Membrane carbonic anhydrase IX (CAIX) expression and relapse risk in resected stage I-II non-small cell lung cancer

Abstract

Background

Adjuvant chemotherapy reduces recurrences of non-small cell lung cancer (NSCLC). To determine which patients need adjuvant chemotherapy, we assessed factors associated with time to relapse (TTR).

Methods

In 230 resected stages I-II NSCLCs we correlated immunohistochemistry (IHC) scores for factors associated with cell growth rate, growth regulation, hypoxia, cell survival, and cell death with TTR.

Results

With a median follow-up of 82 (1-158) months for those alive and relapse-free at last follow-up, median time to recurrence was not reached. The 2- and 5-year probabilities of maintaining freedom from recurrence were 80.7% (95% confidence interval [CI]:(75.3%, 86.4%)) and 74.6% (95% CI:(68.6%, 81.2%)), respectively. TTR curves flattened at an apparent cure rate of 70%. In multicovariate Cox models, factors correlating with shorter TTR were membranous carbonic anhydrase IX (mCAIX) staining (any vs none, hazard ratio [HR]=2.083, p=0.023) and node stage (N1 vs N0, HR=2.591, p=0.002). mCAIX scores correlated positively with tumor size, grade, squamous histology, necrosis, mitoses, Ki67, p53, nuclear DNMT1 and cytoplasmic SHARP2, and correlated inversely with papillary histology, EGFR mutation (trend), CTR1, and cytoplasmic HIF-1α, VEGF, DNMT1, and ERCC1.

Conclusion

Nodal stage and mCAIX IHC were the strongest independent predictors of shorter TTR in resected NSCLCs. mCAIX correlated with tumor size, markers of tumor proliferation and necrosis, and tumor genetic characteristics, and paradoxically correlated inversely with the hypoxia markers HIF-1α and VEGF. Presence of mCAIX could help determine patients with high-risk of recurrence who might require adjuvant chemotherapy.

Introduction

Lung cancer is the world's leading cause of cancer death,¹ and US lung cancer 5-year relative survival rate has only increased from 12% in 1975–77 to 16% currently.² The poor prognosis of lung cancer is due in part to a high proportion of patients presenting initially with advanced disease, but even patients with operable early stage disease are at moderately high risk of relapse.³ Adjuvant chemotherapy reduces the probability of relapse after lung cancer resection, but we are currently not able to accurately determine which patients actually need adjuvant therapy. In patients with resected stages I and II non-small cell lung cancer (NSCLC), we assessed association with time to relapse (TTR) and overall survival (OS) of various tumor markers related to tumor growth rate, growth control, tumor cell survival and death, and hypoxia. In defining clinically important biomarkers to predict tumor biological behavior, OS is a more precise endpoint than TTR but it has the major disadvantage that it is affected by several factors unrelated to tumor biology, including patient age, comorbidities and therapy details.⁴ Hence, our primary objective was to define factors associated with TTR (defined as the time from surgery to tumor recurrence, with patients censored at time of last follow-up, death from other causes, or development of a new primary malignancy (in lung or any other site) that was associated with metastases, if they remained free of evidence of recurrence of their initial primary lung cancer at that time).

Materials and Methods

Using an IRB-approved laboratory protocol, we accessed the University of Texas/MD Anderson Cancer Center Lung SPORE Tissue Bank and selected resected archival formalin-fixed, paraffin-embedded (FFPE) tumor samples from 230 stage I-II NSCLC patients with squamous cell (n=87) or adenocarcinomas (n=143) who had given informed consent at the time of tissue collection and who had not received any adjuvant or neoadjuvant chemotherapy or radiotherapy.

Tissue microarrays were prepared using standard methods:⁵ a biopsy needle was inserted into each FFPE tumor specimen 3 times to obtain 3 tissue cores, each measuring 1 mm in diameter by 2–3 mm long. Based on prior assessment of H&E slides, the TMA cores were obtained preferentially from areas with high tumor cell content and with minimal necrosis or fibrosis, with one sample taken from around the center of the tumor, one from the periphery, and one from an intermediate area. Serial sections were cut from the TMA constructed from these cores and mounted on glass slides. Slides were deparaffinized in Xylene for 10 minutes 3 times. The tissue sections were hydrated in graded ethanols 100%, 90%, 70% and 50% for 5 minutes each time. Heat-induced epitope retrieval was performed in a Dako antigen retrieval bath at 121°C for 30 minutes and 90°C for 10 minutes using a Decloaking chamber (Biocare), followed by a 30 min cool down. Prior to antibody immunostaining, endogenous peroxidase activity was blocked with 3% H₂O₂ in methanol for 30 min. In order to block non-specific antibody binding, tissue sections were incubated with 10% FBS serum in TBSt for 30 min. Incubation with primary antibodies is as presented in Supplementary Table 1. This was followed by incubation with Envision plus labeled polymer, anti-rabbit-HRP antibody (Dako) for 30 min at room temperature. Staining development was performed with DAB, with timed monitoring using a positive control sample. The slides were then counterstained with hematoxylin, dehydrated, cleared and mounted.

Experienced lung pathologists (MIN and IIW, blinded for patient outcome) then manually recorded percent of tumor cells staining with 0 (absent), 1+ (mild), 2+ (moderate) and 3+ (strong) intensity. As appropriate, factor staining intensity was assessed for one or more of tumor cell nucleus, cytoplasm and membrane. Immunohistochemistry (IHC) scores (0–300) were then calculated for each relevant cell region by multiplying stain intensity (0–3) by percent tumor cells staining with each intensity. For each tumor specimen, results from the 3 cores were then averaged. If for a given patient results for only a single core were evaluable by IHC, then that single value was used, while the results for 2 cores were averaged if only 2 of the 3 cores were evaluable.

Molecular factors that we assessed included nuclear factors (p53, p21^WAF1/CIP1 and Ki67), cytoplasmic factors (Cyclo-oxygenase-2 [COX2] and Decoy Receptor-2 [DcR2]); nuclear and cytoplasmic factors (Copper Transporter-1 [CTR1], DNA Methyltransferase 1 [DNMT1], Hypoxia Inducible Factor-1α [HIF1α], Retinoblastoma [Rb], phospho-Rb, Enhancer-of-Split-and-Hairy-Related Protein [SHARP2, also known as “differentially expressed in chondrocytes protein-1”/DEC1], Survivin, Vascular Endothelial Growth Factor [VEGF], p14^ARF, p16^INK4 and Excision Repair Cross-Complementing Rodent Repair Deficiency, Complementation Group 1 [ERCC1]); and cytoplasmic and membrane Carbonic Anhydrase IX (CAIX) and Transforming Growth Factor-β (TGF-β). We also defined number of apoptotic cells (by TUNEL assay) per 10 high-powered fields, and for a subset of patients we had information on mutation status for epidermal growth factor receptor (EGFR) and Kras genes.

Tumor specimens were assessed by pathologists IIW and MIN for tumor histopathologic type, percent of adenocarcinomas made up of acinar, lepidic, mucinous, papillary, solid and micropapillary regions, number of mitoses per 10 high-powered fields, presence of necrosis, and invasion of pleura or lymphovascular structures. Number of nodes examined, number of positive nodes, tumor diameter, and pathologic stage were also recorded.

Statistical Methods

Patient gender, age, race, and smoking history information was recorded and summarized by descriptive statistics. Continuous biomarkers were summarized by mean, standard deviation, median and range. The difference of biomarkers between/among patient characteristic groups was tested by Wilcoxon rank sum test or Kruskal-Wallis test, when appropriate. Correlation between biomarkers was assessed using Spearman correlation analysis. Two experienced clinicians (DJS and CB) reviewed patient records and scans to assess TTR and overall survival (OS) from surgery.

For the purposes of this study, TTR was defined as the time from surgery to relapse or last follow up (LFU), with censoring at LFU, at death from other causes, or at diagnosis of a second primary associated with metastases if the patient was clinically relapse-free from their initial NSCLC. For example, if a patient was diagnosed with colorectal cancer, with intraabdominal nodal and liver metastases, this was interpreted for the purposes of this study as representing metastatic colorectal cancer unless there were compelling clinical, radiological or histopathologic data to indicate that it did instead indicate recurrent lung cancer, and the patient was censored for lung cancer recurrence at the time of diagnosis of the colorectal cancer. Similarly, while development of more than one lung nodule was generally interpreted as recurrence of the original lung cancer, development of a solitary lung nodule or mass that appeared clinically and radiologically to be more likely a new lung cancer primary was coded as a new primary rather than a recurrence. For example, a nodule or mass developing near a resection margin was coded as a recurrence of the original lung cancer while development of a single spiculated nodule in a different lobe was interpreted as a new primary. The investigators were blinded with respect to biological markers when making these designations.

Patients who died of other apparent causes and who had not had clinically suspected or confirmed relapse by the time of death were censored for TTR at their last relapse-free follow-up if they had not had sufficient evaluation shortly prior to their death to conclude with reasonable clinical certainty that they were free of relapse at the time of death. Again, this designation was done in a blinded fashion. We recognized that in both this designation and in the designation of recurrence vs new primary we would miss some recurrences, but chose to err on the side of being more certain of recurrence vs being more certain about lack of recurrence.

TTR and OS were estimated using the Kaplan-Meier method, and the log-rank test was employed to compare TTR between mCAIX staining groups. Cox models were used to assess factors associated with TTR. Only the significant factors with p<0.05 from univariate Cox model were included in the multicovariate Cox model, and backward selection was used to eliminate the ones that were not significant. P-value less than 0.05 was considered as significant. The analysis was performed using SAS 9.3(SAS Institute Inc., Cary, NC, USA). Factors associated with TTR in multicovariate analyses were then tested for their association with OS.

Results

The median age of the 230 patients was 68.2 (range, 33.5 – 90.3) years, the majority were Caucasian (214/230, 93.0%), 122 (53.0%) were female and 108 were male (47.0%), 92 (40.0%) were current smokers, 103 (44.8%) were former smokers and 35 (15.2%) were never smokers, and 182 (79.1%) had stage I and 48 (20.9%) had stage II NSCLC. Patient characteristics are outlined in Supplementary Table 2 and tumor IHC scores are presented in Supplementary Table 3. Median (range) follow-up for TTR was 82 (1-158) months for those who were alive and remained relapse-free at last follow-up. Median (range) follow-up for OS for those who remained alive at last follow-up was 80 (1-158) months. A total of 52 patients had confirmed recurrence and 113 patients died. TTR for the population is presented in Supplementary Figure 1, and OS is presented in Supplementary Figure 2, with the median TTR not being reached and with median OS being 79 months. The 2- and 5-year probabilities of maintaining freedom from recurrence were 80.7% (95% confidence interval [CI]:(75.3%, 86.4%)) and 74.6% (95% CI:(68.6%, 81.2%)), respectively. When TTR was replotted as an exponential decay (log-linear) curve and subjected to nonlinear regression analysis as previously described,⁶ those on the terminal flat portion of the curve (the “cured” fraction) constituted 70% of the overall population, and the estimated half-life to relapse for the 30% relapsing was 20 months (Supplementary Figure 3).

In Table 1 are presented univariate Cox model analyses for impact of patient and tumor variables on TTR. Nodal stage (N1 vs N0), higher number of nodes positive, larger tumor diameter, and pathological stage (II vs I) each correlated with short TTR. Among the IHC markers, TTR was significantly shorter if any membrane CAIX (mCAIX) staining was detected compared to patients whose tumors had no detectable mCAIX staining (Figure 1). The 24- and 60-month probabilities of freedom from recurrence were 75.1% (95% CI: (67.3%, 83.7%)) and 66.0% (95% CI: (57.3%, 76.0%)), respectively, for patients with any mCAIX staining; and 87.5% (95% CI: (80.2%, 95.5%)) and 84.4% (95% CI: (76.4%, 93.4%)), respectively, for patients with negative mCAIX staining (p=0.014, log-rank test).

Table 1.

Univariate Cox Model Analysis for Time to Relapse

Covariate	Estimate	Stand Error	Hazard Ratio	HR 95% CI	p
Histology: Adenocarcinoma vs Squamous	−0.187	0.285	0.830	0.474–1.452	0.513
Gender: Male vs Female	0.153	0.277	1.165	0.676–2.007	0.582
Age	0.026	0.014	1.027	0.999–1.055	0.062
Race: Caucasian vs others	0.544	0.721	1.723	0.419–7.086	0.451
Smoking: Ever vs Never	0.653	0.471	1.922	0.764–4.834	0.165
Smoking: Current vs Never	0.637	0.496	1.890	0.716–4.993	0.199
Former vs Never	0.668	0.490	1.951	0.746–5.098	0.173
Nodes: N1 vs N0	1.096	0.302	2.991	1.655–5.406	0.0003
No. nodes positive	0.383	0.106	1.467	1.191–1.808	0.0003
Tumor diameter (cm)	0.123	0.058	1.131	1.009–1.267	0.035
Pathologic Stage: II vs I	0.945	0.302	2.573	1.425–4.647	0.002
CAIX, membrane*	0.002	0.001	1.002	0.999–1.004	0.244
CAIX, cytoplasm*	0.002	0.002	1.002	0.999–1.005	0.249
COX2, cytoplasm*	0.001	0.001	1.001	0.998–1.004	0.502
CTR1, cytoplasm*	0.001	0.002	1.001	0.998–1.005	0.436
CTR1, nuclear*	0.000	0.002	1.000	0.997–1.004	0.873
DNMT1, cytoplasm*	−0.002	0.002	0.998	0.993–1.002	0.256
DNMT1, nuclear*	0.004	0.004	1.004	0.996–1.012	0.352
DcR2, cytoplasm*	0.004	0.002	1.004	0.999–1.008	0.134
ERCC1, cytoplasm*	−0.002	0.002	0.998	0.993–1.002	0.343
ERCC1, nuclear*	−0.001	0.002	0.999	0.995–1.002	0.558
HIF1α, cytoplasm*	0.001	0.002	1.001	0.998–1.004	0.544
HIF1α, nuclear*	0.007	0.006	1.007	0.995–1.019	0.245
Rb, cytoplasm*	−0.003	0.002	0.997	0.993–1.002	0.224
Rb, nuclear*	0.001	0.002	1.001	0.997–1.004	0.738
SHARP2, cytoplasm*	−0.003	0.003	0.997	0.992–1.002	0.219
SHARP2, nuclear*	0.002	0.002	1.002	0.999–1.005	0.262
SURVIVIN, cytoplasm*	0.002	0.002	1.002	0.998–1.005	0.426
SURVIVIN, nuclear*	0.002	0.001	1.002	0.999–1.005	0.24
TGFβ, membrane*	−0.004	0.003	0.996	0.991–1.002	0.16
TGFβ, cytoplasm*	−0.004	0.002	0.996	0.991–1.000	0.065
VEGF, cytoplasm*	−0.002	0.002	0.998	0.994–1.003	0.434
VEGF, nuclear*	0.005	0.008	1.005	0.989–1.021	0.556
p14ARF, cytoplasm*	−0.003	0.004	0.997	0.990–1.004	0.431
p14ARF, nuclear*	0.005	0.006	1.005	0.994–1.016	0.385
p16INK4, cytoplasm*	−0.002	0.001	0.998	0.996–1.001	0.268
p16INK4, nuclear*	−0.001	0.001	0.999	0.997–1.002	0.513
phospho Rb, cytoplasm*	0.000	0.004	1.000	0.991–1.008	0.923
phospho Rb, nuclear*	0.001	0.002	1.001	0.996–1.005	0.751
p21WAF1/CIP1, nuclear*	0.003	0.002	1.003	0.999–1.008	0.109
p53, nuclear*	−0.001	0.001	0.999	0.997–1.002	0.491
Ki67, nuclear*	0.002	0.002	1.002	0.998–1.005	0.377
CAIX, membrane, >0 vs 0	0.776	0.323	2.172	1.154–4.090	0.016
CAIX, cytoplasm, >0 vs 0	0.520	0.521	1.681	0.605–4.672	0.319
COX2, cytoplasm, >0 vs 0	0.192	0.474	1.212	0.478–3.072	0.685
CTR1, nuclear, >0 vs 0	0.091	0.313	1.095	0.593–2.022	0.772
DNMT1, cytoplasm, >0 vs 0	−0.200	0.310	0.818	0.446–1.503	0.518
DNMT1, nuclear, >0 vs 0	0.480	0.311	1.617	0.880–2.972	0.122
DcR2, cytoplasm, >0 vs 0	0.393	0.598	1.482	0.459–4.788	0.511
ERCC1, cytoplasm, >0 vs 0	−0.274	0.291	0.760	0.430–1.345	0.346
ERCC1, nuclear, >0 vs 0	−0.367	0.289	0.693	0.393–1.221	0.204
HIF1α, cytoplasm, >0 vs 0	−0.160	0.521	0.852	0.307–2.366	0.759
HIF1α, nuclear, >0 vs 0	0.036	0.341	1.037	0.532–2.021	0.916
Rb, cytoplasm, >0 vs 0	0.468	1.011	1.597	0.220–11.58	0.643
Rb, nuclear, >0 vs 0	0.343	0.368	1.409	0.686–2.896	0.351
SHARP2, cytoplasm, >0 vs 0	0.216	0.723	1.241	0.301–5.123	0.766
SHARP2, nuclear, >0 vs 0	−0.190	0.524	0.827	0.296–2.311	0.717
SURVIVIN, nuclear, >0 vs 0	−0.194	0.722	0.824	0.200–3.391	0.789
TGFβ, membrane, >0 vs 0	0.198	0.285	1.219	0.697–2.130	0.488
TGFβ, cytoplasm, >0 vs 0	−0.038	0.316	0.963	0.519–1.788	0.905
VEGF, nuclear, >0 vs 0	0.138	0.357	1.148	0.571–2.309	0.699
p14ARF, cytoplasm, >0 vs 0	−0.157	0.308	0.854	0.468–1.561	0.609
p14ARF, nuclear, >0 vs 0	−0.133	0.595	0.875	0.273–2.811	0.823
p16INK4, cytoplasm, >0 vs 0	−0.342	0.286	0.711	0.405–1.246	0.233
p16INK4, nuclear, >0 vs 0	−0.335	0.287	0.715	0.408–1.254	0.242
phospho Rb, cytoplasm, >0 vs 0	0.295	0.318	1.343	0.720–2.505	0.354
phospho Rb, nuclear, >0 vs 0	0.002	0.338	1.002	0.517–1.941	0.996
p21WAF1/CIP1, nuclear, >0 vs 0	−0.103	0.474	0.902	0.356–2.286	0.829
p53, nuclear, >0 vs 0	−0.076	0.281	0.927	0.535–1.607	0.787
Ki67, nuclear, >0 vs 0	−0.085	0.522	0.919	0.331–2.555	0.871

^*As a continuous variable.

Figure 1.

Microphotographs illustrating CAIX immunohistochemistry expression in malignant cells of lung cancer tissue specimens (×40). A, combined CAIX strong membrane and cytoplasmic expressions in a squamous cell carcinoma. B, strong CAIX membrane expression in a squamous cell carcinoma. C, moderate CAIX cytoplasmic expression in an adenocarcinoma. D, lack of CAIX expression in a poorly differentiated adenocarcinoma.

In multicovariate Cox model analysis, nodal stage (N1 vs N0) (HR=2.561 (95% CI: 1.409–4.766), p=0.002) and presence vs absence of mCAIX staining (HR=2.083 (95% CI: 1.104–3.929), p=0.023) emerged as independent prognostic variables for TTR (Table 2). If tumor diameter was forced into the model (HR=1.068 (95% CI: 0.928–1.231), p=0.36) mCAIX remained important (HR=1.923 (95% CI: 0.999–3.705), p=0.05). Figure 2a presents TTR curves for patients with vs without detectable mCAIX staining and with vs without node involvement. The 60-month probability of freedom from recurrence for mCAIX negative/N0 patients, mCAIX positive/N0 patients, mCAIX negative/N1 patients, and mCAIX positive/N1 patients was 0.866 (95% CI (0.783, 0.957)), 0.713 (95% CI (0.621–0.818)), 0.70 (95% CI (0.467, 1.0)) and 0.399 (95% CI (0.213, 0.748), respectively.

Table 2.

Multicovariate Cox Model Analysis for Time to Relapse

Variable	Parameter Estimate	Standard Error	p-value	Hazard Ratio	95% Hazard Ratio CI
Nodes (N1 vs N0)	0.952	0.311	0.002	2.591	1.409	4.766
CAIX. Membrane, >0 vs 0	0.734	0.324	0.023	2.083	1.104	3.929

Figure 2.

Kaplan Meier curves for relapse and survival by nodal and membrane carbonic anhydrase IX (mCAIX) staining status (E/N: no. events/no. cases). Figure 2a: Time to Relapse: The 60-month probability of freedom from recurrence was 0.866 (95% CI (0.783, 0.957)), 0.713 (95% CI (0.621–0.818)), 0.70 (95% CI (0.467, 1.0)) and 0.399 (95% CI (0.213, 0.748), for N0/mCAIX negative (red), N0/mCAIX positive (blue), N1/mCAIX negative (green) and N1/mCAIX positive (black) patients, respectively. Figure 2b: Overall Survival: The 60-month probability of being alive was 0.745 (95% CI (0.647, 0.858)), 0.639 (95% CI (0.0.550, 0.742)), 0.727 (95% CI (0.506, 1.0)) and 0.272 (95% CI (0.149, 0.497) for N0/mCAIX negative (red), N0/mCAIX positive (blue), N1/mCAIX negative (green), and N1/mCAIX positive (black) patients, respectively.

Five N0 patients (3 stage IA patients with tumors measuring 1.6–2.2 cm and 2 stage IB patients, both with tumors measuring 4.5 cm) were judged clinically and radiologically to have developed second primary cancers rather than recurrences. Of these, 2 were negative for mCAIX on IHC and 3 were positive. If they were excluded from the analyses, TTR continued to be significantly associated with mCAIX (HR=1.903 (95% CI: 1.114–3.250), p=0.0185). None of the 5 had their new primaries compared to their old ones histopathologically.

We then assessed characteristics that correlated with mCAIX IHC scores. In Table 3 are median mCAIX scores for different patient groups. mCAIX scores were significantly higher in squamous cell carcinomas than adenocarcinomas, and less differentiated tumors than better differentiated tumors, with a trend toward higher scores (p=0.074) in patients with EGFR wild type tumors compared to those with EGFR mutations. There was also a trend towards higher mCAIX scores in patients with higher stage tumors (p=0.063 for N0 vs N1).

Table 3.

Membrane carbonic anhydrase IX in different patient groups

Group	N	mCAIX* (mean ± standard deviation, median)	P value
Histopathologic Type:
Adenocarcinoma	133	60.4 ± 91.1, 0	<0.0001
Squamous cell carcinoma	77	118.0 ± 104.5, 103
Nodal Stage:
N0	169	75.6 ± 97.1, 30	0.063
N1	41	106.2 ± 108.4, 60
Pathological Stage:
I	164	75.3 ± 96.6, 30	0.077
II	46	103.9 ± 109.2, 60
Differentiation:
Well	47	63.2 ± 99.4, 0	0.043
Moderate	87	76.6 ± 96.2, 20
Poor	75	97.9 ± 103.6, 60
EGFR:
Mutation	12	27.5 ± 56.5, 0	0.074
Wild type	129	72.3 ± 95.4, 20
KRAS:
Mutation	12	82.1 ± 107.2, 35	0.88
Wild type	128	66.5 ± 91.7, 18.33
Gender:
Female	110	79.6 ± 102.4, 20	0.3879
Male	100	83.7 ± 97.6, 42.5
Race:
Caucasian	194	77.3 ± 96.2, 30	0.090
Other	16	133.5 ± 130.2, 72.5
Smoking History:
Current	84	81.0 ± 97.4, 37.5	0.605
Former	94	87.3 ± 104.3, 42.5
Never	32	66.1 ± 94.4, 15

^*mCAIX range for each category was 0–300

In Table 4 are correlations between mCAIX scores and other continuous variables. mCAIX scores correlated directly with tumor diameter, cytoplasmic SHARP2 scores, number of mitoses, tumor necrosis, and with scores for cytoplasmic SHARP2, nuclear Ki67, nuclear DNMT1, and nuclear p53 scores, and correlated inversely with percent of tumor that had papillary characteristics (among adenocarcinomas) and with scores for cytoplasmic HIF1α, cytoplasmic VEGF, cytoplasmic DNMT1, nuclear and cytoplasmic CTR1, cytoplasmic ERCC1, nuclear and cytoplasmic p16, and nuclear p14. mCAIX scores did not correlate with apoptosis or with scores for nuclear HIF1α, nuclear VEGF, SHARP2, or other pro-cell-survival or tumor suppressor gene-related factors.

Table 4.

Factors correlating with membrane carbonic anhydrase IX score

Factor*	N	Spearman coefficient	P value
Tumor diameter	204	0.243	0.0005
Adenocarcinoma composition (% of tumor made up by each histopathologic subtype):
% acinar	133	0.121	0.167
% lepidic	133	−0.126	0.148
% mucinous	133	0.018	0.836
% papillary	133	−0.212	0.014
% solid	133	0.094	0.282
% micropapillary	133	−0.067	0.44
Hypoxia-associated markers:
nHIF1α	208	−0.016	0.818
cHIF1α	208	−0.153	0.027
nVEGF	188	0.120	0.100
cVEGF	188	−0.143	0.051
nSHARP2	187	−0.044	0.551
cSHARP2	187	0.255	0.0004
Proliferation-associated markers:
Mitoses	209	0.203	0.003
nKi67	200	0.212	0.003
nDNMT1	201	0.192	0.006
cDNMT1	201	−0.240	0.0006
nCTR1	198	−0.208	0.003
cCTR1	198	−0.228	0.001
Cell death- or survival-associated markers:
Necrosis	209	0.260	0.0001
Apoptosis	195	0.060	0.408
cDcR2	191	0.006	0.929
nSURVIVIN	205	0.052	0.463
cSURVIVIN	205	−0.074	0.294
cTGFbeta	203	−0.135	0.055
mTGFbeta	203	−0.125	0.077
DNA repair and inflammation markers:
nERCC1	203	−0.049	0.49
cERCC1	203	−0.151	0.031
cCOX2	174	0.061	0.42
Tumor Suppressor Genes and Related Molecules:
nP53	209	0.157	0.024
nP21	196	0.081	0.256
nRB	204	−0.029	0.677
cRB	204	−0.125	0.074
nPhosphoRB	176	0.136	0.071
cPhosphoRB	176	−0.018	0.814
nP16	202	−0.14	0.043
cP16	202	−0.14	0.041
nP14	209	−0.135	0.051
cP14	209	−0.09	0.191

^*c: cytoplasmic m: membrane n: nuclear

IHC data for mCAIX were evaluable for all 3 cores for 63.2% of patients, for 2 cores for 23.4%, and for only 1 core for 13.4% of patients. Using data from patients in whom at least 2 cores were evaluable for mCAIX, we also assessed impact of heterogeneity of tumor mCAIX IHC expression on TTR. Whether just considering patients with 3 evaluable cores (37% mCAIX negative for all cores, 45% mCAIX positive for all cores and 19% heterogeneous with 1–2 cores negative and the remaining positive for mCAIX), or also adding in patients with just 2 evaluable cores (38% mCAIX negative in both cores, 45% positive in both cores and 17% positive in one and negative in the other), the TTR curve for heterogeneous patients was intermediate between the TTR curves for homogeneously negative patients and for homogeneously positive patients (p=0.0471, log-rank test for trend). If we separately assessed the patients with only a single evaluable core, 47% were negative for mCAIX and 53% were positive. Confidence intervals were wide, but TTR was significantly worse for positive patients than for negative patients (HR=6.413 (95% CI: 1.363–30.17), p=0.0187).

In patients with at least 2 evaluable cores, the proportion of patients with mCAIX heterogeneity was similar for patients with adenocarcinomas vs squamous carcinomas (17% vs 16%), while the proportion of patients with all cores being positive was lower in adenocarcinomas than in squamous carcinomas (35% vs 60%, p=0.0007). Similarly, although only 6% of patients with no necrosis noted in their samples had heterogeneity between cores compared to 17% of patients with at least some degree of tumor necrosis, this difference was not significant (p=0.49), while the proportion of patients who had all cores positive for mCAIX was higher in those with necrosis than in those without necrosis (48% vs 21%, p=0.0287).

While we were primarily interested in TTR, we then assessed whether the factors associated with TTR in multicovariate Cox models were also associated with OS. When nodal stage and mCAIX were taken together as the only factors considered for a multicovariate OS model, the HR for N1 vs N0 was 1.918 (95% CI (1.223, 3.008), p=0.0133) and the HR for mCAIX positive vs negative was 1.762 (95% CI (1.142, 2.720), p=0.0105). The 60-month probability of OS for mCAIX negative/N0 patients, mCAIX positive/N0 patients, mCAIX negative/N1 patients, and mCAIX positive/N1 patients was 0.745 (95% CI (0.647, 0.858)), 0.639 (95% CI (0.0.550, 0.742)), 0.727 (95% CI (0.506, 1.0)) and 0.272 (95% CI (0.149, 0.497), respectively (Figure 2b).

Discussion

In determining who should be considered for adjuvant therapies, it helps to be able to define those patients who are at highest risk of tumor relapse. In this study, we found that for patients with resected stage I-II NSCLC, there appeared to be a plateau on the TTR curve, with approximately 70% of patients projected to be on this “cured” plateau. TTR differs from “relapse-free survival” since our patients were censored if they died of other apparent causes, while either relapse or death from any causes is counted as a relapse-free survival event. We wished to maximize the probability of defining factors associated with tumor biology without contamination from factors associated with death from comorbidities or second primary malignancies. Patients who died of uncertain causes without recent reevaluation of relapse status were censored at the time of last evaluation for relapse if they were relapse-free at that time. Our approach (designed to improve biological and clinical relevance of our assessments) decreased statistical power by decreasing the number of evaluable “events” and by decreasing the length of follow-up for censored patients, and would have missed any relapses occurring after last follow-up.

In keeping with the well-established impact of stage on outcome,³ node involvement emerged as the most important predictor of relapse. Presence of membrane staining for CAIX was the only other factor that correlated with outcome in multicovariate analysis. While mCAIX correlated with both tumor size and with Ki67, mCAIX correlated independently with TTR while size and Ki67 did not.

CAIX mRNA^7,8 and protein (by IHC)^9–12 are frequently expressed in resected NSCLCs, with CAIX expression being particularly common in squamous cancers.^9,13 It is unknown whether this association of CAIX with squamous lung cancers is related to the frequent loss of the von Hippel-Lindau (VHL) tumor suppressor gene that is seen in NSCLC.¹⁴ VHL loss of heterozygosity (LOH)¹⁵ and methylation/downregulation¹⁶ are common in NSCLC, and occur in squamous cell carcinomas much more frequently than in adenocarcinomas. In other tumor types, mCAIX expression was noted in pheochromocytomas only if they were associated with VHL germline mutations,¹⁷ while in renal cell carcinomas, CAIX expression was perinecrotic in tumors with intact VHL systems but was diffuse in VHL-defective tumors.¹⁸

Tumor subtype may also be important. We found particularly low expression in lung adenocarcinomas with a high papillary component. Of interest, CAIX expression is also lower in papillary renal cell carcinomas than in clear cell kidney carcinomas,^19,20 and CAIX expression also varies with subtype in ovarian²¹ and breast^22–25 cancers.

CAIX catalyzes the reversible hydration of carbon dioxide to carbonic acid, and is upregulated in cancers to help maintain physiologic intracellular pH despite high glycolysis rates.²⁶ While maintaining a physiologic intracellular pH, it acidifies the extracellular space.²⁶ CAIX may promote a metastatic or invasive phenotype by reducing cell adhesion,²⁷ increasing cell invasiveness²⁸ and mobility and migration,²⁷ inducing angiogenesis,²⁷ and activating proteases.²⁷

CAIX expression in normal tissues is generally substantially less than in tumors.^7,8,12,13 Hypoxia leads to increased CAIX gene expression,²⁹ transcription,³⁰ and protein expression^31,32 in several types of tumor cell lines, and low glucose and low bicarbonate both increase CAIX transcription and protein expression in hypoxic cells.³⁰

In resected NSCLC, tumor oxygenation correlated negatively with CAIX IHC staining,³³ although there has not been consistent correlation between tumor hypoxia and CAIX expression in vivo,³² and for at least some tumor types, there may be both hypoxia-driven and hypoxia-independent CAIX signaling pathways.^34,35

In keeping with our findings, most (but not all³⁶) other NSCLC studies assessing CAIX IHC^{11,13,37–39} or mRNA^7,33,40 tumor expression or plasma levels³⁷ have also reported an association of high expression with worse overall^{11,13,33,37,38,40} or disease-free^7,13,33,39 survival, with the greatest negative impact on survival being noted in later stage NSCLCs and in squamous cell carcinomas.³³ While we found the strongest correlation between mCAIX staining and TTR, others had previously noted perinuclear CAIX staining to be particularly important.^11,38

High tumor cell CAIX expression has also been associated with worse prognosis in sarcomas,^41,42 gliomas,^28,43 neuroblastomas,⁴⁴ papillary renal cell carcinomas,⁴⁵ and carcinomas of the breast,^23,24,46,47 head and neck,⁴⁸ nasopharynx,⁴⁹ ovary,²¹ cervix,⁵⁰ and rectum,⁵¹ but CAIX expression did not correlate with outcome in meningiomas⁵² or in carcinomas of the pancreas⁵³ or prostate,⁵⁴ and was paradoxically associated with improved outcome in renal clear cell carcinomas.^19,55,56 High CAIX levels in tumor stromal cells may also be associated with poor outcome.^57,58

In our study, mCAIX expression correlated with tumor size, with markers of proliferation (including Ki67 and number of mitoses), with poor differentiation, and with necrosis (but not apoptosis), and we noted a trend (p=0.063) towards higher mCAIX expression in N1 vs N0 tumors. Other studies that included a variety of tumor types also noted a correlation of CAIX expression with tumor size,^47,50 mitoses,⁴⁴ Ki67,^10,43,47 lack of differentiation,^{13,23,25,43,46,47,58} necrotic^{11,23,31,39,59} or perinecrotic areas,^{18,31,43,59,60} and higher stage,^13,31,44 although some of these factors failed to correlate with CAIX expression in still other studies.^{9,20,21,46,50,55,61,62}

CAIX and VEGF are both target genes of the transcription factor HIF-1α^63,64 that is generally induced by hypoxia but that may also be induced by Src.⁶⁴ While mCAIX expression in our patients did correlate with cytoplasmic expression of the hypoxia-inducible factor SHARP-2, it did not correlate with HIF-1α or VEGF nuclear expression, and paradoxically correlated inversely with HIF-1α and VEGF cytoplasmic expression. Results have been variable in other studies. In various tumor types, CAIX expression has correlated with HIF-1α expression in some studies,^{39,43,47,49,52,65} but not in others.^61,65,66 HIF-1α may also lose its transcriptional ability (eg, through repression by p53) such that CAIX induction does not happen despite high HIF-1α expression.⁶⁷ Furthermore, CAIX expression may correlate with HIF-1α expression in tumors in which the HIF-1α expression is perinecrotic, but not in tumors in which HIF-1α expression is diffuse throughout the tumor.⁶⁰ Also, CAIX has a much longer half-life in tissues (about 38 hours³⁰) than does HIF-1α, and HIF-1α expression will rapidly decrease in tumor areas that have low nutrient levels, while CAIX will persist due to its longer half-life.⁶⁸ In addition, in the absence of hypoxia and HIF-1α, CAIX expression may be upregulated by high cell density via the PI3K pathway,⁶⁹ and increased expression of CAIX in the absence of hypoxia may also occur with hypomethylation of the CAIX gene promoter.⁷⁰

Similarly, some previous studies have found positive correlations between CAIX and VEGF expression,^33,43,52,59 while others have not,^34,50,71 and still others have found an inverse correlation between CAIX and VEGF expression,⁵⁶ similar to what we found. Lack of a consistent correlation between CAIX and VEGF may be due in part to the fact that upon reoxygenation of tissues, VEGF mRNA declines rapidly, while CAIX mRNA expression persists for >72 hours.⁵⁹

In addition to serving as a prognostic marker, CAIX could also potentially serve as a therapeutic target or as a predictive marker for efficacy of other therapies. Therapies targeting CAIX were effective in preclinical models,²⁷ and some are in early stages of clinical investigation.⁷² Moreover, high CAIX expression in NSCLC may be associated with decreased efficacy of radiotherapy,⁹ and interference with CAIX strongly augments the efficacy of both radiotherapy and some chemotherapy agents in preclinical systems.²⁸ Conversely, activity of targeted agents and cytokines may be augmented in renal cell carcinomas with high CAIX expression,⁷³ and tumor uptake and efficacy of some chemotherapy agents that are weak acids could potentially be augmented by the acidic milieu promoted by CAIX.⁷⁴

In summary, high CAIX expression is associated with poor prognosis across a wide range of tumor types, and we found that membrane CAIX expression in particular was associated with an increased risk of relapse in our stage I-II NSCLC patients. Based on these observations, it would be reasonable to assess mCAIX expression further as a prognostic factor in NSCLC, and given the range of studies that have found an association between CAIX expression and poor outcome in NSCLC, it would be reasonable in advanced NSCLC to assess efficacy of new investigational agents targeting CAIX.