Benefit to neoadjuvant anti-human epidermal growth factor receptor 2 (HER2)-targeted therapies in HER2-positive primary breast cancer is independent of phosphatase and tensin homolog deleted from chromosome 10 (PTEN) status

The benefit of neoadjuvant dual HER2 blockade with the combination of trastuzumab and lapatinib in HER2-positive breast cancer is independent of PTEN status determined by immunohistochemistry. The lack of standardization of PTEN scoring methodology and cut-off threshold may affect correlations with clinical end points.

Abstract

Background

Assessment of phosphatase and tensin homolog deleted from chromosome 10 (PTEN) might be an important tool in identifying human epidermal growth factor receptor 2 (HER2)-positive breast cancer patients unlikely to derive benefit from anti-HER2 therapies. However, studies to date have failed to demonstrate its predictive role in any treatment setting.

Patients and methods

Prospectively collected baseline core biopsies from 429 early-stage HER2-positive breast cancer patients treated with trastuzumab, lapatinib, or their combination in the Neo-ALTTO study were stained using two anti-PTEN monoclonal antibodies (CST and DAKO). The association of PTEN status and PI3K pathway activation (defined as either PTEN loss and/or PIK3CA mutation) with total pathological complete response (tpCR) at surgery, event-free survival (EFS), and overall survival (OS) was evaluated.

Results

PTEN loss was observed in 27% and 29% of patients (all arms, n = 361 and n = 363) for CST and DAKO, respectively. PTEN loss was more frequently observed in hormone receptor (HR)-negative (33% and 36% with CST and DAKO, respectively) compared with HR-positive tumours (20% and 22% with CST and DAKO, respectively). No significant differences in tpCR rates were observed according to PTEN status. PI3K pathway activation was found in 47% and 48% of patients (all arms, n = 302 and n = 301) for CST and DAKO, respectively. Similarly, tpCR rates were not significantly different for those with or without PI3K pathway activation. Neither PTEN status nor PI3K pathway activation were predictive of tpCR, EFS, or OS, independently of treatment arm or HR status. High inter-antibody and inter-observer agreements were found (>90%). Modification of scoring variables significantly affected the correlation between PTEN and HR status but not with tpCR.

Conclusion

These data show that PTEN status determination is not a useful biomarker to predict resistance to trastuzumab and lapatinib-based therapies. The lack of standardization of PTEN status determination may influence correlations between expression and relevant clinical end points.

Clinical Trials

This trial is registered with ClinicalTrials.gov: NCT00553358.

introduction

Human epidermal growth factor receptor 2 (HER2) is a receptor tyrosine kinase overexpressed in ∼25% of invasive breast carcinomas [1]. Blocking HER2-mediated signalling either with monoclonal antibodies [2, 3] or with small molecule kinase inhibitors [4, 5] is associated with significant clinical benefit and, as a consequence, several of these agents have become standard of care for treating HER2-positive breast cancer patients. Despite the potent anti-tumour activity of these agents, primary and acquired drug-resistance occurs in a significant proportion of patients [6]. Hyperactivation of the PI3K pathway by PIK3CA activating mutations or loss of the phosphatase and tensin homolog deleted from chromosome 10 (PTEN) have been associated with resistance to trastuzumab- and lapatinib-based therapies [7–12].

PTEN is a negative regulator of PI3K/AKT signalling and its loss has been observed in 13%–86% of HER2-positive breast cancers [11–17]. According to preclinical findings, assessment of PTEN might be an important tool in identifying patients unlikely to derive substantial benefit from trastuzumab and lapatinib-based therapies [8–10]. However, studies to date have failed to provide conclusive evidence on the predictive role of PTEN in HER2-positive breast cancer in either the neoadjuvant, adjuvant, or metastatic settings [12–18]. The lack of standardization in PTEN status determination in formalin-fixed paraffin-embedded (FFPE) tissue samples and the small data sets analysed in previous studies may have contributed to the reported high variability in PTEN loss rates and the conflicting results regarding its predictive role of anti-HER2 sensitivity.

In this study, we assessed the incidence of PTEN protein expression and its correlation with patient clinicopathologic features and response to therapy, measured by the rated of total pathological complete response (tpCR), event-free survival (EFS), and overall survival (OS) in HER2-positive breast cancer patients enrolled in the Neo-ALTTO trial (BIG 1-06), a randomized, multi-centre, open-label, neoadjuvant phase III trial designed to assess the efficacy of dual inhibition of HER2 [19]. In addition, we have investigated the influence of the antibodies, scoring methods, and cut-off criteria used, together with the impact of inter-observer variability on PTEN status determination.

methods

patient population and samples

Neo-ALTTO, a phase III parallel-group, open-label, randomized neoadjuvant study of trastuzumab, lapatinib, or their combination included patients with newly diagnosed HER2-positive invasive breast cancer amenable to surgery. Full eligibility criteria can be accessed elsewhere [19]. Patients received anti-HER2 therapy for 6 weeks, and paclitaxel was then added to the regimen for a further 12-week period until definitive surgery for a total period of 18 weeks of anti-HER2 therapy.

PTEN testing methods

FFPE baseline core biopsies were cut and stained with two different anti-PTEN monoclonal antibodies (clone 6H2.1 from DAKO and clone 138G6 from Cell Signaling Technology—CST). Two different pathologists independently scored the slides using the Hscore system. PTEN loss was defined as Hscore < 50 assessed in the invasive tumour cell component. Discordant cases were re-evaluated by the two observers and a unique reconciled score (RS) was generated and used for primary correlative analyses. Because there is no accepted standard definition for PTEN positivity or loss, we also examined an alternative cut-point of ≤10% positive staining (any level of cytoplasmic intensity, % score system) for PTEN loss (see supplementary Data for further details, available at Annals of Oncology online).

statistical analysis

Differences in clinicopathological features by PTEN expression status were calculated using the Wilcoxon two-sample test or χ² tests. The Neo-ALTTO primary end point was the rate of pathological complete response in the breast (bpCR). Secondary end points included rate of locoregional response (tpCR), EFS, and OS [19, 20]. The differences in tpCR rates by PTEN expression status were calculated and tested using a χ² test of association. Logistic regression models were fitted to tpCR, EFS, and OS and included HR status and treatment arm, as well as PTEN expression status or PI3K pathway activation (defined as either PTEN loss and/or PIK3CA mutation).

As a hypothesis-generating, exploratory analysis, the treatment effect was further examined by performing χ² tests of association on tpCR and PTEN expression status and PI3K pathway activation for each treatment arm separately. To address whether any of the continuous measures of PTEN may predict tpCR, each score was fitted in a logistic regression model with treatment arm and HR status. For inter-observer and inter-antibody agreements, intraclass correlation coefficients (ICCs) were calculated for pairs of continuous PTEN score and κ statistics for pairs of binary PTEN scores.

results

study patients

Out of 455 HER2-positive breast cancer patients enrolled in the Neo-ALTTO trial with available pCR data, 429 had FFPE baseline tumour samples available for IHC. Among these, 364 were evaluable for observer 1 (OBS1) and 360 and 366 were evaluable for observer 2 (OBS2, CST and DAKO, respectively). After reconciliation of discordant cases between the two observers, the total number of samples with available RS score for primary correlative analyses were 361 and 363 for CST and DAKO, respectively (supplementary Figure S1, available at Annals of Oncology online).

primary correlative analysis of PTEN status with clinicopathologic parameters, tpCR, and survival follow-up

Clinicopathologic characteristics of the patients whose tumours had loss or normal PTEN cytoplasmic staining in any invasive tumour cell are listed in Table 1.

Table 1.

PTEN expression according to baseline patient characteristics

	CST				DAKO
	PTEN normal	PTEN loss	% PTEN loss	P-value	PTEN normal	PTEN loss	% PTEN loss	P-value
All population	265	96	27		258	105	29
Age, years
Median	50	48.5		0.53	50	50		0.73
Range	57	50			57	50
Tumour size, cm
≤5	164	61	27	0.77	156	68	30	0.45
>5	101	35	26		102	37	27
Hormone receptor status
Pos	143	36	20	0.006	141	40	22	0.004
Neg	122	60	33		117	65	36
Histology tumour grade
Well/intermediate	110	34	24	0.40	99	42	30	0.55
Poor	114	49	30		118	51	30
Missing	41	13	24		41	12	23
PI3K gene status
WT	160	70	30	0.002	156	73	32	0.001
MUT	63	9	12.5		63	9	12.5
Missing	42	17	29		39	23	37

P-values of 0.05 or less are indicated in bold.

PTEN loss (Hscore < 50) was observed in 27% and 29% of patients for CST and DAKO, respectively. No differences in expression rates were observed among the three different treatment arms (supplementary Figure S2A and B, available at Annals of Oncology online). PTEN loss was more frequently found in HR-negative (33% and 36% with CST and DAKO, respectively) compared with HR-positive tumours (20% and 22% with CST and DAKO, respectively) (Table 1, supplementary Figure S2C and D, available at Annals of Oncology online).

We analysed the correlation between PTEN status and tpCR, as defined by no invasive cancer in the breast and no pathological involvement of axillary nodes. Expression of PTEN by IHC was not correlated with tpCR (Table 2, supplementary Figure S3, available at Annals of Oncology online). Analysis by treatment arm showed the largest difference in tpCR rates (not statistically significant) was observed in the lapatinib arm with a lower frequency of pCR in tumours with PTEN normal (18% and 17% for CST and DAKO, respectively) compared with those with PTEN loss (31% and 32% for CST and DAKO, respectively). PTEN was not a statistically significant predictor of tpCR in logistic regression models adjusted for treatment arm and HR status (supplementary Table S1, available at Annals of Oncology online). Similar results were observed when using bpCR as analysis end point (data not shown). No significant differences in EFS and OS according to PTEN or PI3K pathway activation status were observed in the Cox models adjusted for treatment arm and HR status (supplementary Table S2, available at Annals of Oncology online).

Table 2.

tpCR rates according to PTEN status in all treatment arms and for each individual treatment arm

Treatment arm	PTEN status	CST					DAKO
		tpCR					tpCR
		No		Yes			No		Yes
		Count	Row %	Count	Row %	P-value	Count	Row %	Count	Row %	P-value
All arms	Normal	173	68	82	32	0.89	169	68	78	32	0.47
	Loss	63	67	31	33		67	64	37	36
L	Normal	75	82	16	18	0.10	69	83	14	17	0.06
	Loss	22	69	10	31		25	68	12	32
T	Normal	55	69	25	31	0.85	56	69	25	31	0.58
	Loss	24	71	10	29		26	74	9	26
L+T	Normal	43	51	41	49	0.38	44	53	39	47	0.77
	Loss	17	61	11	39		16	50	16	50

L, lapatinib; T, trastuzumab; tpCR, total pathologic complete response.

correlation between PTEN protein loss and PIK3CA mutation status

PIK3CA mutation status was determined using Mass spectrometry on frozen biopsies from the same patients [7]. PTEN IHC results were available in 302 and 301 cases with PIK3CA mutation results for CST and DAKO, respectively (Table 1). PTEN loss was more frequently observed in tumours with PIK3CA wild-type status (30% and 32% for CST and DAKO, respectively) compared with those with PIK3CA mutations (12.5% for both antibodies; χ² P = 0.002 and P = 0.001 for CST and DAKO, respectively).

As PTEN status alone was not predictive of tpCR, we investigated if it could improve the predictive value of PIK3CA mutations previously shown in the same patient population [7]. PI3K pathway activation was found in 47% and 48% of patients (all arms) for CST and DAKO, respectively. tpCR rates were not significantly different for those with or without PI3K pathway activation (supplementary Tables S3 and S4, available at Annals of Oncology online). In logistic regression models adjusted for treatment arm and HR status, PI3K pathway activation was not predictive of tpCR (data not shown).

inter-observer and inter-antibody agreements

Strong inter-observer agreement was found independently of the antibody and scoring system used with disagreement rates ≤10%. ICCs ranged between 0.82 and 0.88 depending on the antibody and the scoring system used. Similar results were observed with κ-statistics (range: 0.70–0.75) (Figure 1A and B).

Figure 1.

Inter-observer and inter-antibody agreements. (A and C) Scattered plots showing Hscores comparison between observer 1 and observer 2 (A, left: CST; right: DAKO) and between CST and DAKO antibodies (C, left: OBS1; right: OBS2). (B) Inter-observer disagreement rates, ICC, and κ-statistic for CST and DAKO antibodies according to the scoring system used. (D) Inter-antibody disagreement rates, ICC, and κ-statistic for OBS1 and OBS2 antibodies according to the scoring system used. (E–G) Representative comparative images of one negative case (E: CST; F: DAKO) and one positive case (G: CST; H: DAKO). OBS 1, observer 1; OBS2, observer 2; HS, Hscore; %, % positivity score.

A strong correlation was found between CST and DAKO antibodies independently of the observer. The observed inter-antibody discordance rates were ≤10%, with percentages of disagreement ranging from 7.9% to 9.4% depending on the observer and scoring criteria used. ICCs were 0.83 for OBS1 and 0.87 and 0.83 for OBS2 using Hscore and % score systems, respectively. Similarly, κ-statistics ranged between 0.67 and 0.80 depending on the observer and scoring system (Figure 1C and D). Representative images of one positive and one negative case with each of the two antibodies are shown in Figure 1E–H.

impact of PTEN scoring variability

Next, we studied how frequencies of PTEN loss and correlations with clinicopathologic parameters such as HR status and pCR varied according to the different scoring method, cut-off threshold, antibody, and observer. Expression of PTEN was classified as ‘loss’ in 32%, 29%, 24%, 19% of cases by OBS1 and in 23%, 19%, 16%, and 13% of cases by OBS2 using DAKO Hscore < 50, CST Hscore < 50, DAKO ≤10%, and CST ≤10% criteria, respectively. Associations with HR status were statistically significant in four out eight events studied (three with OBS1 and one with OBS2). No association was found with pCR independently of the variable used (supplementary Table S5, available at Annals of Oncology online). To address whether any of the continuous measures of PTEN may predict pCR, each score was fitted in a logistic regression model with treatment arm and HR status. None of the PTEN scores were statistically significant (supplementary Table S6, available at Annals of Oncology online).

discussion

To the best of our knowledge, this report is the first study that investigated the impact of PTEN expression on the benefit of dual HER2 blockade with the combination of trastuzumab and lapatinib.

Due to the association between pCR and long-term outcome in HER2-positive breast cancer [21], neoadjuvant studies provide an optimal setting for identifying possible markers for therapy-resistant tumours. We used the tpCR definition instead of bpCR as locoregional response has been better associated with EFS and OS than response in breast alone [21]. Despite the lack of correlation found in our analysis, our data provide valuable insights into the biological and methodological issues around this controversial biomarker.

In Neo-ALTTO, PTEN loss rates varied between 27% and 29% of patients, depending on the antibody used. PTEN loss was more frequently found in HR-negative compared with HR-positive tumours, analogous to previous reports [13, 17, 22]. Despite preclinical evidence supporting PTEN loss as a mechanism of resistance to trastuzumab and lapatinib-based therapies [8, 11], our study demonstrated no significant differences in the pCR rates according to PTEN status in patients receiving neoadjuvant trastuzumab, lapatinib, or the combination of both agents. The higher pCR rate in HR-negative tumours recorded in the Neo-ALTTO study in all treatment arms [19]—similar to other neoadjuvant trials with anti-HER2 agents in HER2-positive disease [23]—may be a possible explanation for the lack of correlation between PTEN loss and pCR. Similarly, this may also explain the unexpected higher pCR in tumour with PTEN loss compared with PTEN normal observed in the lapatinib arm.

We have recently reported that the presence of PIK3CA mutations is associated with reduced clinical benefit from dual HER2 blockade in the Neo-ALTTO patient population [7]. Nevertheless, when we assessed PI3K pathway activation, defined as the presence of either PTEN loss or PIK3CA mutations, the predictive value of PIK3CA mutations not only did not improve, but worsened. Our results appear to conflict with a somewhat general consensus that PTEN loss correlates with resistance to anti-HER2 therapies. However, it is important to stress that previously available correlative data have been inconsistent.

In the neoadjuvant setting, previously published studies have reached different conclusions. Yonemori et al. [15] showed that PTEN expression was not associated with the clinical outcome of trastuzumab-containing neoadjuvant chemotherapy. In contrast, in the study by Dave et al., a trend towards a lower pCR response rate was observed in patients with low PTEN (15.4%) compared with those with high PTEN (44.4%) in the trastuzumab regimen. When the authors included PIK3CA gene status in the analysis, the association between low pCR and pathway activation was statistically significant (P = 0.015). The opposite effect was observed with the lapatinib regimen, with 92.3% of patients (12 of 13 patients) with low PTEN achieving pCR compared with 41.2% of patients (7 of 17 patients) with normal PTEN (P = 0.007) [12]. Similar to Dave et al., in a retrospective analysis conducted on 108 biopsies from the GeparQuattro study, low PTEN was significantly associated with lower pCR to trastuzumab-based chemotherapy [24]. In the adjuvant and metastatic settings, although more data are available, the correlation between PTEN status and clinical outcome remains unclear. Results from published studies on the role of PTEN in trastuzumab or lapatinib sensitivity in HER2-positive breast cancer are reviewed and summarized in supplementary Table S7, available at Annals of Oncology online.

The inconsistent literature correlative data of PTEN protein expression in HER2-positive breast cancer tumours might be attributable to different factors independent from the biology of the disease. First, there is a lack of standardization in PTEN detection methods. IHC provides a means for identifying loss of PTEN protein expression resulting from different mechanisms and, therefore, is currently the preferred technique to study PTEN expression in human tumours. The two most widely used anti-PTEN antibodies are the clone 138G6 from CST and the clone 6H2.1 from DAKO. A recent article comparing sensitivity and specificity of 11 commercially available antibodies showed that CST (138G6) exhibited the best correlation between staining and genetic loss of PTEN [25]. In our study, we observed high concordance (>90%) between the CST and DAKO antibodies, independent of the observer. In addition, high inter-observer agreement (>90%) for PTEN status determination were found for both antibodies, thus suggesting that they may provide overlapping results when conducted under standardized conditions in a ISO15189 accredited laboratory.

Second, as for the detection method, no standard definition of PTEN loss by IHC exists. Published studies used either simple (presence or absence of any staining) or composite scoring systems such as IRS and Hscore. In addition, the use of different cut-off thresholds, even when the same scoring system was applied, may explain the extremely variable frequencies of PTEN loss reported. In our study, we evaluated two different scoring methods. The Hscore encompasses the per cent of cells staining and the intensity and ranged from 0 to 300. The second method was purely binary, and split the expression into two categories based on a 10% cut-off. The lack of a standardized definition of PTEN cut-off thresholds for defining negativity and positivity generated loss call rates ranging from 13% to 32% within the same data set and had an impact on correlation between PTEN and HR. This is in line with the observation by Perez et al., where PTEN loss was associated with a lower or higher rate of HR depending on the cut-off threshold used [13]. Although the scoring methods and thresholds used did not have an effect on the correlation with pCR in our study, they may have had an impact when smaller sample sets were analysed.

Third, with the exception of results from the N9831 adjuvant trial, published evidence of the association—or lack thereof—between PTEN and anti-HER2 treatment sensitivity have been generated from small sample sets. Lastly, most of the patients in published studies received anti-HER2 agents combined with considerably heterogeneous chemotherapeutic regimens (concomitant versus sequential, taxanes versus other drugs, etc).

In conclusion, data from the Neo-ALTTO study indicate that PTEN protein expression alone (or in combination with PIK3CA mutation) is not significantly associated with differential benefit from trastuzumab, lapatinib, or dual HER2 blockade when combined to chemotherapy. These results complement recently published data from the GeparQuattro, GeparQuinto, and GeparSixto neoadjuvant studies [26] by suggesting that PIK3CA status determination, but not PTEN, may be an important biomarker to identify patients unlikely to derive substantial benefit from these treatments and, therefore, may be candidates for alternative therapies (e.g. PI3K inhibitors).

funding

This work was supported by GlaxoSmithKline (GSK) (grant number: do not apply).

disclosure

PGN received speaker honoraria from Novartis. CE is a GSK employee. MJP-G is a Board member of PharmaMar and received Research Grants and Consultant honoraria from Amgen, Astellas, AstraZeneca, Bayer, Eli Lilly, Invivis, MSD, Novartis, Pfizer, Roche-Genentech, Sanofi Aventis, Symphogen, Synthon, Verastem. MS received speaker honoraria from GlaxoSmithKline. JB acts as a consultant/advisor for Roche/Genentech. All remaining authors have declared no conflicts of interest.