Tumor Metabolism and Blood Flow as Assessed by Positron Emission Tomography Varies by Tumor Subtype in Locally Advanced Breast Cancer

Genomic profiling has identified different molecular subtypes of breast cancer, including basal-like, human epidermal growth factor receptor 2 (HER2) positive, luminal A, and luminal B (1–3). These molecular subtypes may be approximated by standard immunohistochemistry (IHC) as follows (2): basal-like (triple negative, TN) as estrogen receptor (ER) negative, progesterone receptor (PR) negative, and HER2 negative; HER2 like as ER negative, PR negative, and HER2 overexpressing; luminal A is ER and/or PR positive, HER2 negative; and luminal B is ER and/or PR positive, HER2 overexpressing. Differential response to neoadjuvant chemotherapy (NC) has been observed among breast cancer subtypes defined by either IHC (2) or by molecular profiling (4). Despite higher rates of pathologic complete response (pCR) to NC, basal and HER2 subtypes are more likely to recur, often rapidly, more often with brain and visceral metastases, and have poorer prognoses than do luminal subtypes (2, 3, 5–8). A greater understanding of the biological basis of these clinical observations may elucidate factors contributing to therapeutic responsiveness and resistance, and ultimately lead to more individualized and tumor-specific treatments.

Functional imaging of tumor blood flow (BF) and metabolism, using [¹⁵O]-water and [¹⁸F]-FDG positron emission tomography (PET), respectively, has been studied in breast and other tumor types (9–13, 35). Studies at this institution have shown that BF and metabolism in locally advanced breast cancer (LABC) are highly variable and provide a means of predicting response and disease relapse in patients receiving NC for LABC (9–11). Specifically, we found that LABC tumors with a pretherapy flow-metabolism mismatch [high metabolic rate of FDG (MRFDG) to BF] were more resistant to therapy, predicting a low likelihood of pCR and a high likelihood of early relapse (10, 14). Additionally, resistant tumors were more likely to have a higher perfusion rate over the course of therapy, failure to diminish tumor perfusion at midtherapy-predicted poor response, early relapse, and death (9, 11). In multivariable models, PET measures predicted relapse and survival independently of other established prognostic measures (15–17) including primary tumor pCR and posttherapy pathologic nodal status (14). Similar associations between flow-metabolism mismatch and more aggressive tumor biology have been observed in other tumor types using functional PET imaging (12, 13, 18).

To investigate patterns of BF and glucose metabolism for different breast cancer subtypes, we analyzed PET measures of tumor glucose metabolism and BF in a cohort of patients with LABC and approximated intrinsic subtype using standard IHC criteria. At least one other study has looked at FDG uptake by breast cancer subtype (19); however, to our knowledge, this is one of the first studies to use functional imaging to investigate the relationship between tumor BF and metabolism by breast cancer subtype. Functional imaging using dynamic PET offers a unique opportunity to study tumor physiology in vivo and may provide a portrait of tumor physiology that is complementary to differences in patterns of gene expression measured by genomic profiling. Noninvasive functional imaging by PET may identify patterns of metabolism and BF associated with responsive and resistant tumors that may ultimately allow for the selection of cytotoxic and biological therapies based on in vivo tumor biology.

Patients presenting to the University of Washington Breast Cancer Specialty Clinic with histologically confirmed breast carcinoma scheduled to undergo NC were eligible for the study. Patients were clinically staged according the American Joint Committee on Cancer tumor-node-metastasis classification of malignant tumors (20, 21). The majority of patients presented with clinical staging consistent with LABC, but a strict definition of inoperable LABC, was not required for inclusion in this analysis. Seven patients in this cohort presented with inflammatory breast cancer as determined by clinical staging on presentation. The enrollment period was from November 1995 through December 2005. Patients were excluded if they were pregnant, unwilling, or unable to undergo PET examinations. Patients were also excluded if they were not surgical candidates. A total of 71 patients were identified with baseline [¹⁵O]-water and [¹⁸F]-FDG PET imaging, demographic and tumor information, and pathologic response for inclusion in this study. Analyses done on a subset of this cohort have been previous published (9–11), but did not examine molecular subtype as currently defined. Written informed consent for PET studies and follow-up was obtained according to the University of Washington Human Subjects Committee guidelines.

According to institutional standards, tumor tissue was sampled under ultrasound or stereotactic guidance for diagnosis before initiation of NC. Tumor subtype by IHC was assessed on diagnostic biopsy material. ER expression was assessed using the 1D5 antibody (DAKO) and was expressed as negative (<5%), 1+ (>5-20% of tumor cells were ER positive), 2+ (20-80% positive), or 3+ (>80% positive). PR expression was assessed using the PR88 antibody (BioGenex) and the level of expression was scored in a fashion identical to that for ER. HER2 was defined as positive if IHC showed 3+. It was considered negative in those tumors scoring 1+. For those tumors with a 2+ score on IHC, gene amplification using fluorescence in situ hybridization was used to determine HER2 status. HER2 was considered positive if the ratio of HER2 gene copies to chromosome 17 signals was >2. Cellular proliferation was assessed by measurement of Ki-67 antigen by monoclonal antibody (DAKO) with scoring by institutional standards as low (<10%), intermediate (10-20%), and high (>20%). Tumor grade was based on expert breast pathology review.

TN, or basal, subtype was defined as ER and PR negative without HER2 overexpression by IHC or amplification by fluorescence in situ hybridization. Luminal tumors were defined as those expressing either ER or PR, without regard to HER2 status. For the purposes of this analysis, luminal A and B were not subdivided. HER2-like tumors were ER and PR negative with HER2 overexpression or gene amplification.

PET radiotracer production, imaging methods, and data analysis have been previously described (9–11, 22–24). Briefly, images were acquired on an Advance PET scanner (General Electric Medical Systems) and reconstructed by filtered back projection with an approximate resolution of 10 to 12 mm (22, 23). For [¹⁵O]-water studies, patients received 724 to 1,902 MBq in a 1- to 4-mL volume through bolus i.v. injection. Dynamic images were acquired for 7.75 minutes after injection. For [¹⁸F]-FDG studies, 218 to 396 MBq was infused over 2 minutes in a 7- to 10-mL volume, and dynamic images were acquired for 60 minutes after the start of infusion. Regions of interest measuring mean uptake were 1.5-cm diameter circles drawn on three adjacent slices over the left ventricle and the portion of the tumor with maximal uptake on summed [¹⁸F]-FDG images. Regions of interest from the summed [¹⁸F]-FDG images were copied onto the corresponding dynamic [¹⁸F]-FDG and [¹⁵O]-water images to determine BF and tumor time-activity curves. BF estimates from [¹⁵O]-water and [¹⁸F]-FDG kinetic parameter estimates (MRFDG and K₁) were obtained using a model optimization software (Berkley Madonna) as previously reported (24). The measurement error of these parameters has been previously estimated as 13% for BF, 11% for FDG K₁, and 3% for K_i, in which MRFDG equals K_i multiplied by the glucose concentration (24).

Descriptive displays and linear regression analysis were used to explore how functional parameters differed by subtype. Imaging parameters were log transformed to satisfy the normality assumptions for regression analysis. Overall statistical significance of subtype (three levels) as a predictor was assessed using likelihood ratio tests. Pairwise comparisons between subtypes used the Tukey-Kramer test to control for multiple comparisons. Multivariable models were fit to examine the relationship between subtype and imaging parameters, holding tumor grade constant. Relationships among subtype, tumor grade, and proliferation (Ki-67) were assessed using Fisher's exact test. Logistic regression models were fit to describe the associations between functional parameters and response to chemotherapy, for subgroups defined by subtype. Statistical analyses were conducted using the SAS/STAT software version 9.1 (SAS Institute, Inc.) and R version 2.8.1 (R Foundation for Statistical Computing).

Table 1 shows patient characteristics of the entire cohort of 71 patients with LABC. The average age at diagnosis was 47 years (range, 30-76). The majority of patients were diagnosed with infiltrating ductal carcinoma. Seven patients in this cohort presented with inflammatory breast cancer; five tumors were TN and two were HER2 like. The average tumor size was 4.9 cm (range, 1.1-14 cm). Most tumors did not overexpress the Her2/neu oncogene. By IHC, 25 of 71 (35%) patients had TN tumors, 6 (8%) had HER2 positive, and 40 (56%) were of the luminal subtype.

The majority of patients (n = 64, 90%) received institutional standard weekly metronomic doxorubicin with daily oral cyclophosphamide or other doxorubicin-based regimens. Doxorubicin and cyclophosphamide was followed by paclitaxel in 24 of 64 (38%) patients. Other chemotherapy regimens included cyclophosphamide, methotrexate, and 5-fluorouracil in four (6%); paclitaxel and trastuzumab in one; paclitaxel and vinorelbine in one; and docetaxel and vinorelbine in one. The mean chemotherapy duration was 18 weeks (range, 8-38 wk). Fifty-seven (80%) patients underwent mastectomy; 14 (20%) patients underwent lumpectomy of the primary breast tumor. Surgical management of the axillary nodes was per institutional standard with 68 (96%) of patients having an axillary node dissection.

Table 2 summarizes tumor characteristics by subtype. TN tumors were more likely to be grade 3 (odds ratio, 8.7; P < 0.001) and more likely to have a high Ki-67 index (odds ratio, 11.4; P = 0.009), compared with luminal tumors. Rates of pCR (28% for TN, 13% for luminal) were not significantly different (odds ratio, 2.7; P = 0.19) but were similar to pCR rates reported by others (2, 4, 8, 36).

Functional imaging parameters by subtype are described in Table 2. These data are displayed in full in Figs. 1 to 3, with boxplots overlaid to show the median and quartile values. Figure 1 shows MRFDG values before NC. MRFDG differed by subtype (P = 0.009), although the global effect of subtype was no longer statistically significant at the 0.05 level when controlling for tumor grade (P = 0.07). In both models (univariate and controlling for tumor grade), MRFDG for TN disease was on average 67% greater than for lesions with a luminal subtype [95% confidence interval (95% CI), 9-156%]. Mean FDG uptake by standardized uptake value normalized by body weight showed similar effect with standardized uptake value of TN tumors 54% greater on average than for the luminal subtype (95% CI, 10-116%; data not shown). Figure 2 shows the distribution of breast tumor BF. The subtypes did not differ in average BF levels (P = 0.73). Figure 3 shows that like MRFDG, the MRFDG/BF ratio differed between subtypes (P = 0.004). The difference persisted when controlling for tumor grade (P = 0.03). Pairwise comparisons found that the average MRFDG/BF ratio was 53% greater in TN compared with luminal cases (95% CI, 9-114%) and 88% greater in TN compared with HER2 (95% CI, 3-242%).

Phenotypic differences in the relationship between tumor metabolism and BF are suggested by Fig. 4, which plots MRFDG against BF and denotes patients who achieved pCR by darker plotting characters. Breast cancers of the luminal subtype with relatively high metabolism were observed to have correspondingly higher BF. The correlation of log-transformed MRFDG and BF was 0.66 (P < 0.001) in luminal tumors. Higher MRFDG was also associated with higher BF in TN tumors (r = 0.64, P < 0.001). However, although the strength of the relationship was similar (r = 0.65 for both luminal and TN), there was evidence for a difference in magnitude or slope (P = 0.07). When MRFDG was doubled, the predicted BF was 42% greater in luminal tumors, but only 30% greater in TN tumors (in a model constrained to have similar predicted BF for the lowest observed MRFDG for TN tumors, 2.73 μmol/min/100 g).

Figure 4 also suggests that the relationships between metabolism, BF, and pathologic response differ by subtype. The majority of luminal tumors had relatively low MRFDG and BF, and did not achieve pCR following NC. When MRFDG was higher, typically in the more proliferative luminal tumors, there was a matched elevation in BF and pCR occurred frequently. In the luminal subtype, there were greater odds of pCR for patients with high BF compared with patients whose tumors had lower BF (P = 0.009) and a trend for higher MRFDG to be associated with pCR (P = 0.055). For the TN subtype, pCR was not found to be associated with MRFDG (P = 0.38), BF (P = 0.75), or the MRFDG/BF ratio (P = 0.19), although five of seven pCRs occurred in TN tumors with lower MRFDG to BF ratios.

Using functional imaging of tumor metabolism and BF measured by [¹⁸F]-FDG and [¹⁵O]-water PET, this study shows that tumor BF and metabolism differ across breast cancer tumor subtypes. This is congruent with the molecular heterogeneity across tumor types identified by gene profiling (1, 3). MRFDG differed significantly between TN tumors and luminal tumors, with MRFDG being higher in the TN tumors. Similar results were found by Basu et al. (19) who reported enhanced FDG uptake measured by functional PET imaging in TN tumors. Our results add to this prior report by examining BF and the relationship between glucose metabolism and BF among different breast cancer subtypes.

The TN subtype tumors show greater variability in MRFDG than that of the luminal and HER2 subtype, suggesting heterogeneity within the TN subtype. There was no significant difference in BF across subtypes, but the relationship between metabolism and BF, expressed as the MRFDG/BF ratio, was significantly different in luminal versus TN tumors. For example, over the entire group of tumors, both MRFDG and BF varied considerably: MRFDG ranged from 0 to 40 μmol/min/100 g and BF ranged from 0 to 1 mL/min/g. Although over one third of the TN tumors had MRFDG greater than the midpoint of the range (20 μmol/min/100 g), only two of these nine tumors had BF over the range midpoint (0.5 mL/min/g). For luminal tumors on the other hand, only three tumors (8%) had a of MRFDG >20 μmol/min/100 g and all three had BF of >0.5 mL/min/g.

Several studies have shown that a BF metabolism mismatch, defined as high tumor glucose metabolism relative to BF, when compared with normal tissue, is associated with a more aggressive clinical behavior and therapeutic resistance (12, 18). Our data suggest that BF metabolism mismatch is more prevalent in the TN subtype when compared with luminal tumors. In the TN subtype, we observed greater heterogeneity in metabolism and BF, more frequent flow-metabolism mismatch, and no association between pretherapy imaging parameters and pCR. Prior work has shown that although the TN subtype is more likely to respond to systemic chemotherapy, patients with TN tumors that fail to respond to NC are more likely to relapse and die of their disease (2). Our findings in the TN subtype may shed light on some of the factors underlying this observation.

Metabolism and BF are both fundamentally important for normal cell survival and tissue viability. In normal tissue, autoregulatory mechanisms are responsible for the tight control of energy metabolism and BF (25), which tend to be highly correlated. This is not necessarily so with tumor cells. Both the vascular supply and energy metabolism is disorganized within tumor cells (26, 27), resulting in inefficient delivery and use of energy substrates and BF metabolism mismatch. Our data suggest that glucose metabolism and BF are relatively closely matched in the well-differentiated, luminal subtypes. As part of normal mammary function, differentiated breast epithelial cells proliferate and regress in response to the endocrine signaling that is part of female reproductive physiology (28, 29). One could hypothesize that the ability to match tissue perfusion to metabolic demand seen in most normal tissues (24, 25, 30) is preserved in more differentiated forms of breast cancer.

The TN subtype is thought to be molecularly similar to the basal, myoepithelial component of the normal breast that does not respond to endocrine signaling and has a more mesenchymal and less epithelial subtype (33). Our data suggest that tumors of this cell type often have glucose metabolism that is of proportion to tumor perfusion which, compared with luminal tumors, is the result of higher rates of glucose metabolism without comparable increases in tumor perfusion. This could be the result of molecular stimuli supporting enhanced glycolysis (27), an inadequate blood supply (26), or a combination of the two. It is interesting to note that the same pattern of flow-metabolism mismatch that is prevalent in locally advanced TN breast cancers is also seen in ischemically stressed myocardium (30) and could reflect the more mesenchymal origins of many TN breast cancers. This might be the result of differential gene expression for TN for luminal tumors including factors known to affect glucose metabolism such as GLUT-1 and hexokinase. The observation of more frequent flow-metabolism mismatch in TN tumors also suggests the potential for biological therapies targeting tumor vasculature and metabolism, and highlights the potential of functional imaging to elucidate mechanisms of response and resistance to an expanding arsenal of targeted therapies in development.

FDG uptake increases with inflammation (31, 32), raising concern that the high MRFDG and BF observed in some tumors may be due to an inflammatory response induced by diagnostic breast biopsy rather than tumor glycolytic acitivity. FDG and [¹⁵O]-water PET scans were done between 1 and 72 days after breast biopsy (median, 16 d). We found no evidence for an inflammatory response; the correlations between days since biopsy were 0.10 (P > 0.3) for both MRFDG and BF values, and the average MRFDG and BF for scans done within 7 days of biopsy were not different from the average for scans conducted >7 days after biopsy (t test P of >0.8 for both). We conclude that inflammation induced by the diagnostic breast biopsy has minimal influence on tumor glucose metabolism or BF as assessed by PET.

There are several limitations to our study. Ours is a retrospective analysis of prospectively acquired data with a relatively small number of patients. Determination of tumor subtype was based on IHC, which may not necessarily coincide with intrinsic subtype determined gene expression profiling (34). Future studies comparing tumor flow and metabolism measured by imaging to the intrinsic subtype assessed by gene expression arrays may shed more light on this observed phenomenon. Our sample sizes for the HER2 (n = 6) and luminal B (n = 10) subtypes were too small to attempt more than descriptive analysis. Over the course of this study, NC for LABC advanced with more frequent use of taxanes. Incorporation of HER2-targeted therapy, i.e., trastuzumab, was not routine before 2005 and the potential benefits of such biological therapy on pCR were not reflected in this analysis. Imaging parameters are macroscopic measures taken from the most metabolically active portion of the tumor and do not account for the heterogeneity that likely exists within these large tumors. Analysis using parametric imaging methods that can examine the heterogeneity of glucose metabolism and BF across the extent of the tumor may provide more insight.

In summary, we observed a difference in the rate of glucose metabolism and the relationship between glucose metabolism and tumor perfusion in luminal tumors versus TN tumors. Further examination of the relationship between tumor metabolism, perfusion, and gene expression may help explain these differences and may shed light on in vivo tumor biology and response to systemic therapy.

No potential conflicts of interest were disclosed.

Grant Support: CA72064 and CA42045 (D.A. Mankoff), and CA015704 (B.F. Kurland).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.