with RCC and is characterized by VHL inactivation through a biallelic gene loss. The consequences of this genetic event include the overproduction of VEGF by tumour cells even in the absence of hypoxia, and promotion of angiogenesis, tumour growth and metastasis. Large randomized phase III trials of agents directed against the VEGF pathway showed significant clinical activity in RCC [17]. However, these trials provide a general direction for the average patient and do not identify the optimum treatment for a specific individual. Patients’ baseline clinical features and tumour characteristics might greatly influence the outcome to VEGF-targeted therapy. In particular, up to 30% of patients might not benefit from these therapies and show progression at their first re-staging imaging [3,4,18]. Currently, patients with clear-cell mRCC are stratified by clinical prognostic factors developed by Motzer et al.[19] from patients treated with immunotherapy and generally, patients with a good and intermediate prognosis are offered front-line therapy with sunitinib, sorafenib and bevacizumab (with interferon-α), while patients with poor prognosis are offered the mTOR inhibitor temsirolimus. Although treatment selection based on these clinical criteria is helpful, it is largely imperfect and previous prognostic models might not be applicable to these novel therapies [15]. Ideally, a patient selection scheme should be based on pathological, genetic and molecular features of the patient’s tumour. At present there are no known histological or molecular predictors of benefit from VEGF-targeted therapy.

As discussed earlier, previous studies have shown that certain histological features and CAIX expression can predict the response to high-dose interleukin-2. Similarly, VHL inactivation and expression of HIF-1α and HIF-2α (measured by Western blotting) have been shown to predict the response to sorafenib and sunitinib, respectively. Therefore, in the present study, we sought to evaluate the prognostic utility of CAIX expression and histological features in patients treated with VEGF-targeted therapy, and to specifically investigate their predictive utility among patients treated with sunitinib and sorafenib.

Our results suggest that CAIX expression does not predict the response in a series of 94 patients with advanced RCC treated predominantly with sunitinib or sorafenib. We chose change in tumour size, measured on a continuous scale (tumour shrinkage), as the primary endpoint in this analysis, as RECIST-defined RRs are not uniform for the anti-VEGF agents considered. Sunitinib RRs are 30–40% while sorafenib and bevacizumab have RRs of 2–10%[2]. However, all these agents produce tumour shrinkage in a similar percentage of patients, and therefore tumour shrinkage probably represents a more uniform indicator of potential activity for VEGF-targeted therapy in RCC. Interestingly, only in sorafenib-treated patients was there an association of high tumour CAIX expression with a superior tumour shrinkage rate, supporting the initial observation that sunitinib, in contrast to sorafenib, is active in tumours with wild-type VHL (and probably low CAIX expression), and suggesting a predictive value of tumour CAIX expression for sorafenib therapy. A preliminary analysis from a small subset of patients who had received either sorafenib or placebo as part of the TARGET trial [20], a randomized phase III trial of sorafenib vs placebo in patients whose disease had progressed after cytokine therapy, showed a benefit of sorafenib in terms of tumour shrinkage and progression-free survival relative to placebo in tumours with high CAIX expression. The mechanism by which CAIX expression predicts a better outcome in patients with higher expression is unknown, but it is possible that responses in sorafenib-treated patients might depend on an inactivated pVHL which translates into higher CAIX expression. This can render a tumour more VEGF-dependent and consequently more susceptible to a weaker VEGF receptor blocker such as sorafenib. However, sunitinib might have additional non-VHL-related inhibitory mechanisms on RCC tumour and/or endothelial cells, or possibly a superior inhibition of the VEGF receptor with less dependence on the tumoral VHL status and consequently CAIX expression.

Based on our previous experience assessing the detailed histological features described by Upton et al.[13] in the RCC tumour specimens, we also sought to identify pathological predictors that can distinguish patients with a favourable outcome from VEGF-targeted agents. Previous experience with immunotherapy suggested that responses to interleukin-2 are mostly restricted to patients with an alveolar phenotype, with little efficacy in patients with tumours showing papillary or granular features. In the present study there were no pathologic features that clearly predicted a better outcome overall. However, there was an apparent direct correlation between the extent of the tumour clear-cell component and the degree of tumour shrinkage, likelihood of a partial response, and duration on VEGF-targeted therapy. This could be explained by the fact that patients with tumours expressing more clear-cell features might have complete inactivation of VHL and loss of the protein product, possibly due to ‘loss of function’ mutations. In the study by Choueiri et al.[15], loss of function mutations were shown to give better responses even after adjusting for clinical adverse factors on multivariable analysis.

VEGF-targeted therapy has substantial antitumour activity and is now considered a standard therapy for patients with advanced RCC. Analysis of predictive and prognostic histological and molecular features of tumours in patients receiving VEGF-targeted therapies might provide an understanding of the mechanism of response and resistance, help to identify patients most likely to benefit from specific therapies, and potentially identify patient populations for whom new or different therapeutic approaches are necessary. Ongoing studies are focused on analysing additional tumour tissue from an expanded cohort of patients at our institutions, as well as patients treated on large clinical trials, with the goal to better understand the biology of response to VEGF-targeted agents in metastatic RCC.


Supported in part by the Dana-Farber/Harvard Cancer Center Kidney SPORE NCI P50CA101942. The authors acknowledge Ashley Brick and Miranda Rogers from Dana-Farber Cancer Institute, and Christabel Kwabi, Karishma Shah and Andrew Percy from Beth Israel Deaconess Medical Center, for their assistance in collecting renal cancer specimens and their help in creating the database.


None declared.