HIF-1α, TWIST-1 and ITGB-1, associated with Tumor Stiffness, as Novel Predictive Markers for the Pathological Response to Neoadjuvant Chemotherapy in Breast Cancer.

PURPOSE: To investigate the relationship between hypoxia-inducible factor 1-alpha (HIF-1α), Twist family BHLH transcription factor 1 (TWIST-1), and β1 integrin (ITGB-1) expression and tumor stiffness, and evaluate performance of HIF-1α, TWIST-1, and ITGB-1 alone and in combination with Ki-67 for predicting pathological responses to neoadjuvant chemotherapy (NACT) in breast cancer (BC). PATIENTS AND METHODS: This was a prospective cohort study of 104 BC patients receiving NACT. Tumor stiffness and oxygen score (OS) were evaluated before NACT by shear-wave elastography and optical imaging; HIF-1α, TWIST-1, ITGB-1, and Ki-67 expression were quantitatively assessed by immunohistochemistry of paraffin-embedded tumor samples obtained by core needle biopsy. Indexes were compared among different residual cancer burden (RCB) groups, and associations of HIF-1α, TWIST-1, ITGB-1, and Ki-67 with tumor stiffness and OS were examined. The value of HIF-1α, TWIST-1, ITGB-1, and Ki-67, and a possible new combined index (predRCB) for predicting NACT responses was assessed by receiver operating characteristic (ROC) curves.
RESULTS: HIF-1α, TWIST-1, and ITGB-1 expression were positively correlated with tumor stiffness and negatively with OS. Area under the ROC curves (AUCs) measuring the performance of HIF-1α, TWIST-1, ITGB-1, and Ki-67 for predicting responses to NACT were 0.81, 0.85, 0.79, and 0.80 for favorable responses, and 0.83, 0.86, 0.84, and 0.85 for resistant responses, respectively. PredRCB showed better prediction than the other individual indexes for favorable responses (AUC = 0.88) and resistant responses (AUC = 0.92).
CONCLUSION: HIF-1α, TWIST-1, ITGB-1, and Ki-67 performed well in predicting favorable responses and resistance to NACT, and predRCB improved the predictive power of the individual indexes. These results support individualized treatment of BC patients receiving NACT.

Introduction

Neoadjuvant chemotherapy (NACT) plays an indispensable role in the treatment of breast cancer (BC). Pathological complete response (PCR) is used as a surrogate prognostic marker for long-term disease-free survival after NACT in BC.1,2 Approximately 30% of cancers achieve PCR after NACT.3 However, certain risk factors are associated with the development of chemotherapy resistance.4 Moreover, because BC can progress during NACT, determining the optimal time for surgical intervention is difficult. Therefore, early prediction of the pathological response to NACT is critical, and it may help optimize individual chemotherapeutic strategies in BC patients.

Among the established clinicopathological markers, only Ki-67, estrogen receptor (ER), and human epidermal growth factor receptor 2 (HER2) have shown clear clinical applicability,4,5 and of these, only Ki-67 was reported to be an accurate biomarker for predicting PCR to NACT.6 However, the predictive efficacy of a single biomarker remains controversial. Gene expression analysis may improve our understanding of the biological behavior of BC, and although some genes may be useful indicators for the early identification of NACT responses,7,8 the high costs associated with their analysis limit their routine clinical application. Therefore, the identification of an effective biomarker to predict NACT responses and optimize the treatment of BC is an urgent need.

It is reported that BC with high matrix stiffness evaluated by pre-treatment ultrasound elastography is strongly correlated with chemoresistance.9,10 Tumor stiffness is largely determined by the collagen composition of the extracellular matrix (ECM), which has profound effects on BC progression, invasion, metastasis, and chemoresistance.11 Recent studies show that high matrix stiffness could induce epithelial to mesenchymal transition (EMT) and tumor progression by activating Twist family BHLH transcription factor 1 (TWIST-1).12,13 In addition, TWIST-1 over-expression is associated with short survival and a poor response to chemotherapy in patients with cancer.2,14,15

Hypoxia-inducible factor 1-alpha (HIF-1α), which mediates adaptation to hypoxia in cells, has been shown to be associated with matrix stiffness.16 Activation of the hypoxia pathway by HIF-1α contributes to the development of radiotherapy and chemotherapy resistance.17 HIF-1α can also directly upregulate TWIST-1 expression.18,19

Integrins play an important role in maintaining mammary stem cells in the normal breast. Dysregulation of integrin signaling distorts cell–cell or cell–ECM interactions and promotes BC progression by inducing chemoresistance and metastasis.20,21 A recent study showed that β1 integrin (ITGB-1) plays a pivotal role in the regulation of matrix stiffness,22 and ITGB-1 expression is associated with chemoresistance and metastasis in BC.20,21

Most of the studies cited above were based on basic experiments in vitro or in animal models. However, no study has investigated the relationships between HIF-1α, TWIST-1, and ITGB-1 expression and tumor stiffness in BC patients, or the predictive diagnostic performance of these biomarkers for predicting NACT responses. Here, we analyzed 104 patients who received NACT to determine the association of HIF-1α, TWIST-1, and ITGB-1 expression with tissue stiffness, oxygen score (OS), and pathological responses. In addition, we investigated the power of HIF-1α, TWIST-1, and ITGB-1 alone or in combination with Ki-67 to predict the response to NACT in BC.

Materials and Methods

Patients

A total of 112 women were enrolled between February 2014 and July 2019. All patients were diagnosed with invasive BC by ultrasound-guided core needle biopsy (CNB) and received NACT and subsequent surgical intervention. Eight patients were excluded because of changes in the treatment regimen or other unspecified reasons. The study was conducted with the approval of the ethics committee of Shengjing Hospital of China Medical University. All patients provided written informed consent.

Chemotherapy Regimen

Prior to surgery, all patients (n = 104) had received six cycles of NACT. The detailed chemotherapy regimens are as follows: 66 received TEC (docetaxel, epirubicin, and cyclophosphamide); 11 received TE (docetaxel and epirubicin); 9 received FEC (5-fluorouracil, epirubicin, and cyclophosphamide); and 18 classified as HER2+ received the targeted drug herceptin (trastuzumab) in addition to the docetaxel-based regimen.

Shear-Wave Elastography (SWE) Stiffness Evaluation

Tumor stiffness was evaluated using an ultrasound diagnostic imaging system, Aixplorer (SuperSonic Imagine, Aix en Provence, France), with a 4–15 MHz linear transducer. Four SWE images were obtained on two orthogonal planes for each lesion without outside pressure. Gray-scale and SWE images were simultaneously displayed in the split-screen mode. Tissue elasticity was assessed according to a color-coded map, with colors ranging from blue (soft) to red (hard). An optionally sized region of interest (ROI) trace (Q-box trace; SuperSonic Imagine) was drawn to include the lesion and peritumoral stroma. Then, quantitative elasticity values representing the Young modulus in kilopascals (range: 0–300 kpa) were automatically calculated and presented by the SWE system. Finally, the maximum elasticity (Emax) and mean elasticity (Emean) values of a lesion were recorded, and the average results of the four images were used for further analysis.

Breast Relative Oxygen Saturation Evaluation

The OS of the tumors was evaluated using the dynamic optical breast imaging (DOBI) system, TM-A02 (TRKM Medical Technology Co., Ltd, Shenzhen, China), which is equipped with a high-intensity probe (with a dual-wavelength LED illuminator at 730 and 850 nm) and a near-infrared camera (resolution: 570–600 lines; sensitivity: 0.001–0.01 Lux). The probe emits red light, which penetrates breast tissues and shows a different absorption or scatter pattern when meeting a neoangiogenic area compared with that in other tissues. This is attributed to differences in the distribution of oxygenated and deoxygenated hemoglobin.23

OS evaluation was performed in the dark. The patients exposed their upper body and sat facing the machine at a distance of 55–75 cm. First, the examiner palpated the five breast quadrants and the axillary area bilaterally. The probe was placed under the breast to be examined (outer or lower quadrant). The examiner adjusted the sharpness and brightness of the image to ensure that the entire breast was captured and that the vasculature in the breast tissue was clearly displayed. For each image, the probe was retained for at least 2 seconds to complete the acquisition. The system then generated a two-dimensional distribution image and a functional image. The relative oxygen content distribution of the area was represented by a color-coded map on the functional image (with colors ranging from green to red). Then, the ROIs were selected, and the system automatically calculated the OS of the lesion. OS represents the relative oxygen saturation of the ROI.23

Tumor SWE stiffness and the OS of all patients were evaluated one day before NACT.

Immunohistochemistry and Pathology

Pathologic assessments were conducted in two steps.

First, samples from ultrasound-guided CNBs were examined to confirm the histopathological characteristics and molecular subtypes of the tumors.

Immunohistochemistry

Anti-ER (Clone SP1, Roche, USA), anti-PR (Clone 1E2, Roche, USA), and anti-HER2 antibodies (Clone 4B5, Roche, USA), anti- HIF-1α (dilution 1:250; Clone EPR3658, Abcam, USA), anti-TWIST-1 (dilution 1:200; Clone 10E4E6, Abcam, USA), anti-ITGB-1 (dilution 1:200; Clone EPR1040Y, Abcam, USA), and anti-Ki-67 antibodies (dilution 1:200; Clone SP6, Abcam, USA) were used for immunohistochemical staining. Immunohistochemistry procedures were performed according to the manufacturers’ instructions.

Immunohistochemical Evaluation

Positive staining for ER and PR was defined as nuclear staining in ≥1% of the tumor cells. HER2 was assessed based on the intensity of tumor cell membrane staining; HER2-positivity was indicated by a 3+ or 2+ score and was confirmed by fluorescence in situ hybridization (FISH).24

HIF-1α, TWIST-1, ITGB-1, and Ki-67 expression levels were scored by two independent clinical doctors who had no prior knowledge of the prognosis or other clinicopathological variables, using a weighted Histoscore method, also known as the H-score.25 Briefly, the percentage of positive cells per slide (0% to 100%), as the average of ten random fields (400x, diameter: 0.55mm) screened, and the dominant intensity pattern of staining (0, absent; 1, weak; 2, moderate; 3, intense) were measured for each tumor section. H-scores for each sample were determined by multiplying the staining intensity by the percentage of positive cells (range, 0 to 300). Positivity for HIF-1α and TWIST-1 was defined as positive nuclear and cytoplasmic staining. Positivity for ITGB-1 was defined as positive membrane and cytoplasmic staining. Positivity for Ki-67 was defined as positive nuclear staining.

According to the St. Gallen International Expert Panel consensus,26 all lesions were classified into four major molecular subtypes: luminal A, luminal B, triple negative, and HER2-positive.

Second, all surgical specimens were collected to assess NACT responses. We opted for the web-based MD Anderson Residual Cancer Burden (RCB) calculator ().

This method allows calculation of an index that combines pathology measurements of the primary tumor (size and cellularity) and nodal metastases (number and size) as follows: RCB = 1.4 (finvdprim)0.17 + [4(1–0.75LN) dmet]0.17,27 where finv is the proportion of primary tumor area containing invasive carcinoma; dprim is the bidimensional diameters of the primary tumor bed; LN is the number of positive lymph nodes; and dmet is the diameter of the largest nodal metastasis. Using two cut-off points, four RCB categories were proposed as follows: RCB-0 (PCR, RCB = 0), RCB-I (RCB = 0−1.36), RCB-II (RCB = 1.36−3.28), and RCB-III (RCB > 3.28).27 Favorable responses to NACT were classified as RCB-0 or RCB-I, whereas RCB-III represented resistance to NACT (pathological non-responders). Pathologic assessments for all lesions were performed according to the World Health Organization classification standards.

Statistical Analysis

Statistical analyses were performed using SPSS 23.0 software (IBM, USA), GraphPad Prism version 5.0 (GraphPad Software, USA), and Sigmaplot version 14.0 (Systat Software, USA). Continuous variables with normal distribution were expressed as the mean ± standard deviation, skewed distributions were expressed as the median and interquartile range, and categorical variables were expressed as counts and percentages. Comparisons among three groups (PCR+RCB-I, RCB-II, and RCB-III) were performed using the χ2 test, Kruskal–Wallis test or analysis of variance (ANOVA). Post hoc analysis was used for pairwise comparisons among three groups if the results of the Kruskal–Wallis test or ANOVA test were significant. Bonferroni method analysis was also performed for pairwise comparisons among three groups if the results of the χ2 test were significant. Inter-observer reproducibility of Ki-67, HIF-1α, TWIST-1, and ITGB-1 was assessed by computing intra-class correlation coefficients (ICC). The detail parameters were as follows: Model: Two-Way Mixed-Effect model; Type: single measure; Definition: absolute agreement.28 Pearson’s correlation (rp) and Spearman’scorrelation (rs) analyses were used to analyze the relationships among SWE stiffness, OS, HIF-1α, TWIST-1, and ITGB-1 expression. The area under the ROC curve (AUC) values were used to determine the predictive diagnostic performance of HIF-1α, TWIST-1, ITGB-1, and Ki-67. A new predictive biomarker (predRCB) was combined with the largest AUC of new predictors (HIF-1α, TWIST-1, and ITGB-1) and the traditional one (Ki-67) according to the results of the multivariable linear regression model. Differences were considered significant when the two-sided P value was <0.05.

Results

Baseline Characteristics of Patients in the Three Groups

The baseline characteristics are summarized in Table 1. Among the 104 patients who underwent breast and axillary surgery 3–4 weeks after NACT, 23 (22%) showed a favorable response (PCR and RCB-I), 48 (46.2%) showed a moderate response (RCB-II), and 33 (31.7%) showed NACT resistance (RCB-III). In the subpopulations according to molecular subtype, the rate of PCR+RCB-I was 21.7% in the triple negative type, 26.1% in the HER2-positive type, 17.4% in the luminal A type, and 39.1% in the luminal B type. The RCB-III rate was 0% in the triple negative type, 4.2% in the HER2-positive type, 21.2% in the luminal A type, and 72.7% in the luminal B type. There were significant differences among the three RCB groups (P < 0.05) for most clinical indicators, except for HER2 positivity in “Immunohistochemical marker”, T2 in “Tumor size”, Grade 2 and Grade 3 in “Grade”, and IIIB in “Clinical stage”. The imaging indicators Emax, Emean, and OS were also significantly different among the three groups (P < 0.05). Figures 1 and 2 show the SWE and DOBI images of one lesion 1 day before NACT, respectively.

Table 1

Baseline Characteristics of Patients

Characteristic	Total	PCR+RCB-I	RCB-II	RCB-III	P value
RCB scores	2.1 (1.5–3.6)	0 (0–1.2)	2.1 (1.8–2.7)	4.0±0.5	<0.001#,a,b,c
Patients number	104	23	48	33
Age (years)	49.0 (38.0–57.0)	46.0 (43.0–53.0)	49.0 (32.0–57.0)	50.2±11.1	0.187#
Largest diameter (cm)	4.0 (2.3–5.0)	4.0(2.0–6.0)	3.0 (2.3–4.0)	4.0 (3.0–6.0)	0.192#
Immunohistochemical marker
Ki-67 (%)	32.5±18.9	50.0±21.3	33.7±14.9	18.6±9.4	<0.001￥,a,b,c
ER positive, n (%)	62	12 (52.2)	20 (41.7)	30 (90.1)	<0.001*,b,c
PR positive, n (%)	56	8 (34.8)	20 (41.7)	28 (84.8)	<0.001*,b,c
HER2 positive, n (%)	35	7 (30.4)	20 (41.7)	8 (24.2)	0.247*
Molecular subtype, n (%)
Luminal A	13	4 (17.4)	2 (4.2)	7 (21.2)	0.039*,c
Luminal B	51	9 (39.1)	18 (41.8)	24 (72.7)	0.004*,b,c
Triple negative	20	5 (21.7)	15 (31.2)	0 (0)	<0.001*,b,c
HER2 positive	20	6 (26.1)	13 (27.1)	2 (4.2)	0.032*
Pathological types, n (%)
Invasive ductal carcinoma	95	19 (82.6)	47 (97.9)	29 (87.9)	0.04*
Invasive lobular carcinoma	9	4 (17.4)	1 (2.1)	4 (12.1)	0.04*
Tumor size (cT), n (%)
T1	9	2 (8.7)	7 (14.6)	0 (0)	0.047*
T2	72	13 (56.5)	36 (75.0)	23 (69.7)	0.297*
T3	23	8 (34.7)	5 (10.4)	10 (30.3)	0.022*,a
Grade, n (%)
Grade 2	90	19 (82.6)	43 (89.6)	28 (84.8)	0.662*
Grade 3	14	4 (17.4)	5 (10.4)	5 (15.1)	0.662*
Clinical stage
IIA	12(11.5)	7 (30.4)	5 (10.1)	0 (0)	0.001*,b
IIB	40(38.5)	6 (26.1)	27(56.2)	7 (21.2)	0.002*,c
IIIA	45(43.3)	8 (34.8)	16 (33.3)	21 (63.6)	0.002*,c
IIIB	4(3.8)	2 (8.7)	0 (0)	2 (6.1)	0.098*
IIIC	3(2.9)	0 (0)	0 (0)	3 (9.1)	0.040*
SWE stiffness
Emax	147.9±65.5	114.7±34.9	141.2±56.9	180.7±78.9	0.001￥a,b
Emean	49.0±24.3	41.5±13.2	44.7±21.1	60.4±30.3	0.003￥b,c
OS	0.3 (0.2–0.5)	0.6±0.2	0.2 (0.2–0.4)	0.3±0.2	<0.001#,a,b

Notes: *χ2 test; #Kruskal–Wallis test; ￥ANOVA test; asignificant difference for pairwise comparison between PCR+RCBI and RCBII; bsignificant difference for pairwise comparison between PCR+RCBI and RCBIII; csignificant difference for pairwise comparison between RCBII and RCBIII.

Abbreviations: PCR, pathological complete response; RCB, residual cancer burden; ER, estrogen receptor; PR, progesterone receptor; HER2, human epidermal growth factor receptor 2; Emax, maximum elasticity; Emean, mean elasticity; OS, oxygen score; SWE, shear-wave elastography; n (%), number (%).

Figure 1

SWE image with a ROI placed over the stiffest part of the lesion with the Emax and Emean values being automatically calculated by the system.

Abbreviations: Emax, maximum elasticity; Emean, mean elasticity; ROI, region of interest; SWE, shear-wave elastography.

Figure 2

DOBI image with a ROI placed over the center point of the lesion with the OS values being automatically calculated by the system.

Abbreviations: DOBI, dynamic optical breast imaging; OS, oxygen score; ROI, region of interest.

HIF-1α, TWIST-1, ITGB-1, and Ki-67 Expression in the Three Groups

The expression of HIF-1α, TWIST-1, ITGB-1, and Ki-67 was detected in 104 preoperative tumor biopsy specimens (Figure 3A–H). Inter-observer reliability was good, with ICC values of 0.839 (0.771–0.888), 0.837 (0.769–0.886), 0.877 (0.823–0.915), and 0.804 (0.723–0.863) for Ki-67, HIF-1α, TWIST-1, and ITGB-1, respectively. The expression levels of HIF-1α, TWIST-1, ITGB-1, and Ki-67 in the different RCB groups are presented in Table 2. The expression of all biomarkers differed significantly among the three RCB groups (P < 0.01). Specifically, patients in the resistance group showed higher HIF-1α, TWIST-1, and ITGB-1 expression and lower Ki-67 expression, whereas those in the favorable response group showed lower HIF-1α, TWIST-1, and ITGB-1 expression and higher Ki-67 expression.

Figure 3

Left column (A–D) Immunohistochemical analysis of a pre-NACT biopsy specimen from a patient with chemotherapy resistance. (A) High HIF-1α expression with a H score of 180; (B) high TWIST-1 expression with a H score of 160; (C) high ITGB-1 expression with a H score of 190; (D) low Ki-67 expression with a H score of 30. Right column (E–H) Immunohistochemical analysis of a pre-NACT biopsy specimen from a patient with PCR to NACT. (E) Low HIF-1α expression with a H score of 40; (F) low TWIST-1 expression with a H score of 6; (G) low ITGB-1 expression with a H score of 2; (H) High Ki-67 expression with a H score of 270.

Abbreviations: HIF-1α, hypoxia-inducible factor 1-alpha; ITGB-1, β1 integrin; NACT, neoadjuvant chemotherapy; PCR, pathological complete response; TWIST-1, twist family BHLH transcription factor 1.

Table 2

HIF-1α, TWIST1, ITGB1, and Ki-67 Expression in the Three RCB Groups

Biomarkers (H Score)	Total	PCR+RCB-I	RCB-II	RCB-III	P value
HIF-1α	115.61±56.7	73.9±34.1	103.7±42.3	161.8±57.0	<0.001￥,a,b,c
TWIST-1	83.0 (58.5–149.0)	57.7±23.5	88.4±38.3	154.2±62.2	<0.001￥,a,b,c
ITGB-1	100.3±54.8	64.3±37.7	87.0±37.0	144.8±58.8	<0.001 ￥,a,b,c
Ki-67	67.6±46.1	112.7±62.7	67.4±29.8	36.6±19.5	<0.001￥,a,b,c

Notes: ￥ANOVA test; asignificant difference for pairwise comparison between PCR+RCBI and RCBII; bsignificant difference for pairwise comparison between PCR+RCBI and RCBIII; csignificant difference for pairwise comparison between RCBII and RCBIII.

Abbreviations: PCR, pathological complete response; RCB, residual cancer burden; HIF-1α, hypoxia-inducible factor 1-alpha; TWIST-1, twist family BHLH transcription factor 1; ITGB-1, β1 integrin.

Correlations Between SWE Stiffness, OS, HIF-1α, TWIST-1, ITGB-1, and Ki-67 Expression

Correlation analysis revealed negative correlations between Emax and OS (rs = –0.812, P < 0.001) and Emean and OS (rs = –0.715, P < 0.001); positive correlations were observed between HIF-1α and TWIST-1 expression (rs = 0.797, P < 0.001), between HIF-1α and ITGB-1 expression (rp = 0.852, P < 0.001), and between TWIST-1 and ITGB-1 expression (rs = 0.814, P < 0.001); negative correlations were observed between Ki-67 and HIF-1α expression (rp = –0.404, P < 0.001), between Ki-67 and TWIST-1 expression (rs = –0.467, P < 0.001), and between Ki-67 and ITGB-1 expression (rp = −0.358, P < 0.001). The correlations of immunohistochemical features with SWE stiffness and OS at baseline are shown in Table 3. The results showed that HIF-1α, TWIST-1, and ITGB-1 expression levels were positively correlated with SWE stiffness (Emean and Emax) and negatively correlated with OS. In addition, Ki-67 showed no or weak correlation with SWE stiffness and OS at baseline.

Table 3

Relationships Between HIF-1α, TWIST-1, ITGB-1, and Ki-67 Expression and SWE Stiffness and OS

Parameters	HIF-1α	TWIST-1	ITGB-1	Ki-67
Emax	0.708 (<0.001)*	0.621 (<0.001)	0.665 (<0.001)*	–0.244 (0.018)*
Emean	0.655 (<0.001)*	0.580 (<0.001)	0.637 (<0.001)*	–0.198 (0.044)*
OS	–0.644 (<0.001)	–0.622 (<0.001)	–0.618 (<0.001)	0.322 (0.001)

Notes: *Pearson’s correlation analysis; Spearman's correlation analysis for remaining data.

Abbreviations: HIF-1α, hypoxia-inducible factor 1-alpha; TWIST-1, twist family BHLH transcription factor 1; ITGB-1, β1 integrin; Emax, maximum elasticity; Emean, mean elasticity; OS, oxygen score.

Performance of HIF-1α, TWIST-1, ITGB-1, Ki-67, and PredRCB for Predicting the Response to NACT

The AUC values of HIF-1α, TWIST-1, ITGB-1, and Ki-67 for predicting NACT responses and the optimal threshold required are summarized in Table 4. HIF-1α, TWIST-1, ITGB-1, and Ki-67 showed good power for predicting a favorable response (AUC = 0.81, 0.85, 0.79, and 0.80, respectively) and resistance to NACT (AUC = 0.83, 0.86, 0.84, 0.85, respectively). Among the new predictors, TWIST-1 showed the highest predictive power with the largest AUC values for both a favorable response (AUC = 0.85) and resistance (AUC = 0.86).

Table 4

Predictive Diagnostic Performance of HIF-1α, TWIST-1, ITGB-1, Ki-67, and PredRCB for Predicting Responses to NACT

Predictor	Favourable Response							Resistant Response
	AUC	SE	95% CI	P value	Cut-off*	Sensitivity (%)	Specificity (%)	AUC	SE	95% CI	P value	Cut-off*	Sensitivity (%)	Specificity (%)
New
HIF-1α	0.81	0.049	0.71–0.90	<0.001	≤71.0	84.0	65.2	0.83	0.042	0.75–0.91	<0.001	>107.0	87.9	73.8
TWIST-1	0.85	0.042	0.77–0.93	<0.001	≤66.0	77.8	78.3	0.86	0.037	0.79–0.94	<0.001	>112.0	81.8	81.7
ITGB-1	0.79	0.057	0.67–0.90	<0.001	≤57.0	86.4	65.2	0.84	0.042	0.75–0.92	<0.001	>93.0	84.8	71.8
Traditional
Ki-67	0.80	0.049	0.71–0.90	<0.001	≥49.0	95.7	51.9	0.85	0.039	0.77–0.92	<0.001	<49.0	77.5	81.8
Combined
PredRCB	0.88	0.034	0.82–0.95	<0.001	≤2.075	72.8	95.7	0.92	0.027	0.86–0.97	<0.001	>2.717	81.8	87.3

Note: *Cut-offs were generated by the best Youden index.

Abbreviations: HIF-1α, hypoxia-inducible factor 1-alpha; TWIST-1, twist family BHLH transcription factor 1; ITGB-1, β1 integrin; AUC, area under the receiver operating characteristic curve; SE, standard error; CI, confidence interval.

Furthermore, Ki-67 (traditional predictor) and TWIST-1 (the best new predictor) were combined into a new predictor (predRCB), which was generated using a multivariable linear regression model (predRCB = 1.819 + 0.012 × TWIST-1 – 0.01 × Ki-67). Compared with other single predictors, predRCB showed a better performance for predicting both a favorable response (AUC = 0.88) and resistance (AUC = 0.92) (Figure 4).

Figure 4

ROC curves of HIF-1α, TWIST-1, ITGB-1, Ki-67 index and predRCB for predicting (A) a favorable response to NACT and (B) resistance to NACT.

Abbreviations: HIF-1α, hypoxia-inducible factor 1-alpha; ITGB-1, β1 integrin; TWIST-1, twist family BHLH transcription factor 1; RCB, residual cancer burden; ROC, receiver operating characteristic.

In addition, the results of AUC analysis of the predictive performance of predRCB among the different subgroups are shown in Table 5. PredRCB showed good accuracy (AUC > 0.80) in predicting NACT responses in the different subgroups, especially in patients with luminal A type, invasive lobular carcinoma, Grade 3, T3 tumor size, and TEC in NACT regimens.

Table 5

Performance of predRCB for Predicting Responses to NACT in the Different Subgroups

Subtype	Favourable Response				Resistant Response
	AUC	SE	95% CI	P value	AUC	SE	95% CI	P value
Molecular subtype
Luminal A	1.00	0.000	1.00–1.00	0.005	0.93	0.07	0.79–1.00	0.01
Luminal B	0.90	0.044	0.81–0.98	<0.001	0.91	0.04	0.83–0.99	<0.001
Triple negative	0.89	0.101	0.69–1.00	0.010	–	–	–	–
HER2	0.95	0.05	0.85–1.00	0.003	0.78	0.17	0.44–1.00	0.20
Pathological types
Invasive ductal carcinoma	0.86	0.04	0.78–0.94	<0.001	0.92	0.03	0.86–0.97	<0.001
Invasive lobular carcinoma	1.00	0.00	1.00–1.00	0.014	0.95	0.07	0.81–1.00	0.03
Grade
Grade 2	0.88	0.04	0.81–0.95	<0.001	0.90	0.03	0.84–0.97	<0.001
Grade 3	1.00	0.00	1.00–1.00	0.005	1.00	0.00	1.00–1.00	0.003
Tumor size (cT)
T1	0.57	0.20	0.19–0.96	0.770	–	–	–	–
T2	0.84	0.05	0.75–0.94	<0.001	0.88	0.04	0.80–0.96	<0.001
T3	1.00	0.00	1.00–1.00	<0.001	0.96	0.03	0.90–1.00	<0.001
NACT regimens:
TEC	0.93	0.03	0.86–1.00	<0.001	0.92	0.03	0.85–0.98	<0.001
TE	0.64	0.18	0.30–0.99	0.450	1.00	0.00	1.00–1.00	0.114
FEC	–	–	–	–	0.79	0.15	0.48–1.00	0.242
TH	0.75	0.11	0.54–0.96	0.261	1.00	0.00	1.00–1.00	0.008

Note: “–“: Unable to calculate due to small sample size after subgrouping.

Abbreviations: NACT, neoadjuvant chemotherapy; TEC, (docetaxel, epirubicin, and cyclophosphamide); TE (docetaxel and epirubicin); FEC (5-fluorouracil, epirubicin, and cyclophosphamide); TH (docetaxel and herceptin); AUC, area under the receiver operating characteristic curve; SE, standard error. CI, confidence interval.

Discussion

The results of this study can be summarized as follows: (i) Higher tumor stiffness was strongly correlated with higher HIF-1α, TWIST-1, and ITGB-1 expression and lower OS. (ii) HIF-1α, TWIST-1, ITGB-1, and Ki-67 expression exhibited good diagnostic performances for predicting a favorable response and resistance to NACT. (iii) PredRCB had considerably better power than HIF-1α, TWIST-1, ITGB-1, and Ki-67 alone for predicting NACT responses.

BC with higher stiffness evaluated by ultrasound elastography is closely correlated with chemoresistance.9,29,30 Indeed, BC progression, invasion, and resistance to chemotherapeutic drugs are not only determined by the tumor cells themselves, but also by the extracellular microenvironment.31 Basic research studies have confirmed that the cross-linking collagen in the ECM plays an important role in tissue stiffness.32 However, the reason why tumors with high stiffness tend to be resistant to NACT in BC remains unknown.

The two particularly important molecules of the ECM, collagen and hyaluronan (HA), participate in matrix stiffness, especially in the targeting of mechanotransduction in cancer.30,33 Specifically, abnormal collagen composition or increased HA concentration can increase colloidal osmotic pressure, which can increase interstitial fluid pressure (IFP). High IFP can cause collapse of tumor vessels, thereby reducing microvascular perfusion and eventually limiting the delivery of chemotherapeutic drugs.34,35

Hypoxia is another extracellular microenvironment factor that plays an important role in regulating BC progression, invasion, metastasis, and chemotherapeutic resistance.36,37 Zhu et al37 analyzed the concentrations of total hemoglobin (t-Hb), oxygenated (oxy-Hb), and deoxygenated hemoglobin (deoxy-Hb) before chemotherapy by ultrasound-guided near-infrared optical tomography, and determined their association with pathologic responses to chemotherapy in 34 BC cases. The results indicated that hypoxia in BC tissues was strongly correlated with chemoresistance. In this study, we confirmed that invasive BC with lower stiffness and higher OS showed better NACT responses (PCR or RCBI), whereas tumors with higher stiffness and lower OS were associated with NACT resistance (RCBIII). These results are consistent with those of previous studies.9,29,30,37

This study is the first to use non-invasive imaging modalities to confirm the significant negative correlation between tumor stiffness and OS. The results can be explained as follows. (i) Matrix stiffness and neovascularization increase simultaneously during BC progression. However, the growth of vasculature in the tumor region does not meet the increased need for oxygen and nutrients in the neoplasm. An imbalance between blood oxygen demand and supply leads to hypoxia in tumor tissues. (ii) Increased matrix stiffness increases IFP, which causes tumor vascular collapse and compression. This further increases blood flow resistance, which reduces tumor microvascular blood perfusion, consumption of oxygen, and accumulation of waste, eventually leading to the formation of a hypoxic environment.

We hypothesized that increased matrix stiffness causes insufficient blood perfusion, which not only reduces drug delivery, but also induces hypoxia. The response to hypoxia is mainly regulated by HIF-1α in tumors, and HIF-1α is involved in tumorigenesis, growth, and metastasis in BC.38,39 Furthermore, HIF-1α directly binds to and upregulates the expression of TWIST-1.18,19,40 In addition, increased matrix stiffness activates and upregulates the expression of TWIST-1 in BC.12 However, the mechanism by which TWIST-1 promotes BC resistance to NACT still needs to be elucidated. Recently, Yang et al41 revealed that ECM–receptor interaction and the MAPK, PI3K/AKT, P53, and WNT signaling pathways are aberrantly activated in MCF10A-TWIST-1 cells based on iTRAQ-labeling combined with 2D LC-MS/MS analysis. These authors used ingenuity pathway analysis to show that TWIST-1 regulates these downstream proteins through ITGB-1. Thus, TWIST-1/ITGB-1 seems to be the upstream signaling molecules that induce invasion and metastasis in BC. In addition, over-expression of both TWIST-1 and ITGB-1 is associated with increased tumorigenesis, growth, metastasis, and resistance to chemotherapy in several cancers.42,43

Based on the analysis above, we propose a hypothesis to explain why high tumor stiffness is closely correlated with chemoresistance in BC. The IFP increases in tumor environments with high matrix stiffness, leading to heterogeneity and tortuosity of neovascularization. This leads to increased blood flow resistance, which reduces blood perfusion and increases oxygen consumption. Finally, a hypoxic environment is formed in BC cells. The HIF-1α/TWIST-1/ITGB-1 pathway is activated, which causes BC phenotypic changes (eg, increased efflux of chemotherapeutic drugs, increased cytoprotective autophagy, and reduced apoptosis) through different downstream signaling pathways. These factors lead to the development of resistance to chemotherapeutic drugs. This study is the first to show that the expression levels of HIF-1α, TWIST-1, and ITGB-1 are positively correlated with stiffness and negatively correlated with OS in BC patients. In addition, we found a direct association among the expression levels of HIF-1α, TWIST-1, and ITGB-1. Moreover, higher expression levels of HIF-1α, TWIST-1, and ITGB-1 were inter-related in the NACT resistance group (RCBIII). These results strongly support the hypothesis above. However, the mechanism underlying the function of the HIF-1α/TWIST-1/ITGB-1 axis and downstream signaling needs to be investigated in further prospective studies in vitro and in vivo.

We also used these biomarkers to predict the pathological response to NACT in BC. The results demonstrated that HIF-1α, TWIST-1, and ITGB-1 show comparable abilities for the accurate assessment and prediction of NACT responses. To the best of our knowledge, this is the first study to report the performance of HIF-1α, TWIST-1, and ITGB-1 for predicting the response to NACT in BC. Ki-67 is a biological marker associated with cell proliferation and is regarded as the traditional marker for predicting the response to NACT in BC.6 In this study, Ki-67 also showed a good predictive performance. These results strongly support previous findings.6,44

This study also showed that a combination of pathological markers can improve the power of single markers for predicting the response to NACT in BC. TWIST-1 and Ki-67, which showed the best performance for predicting different responses, were combined into a new predictive marker termed predRCB using a linear regression model. The results showed that the performance of predRCB was better than that of the other biomarkers alone for predicting NACT responses. In the subgroup analysis, we confirmed that the predictive power of predRCB was not influenced by tumor classification, subtype, or NACT regimens. This provided strong evidence that our conclusions had generalizability and validity.

Clinically, identifying a method to reverse or block chemoresistance in BC with high tissue stiffness is important. In in vitro cultures mimicking stiffness changes during BC progression, alterations in ECM rigidity might aberrantly activate certain mechanotransduction pathways, resulting in various tumorigenic processes, such as sustained proliferation, EMT, invasion, metastasis, and resistance to cell death.31,45-47 However, alterations in ECM rigidity cannot be induced in the human body because cells normally exist in a physiologic environment with specific rigidity, pressure, and strain. Several studies have investigated whether markers such as HIF-1α, TWIST-1, and ITGB-1 can be used as therapeutic targets to reverse or block resistance to chemotherapy in BC. Some of the results are as follows. (i) HIF-1α: HIF-1α inhibitors such as digoxin and acriflavine show convincing potential therapeutic effects by decreasing tumor growth, vascularization, invasion, and metastasis, as well as chemoresistance in animal models of BC.48,49 (ii) TWIST-1: TWIST-1 is an excellent target for modulating chemoresistance in BC because it is rarely expressed in normal human tissues.50 Therefore, systemic use of TWIST-1 inhibitors could have a significant effect on TWIST-1-overexpressing cancer cells with minimal side effects in other tissues. The inactivation of TWIST-1 by siRNA technology or chemotherapeutic approaches was shown to be successful,51–53 and several inhibitors have been identified to antagonize the upstream or downstream molecules of TWIST signaling.54 Recently, cancer stem cell-targeted nanoparticle delivery has received increased attention. Finlay et al55 verified the in vivo efficacy of mesoporous silica nanoparticle (MSN)-delivered siRNA in a mouse model of melanoma. Similar data were reported in ovarian cancer, where the tumor burden was significantly reduced by 75% in mice treated with siTWIST MSN plus chemotherapy compared with the control chemotherapy-only treated mice.56 (iii) ITGB-1: Increasing evidence suggests that ITGB-1 is a potential molecular therapeutic target in BC.57,58 The therapeutic potential of targeting ITGB-1 using various strategies or inhibitors, such as monoclonal antibodies, peptides, or synthetic peptides, and siRNA is being investigated. Inhibitors that silence ITGB-1 were shown to suppress tumor progression and metastasis in vitro and in vivo.

However, most of these studies are based on basic experiments. Further clinical trials are warranted to determine whether HIF-1α/TWIST-1/ITGB-1 targeting strategies can reverse or block the chemotherapeutic resistance of BC alone or in combination with current therapeutic regimens.

The present study had several limitations. First, the study was based on a relatively small sample size. Second, the endogenous expression of HIF-1α, TWIST-1, ITGB-1, and Ki-67 in the tumor (as a three-dimensional structure) showed an uneven distribution. The expression of HIF-1α, TWIST-1, ITGB-1, and Ki-67 was assessed using tissues obtained by CNB. However, local samples obtained by CNB do not fully reflect the heterogeneity of the entire tumor. Taking the average of multi-site biopsies might reduce this difference. Finally, disease-free survival and overall survival analyses could not be performed because of the short follow-up period. A prospective study with a longer follow-up period should be conducted to confirm the present results.

Conclusion

First, higher HIF-1α, TWIST-1, and ITGB-1 expression levels were strongly correlated with higher tumor stiffness and lower OS in BC patients. Second, HIF-1α, TWIST-1, and ITGB-1 expression exhibited a good diagnostic performance for the early prediction of NACT responses in BC. Third, this study highlighted the potential utility of predRCB, which could improve the diagnostic performance of single markers for the early prediction of different pathological responses and assist in clinical treatment decisions in patients with BC receiving NACT.