1.1.

Clinical Background

NACT is a well-established treatment option for the clinical management of patients with locally advanced BC (more than three positive lymph nodes or tumors $>5\mathrm{cm}$), patients who have medical contraindications to surgery at diagnosis but in whom surgery is anticipated at a later date (i.e., women diagnosed during late pregnancy), select cases of early stage BC patients in which breast-conserving surgery would be suboptimal, and select cases of TNBCs or HER2+ patients. NACT, which is administered preoperatively, is often used to downsize larger tumors, allowing breast-conserving surgery rather than mastectomy.¹ Types of NACT include combinations of anthracyclines (e.g., doxorubicin or epirubicin) and taxanes (e.g., docetaxel or paclitaxel), with other drugs such as fluorouracil and cyclophosphamide. In HER2+ patients, targeted therapies with drugs such as trastuzumab, pertuzumab, lapatinib, and/or bevacizumab are administered along with standard chemotherapy regimens, individualized to the patient. NACT has been shown to be as effective as adjuvant chemotherapy in terms of outcomes and is now considered as a possible standard option and a preferred approach for a select subset of BC patients.¹⁶ Multiple studies have shown that a patient’s response to NACT is correlated with overall prognosis and disease-free survival.¹⁷ Specifically, patients who have a pathological complete response (pCR) after NACT exhibit better overall disease-free survival and better long-term outcome than those with residual disease (RD).^1,18–22 In particular, pCR is most strongly correlated with better overall disease-free survival and better long-term outcome in TNBC and HER2+ BC patients.^1,18,23,24 However, NACT has unpleasant and debilitating side effects, including, but not limited to, nausea, vomiting, fatigue, hair loss, infection, and mouth sores. Bone marrow reserves and renal function decrease with age, increasing the probability of myelosuppression, cardiodepression, peripheral neuropathy, and neurotoxicity.²⁵ Further, NACT has been shown to increase the density of tumor microenvironment of metastasis (TMEM) complexes in human breast tumors. TMEM complexes are thought to be sites of tumor cell intravasation and their density is correlated with metastatic outcome in human mammary carcinomas.²⁶ For these reasons, there is an identified pressing need to administer NACT only to those who will benefit from it and not administer it to those who will not benefit.¹⁹

Though there are different definitions of pCR, many studies concede that true pCR includes a lack of residual invasive disease in both the breast and axillary lymph nodes.^18,27–30 RCB, one method of quantifying RD in a resected specimen, is a score calculated based on the presence or absence of invasive tumor in the primary tumor bed, and presence or absence of tumor in the lymph nodes.^20,21,31,32 Patients are sorted into one of four classes based on their RCB scores, where RCB-0 corresponds to pCR, RCB-I and RCB-II represent intermediate outcomes, and RCB-III indicates extensive RD.³¹ In numerous clinical trials and meta-analyses, RCB class has a significant positive correlation with overall survival and discriminates between patients with favorable and unfavorable prognoses.²¹ Specifically, RCB-0 and RCB-I have been shown to have the same positive five-year prognosis no matter the NACT, adjuvant chemotherapy, or pathological stage, while RCB-II and RCB-III have worse prognoses.³¹ To this end, an improvement in the ability to identify those BC patients who are most likely to have a class of RCB-0 or RCB-I, and therefore a better response to NACT, will assist decision-making regarding surgery, adjuvant chemotherapy, and follow-up care.¹⁹

2.1.

Patient Samples

A total of 29 patients with HER2+ biopsy specimens and 27 patients with TNBC biopsy specimens, all with available HER2 and RCB results, were identified from the pathology files at the University of Rochester Medical Center (URMC) (2009 to 2019). Use of patient samples was approved by the Institutional Review Board at URMC. All TNBC patients received some type of NACT regimen which included standard combinations of a taxane, an anthracycline, cyclophosphamide, etoposide, 5-fluorouracil, methotrexate, and/or platinum therapy. All HER2+ patients received trastuzumab and/or pertuzumab with some type of NACT regimen which included standard combinations of a taxane, an anthracycline, cyclophosphamide, and/or platinum therapy. All biopsy samples collected prior to NACT administration were processed in the URMC pathology laboratory and mounted on slides as $5\text{-}\mu \mathrm{m}$-thick FFPE sections, prior to routine H&E staining. After NACT administration and tumor resection, all patients received either partial or total mastectomy with appropriate surgical evaluation of lymph nodes. Postmastectomy tissues were processed in the URMC pathology laboratory as $5\text{-}\mu \mathrm{m}$-thick FFPE sections and were H&E stained. All tumor H&E slides and immunohistochemistry stains were reviewed by at least two board-certified breast pathologists with manual interpretation of HER2 (rabbit antihuman HER2, Dako HercepTest^™). Fluorescence in-situ hybridization (FISH) was performed on all equivocal HER2 immunohistochemistry results (HER2 IQFISH pharmDx, FDA kit, Dako), and the FISH results were used in lieu of the immunohistochemistry for these cases.

The evaluation of the extent of RD following neoadjuvant therapy was performed on the post-treatment pathology material using the published method of Symmans et al.^21,31,32 and the associated RCB online calculator. Briefly, the gross description along with clinical imaging studies and specimen photographs were used to determine the two largest dimensions of the residual primary tumor bed in the resection specimen. Evaluation of microscopic sections was used to determine (1) the proportion of the primary tumor bed that contained metastatic carcinoma, (2) the number of axillary lymph nodes that contained metastatic carcinoma, and (3) the diameter of the largest metastatic deposit. This information was entered into the online calculator and the RCB score and RCB class were recorded.

2.2.

Imaging

A Spectra Physics MaiTai Ti:sapphire laser (circularly polarized at the sample using a Berek Compensator, 100 fs pulses at 80 MHz, 810 nm, $\sim 3\mathrm{mW}$ at the sample) was directed through an Olympus Fluoview FV300 scanner. The laser was focused through an Olympus UMPLFL20XW water-immersion lens ($20\times$, 0.95 NA), which subsequently captured backward-propagating SHG signal. This SHG signal was separated from the excitation beam using a 670-nm dichroic mirror, filtered (HQ405/30 m-2P, Chroma, Rockingham, Vermont), and collected by a photomultiplier tube (Hamamatsu HC125-02). The forward-scattered SHG was collected through an Olympus 0.9 NA condenser, reflected by a 565-nm dichroic mirror (565 DCSX, Chroma, Rockingham, Vermont) to remove excitation light, filtered (HQ405/30 m-2P, Chroma, Rockingham, Vermont) and captured by an identical photomultiplier tube (Hamamatsu HC125-02). Quantification of “F” and “B” signals while scanning the excitation wavelength from 740 to 880 nm (i.e., just below to just above the emission filter bandpass) enabled us to estimate the extent of contamination of the SHG signal with two-photon-excited fluorescence, revealing that $\sim 1\%$ of the detected forward-scattered photons and $\sim 10\%$ of backward-scattered photons were from two-photon-excited fluorescence.

A microscopist who was blinded to the patient outcome (i.e., eventual RCB class) performed the SHG imaging and analysis. Three fields of view (FOVs) were taken in the “tumor bulk” (solid box, Fig. 1) of each $5\text{-}\mu \mathrm{m}$-thick pretreatment core needle biopsy sample. The “tumor–host interface” was defined as the stroma immediately surrounding the tumor bulk. Three FOVs were also taken in the tumor–host interface, with the tumor bulk just out of the FOV (dashed box, Fig. 1). Forward- and backward-scattered SHG images were collected simultaneously. Examples of the H&E-stained FOV and corresponding SHG images are shown in Fig. 2. Images from these core needle biopsy sections revealed a random network of collagen fibers often encapsulating SHG-dark nodules containing highly cellular areas, consistent with our previous images of primary tumor sections.^15,33 Interestingly, close inspection of forward (F) and backward (B) images often revealed the characteristic “segmented” appearance of fibrils in the B image relative to the F image (Fig. 2), as previously reported in other tissue types.^4,8

Fig. 1

Core needle biopsy sections contain tumor cells and surrounding stroma. Pretreatment core needle biopsy sections contain heterogeneous cellular and noncellular regions. We have defined both the tumor bulk (solid box), which consists primarily of tumor cell clusters, and the tumor–host interface (dashed box), the matrix that surrounds the cell clusters and consists mainly of collagen and structural proteins. Three images were taken in the tumor bulk and three in the tumor–host interface for each H&E-stained core needle biopsy.

Fig. 2

Tumor bulk and tumor–host interface can be imaged with SHG microscopy. Core needle biopsy sections contain the bulk of the tumor and the tumor–host interface consisting mostly of collagen. Here, the same region of the tumor bulk is shown: (a) transmitted light microscopy (H&E stained) and (b) two- photon forward-scattered SHG image. In the same core, a FOV containing the tumor–host interface is shown: (c) transmitted light microscopy (H&E stained) and (d) two-photon forward-scattered SHG image of the same region. Note: (a)–(d) are $\sim 660\mu \mathrm{m}$ across. (e) Backward-scattered and (f) forward-scattered SHG images reveal the fibrillar collagen network, often encapsulating SHG-dark cellular areas. Characteristic segmented fibrils appear in (e) the backward-scattered image, relative to (f) the corresponding forward-scattered image. Note: (e) and (f) are $\sim 420\mu \mathrm{m}$ across.

2.3.

Image Analysis: Intensity Mask-Based F/B

A single value of F/B for a FOV was derived from the F and B images of that FOV using ImageJ^34,35 as previously described.^15,33 Briefly, an F image of forward-propagating SHG and a B image of backward-propagating SHG were collected in each FOV. All images were background-subtracted and a pair of masks was created to distinguish background pixels from collagen fibers. To produce the masks (one mask for the F image and one for the B image), a blinded user selected a threshold to best distinguish pixels within fibers from background pixels. The same two thresholds were used for all F and all B images in a single imaging session. All pixels above the lower threshold were set to 1 and those below the lower threshold to 0, producing binary F and B masks. These binary F and B masks were multiplied together to create a final mask of pixels within fibers. The background-subtracted F and B images were divided to produce a single F/B image, which was multiplied by this final mask. The average value of the nonzero pixels from the resultant image yielded the average F/B of the entire FOV. The F/B values from the three FOVs in each location type were averaged to provide a single F/B value from the tumor bulk and one from the tumor–host interface for each patient.

2.4.

Image Analysis: Coherency Mask-Based F/B

In addition to the selection of pixels based on an intensity threshold, we also investigated the selection of pixels based on their grouping into linear structures, as quantified by pixel “energy” and “coherency,” as previously described.³⁶ SHG images were background-subtracted and day-to-day variations in intensity were normalized, as described above. The open-source toolbox OrientationJ was used to select those pixels most relevant in this analysis. This toolbox enables the production of binary image masks where the pixels that are rejected or passed through the mask are selected based on their energy and/or coherency contribution. For this analysis, pixels passed by the mask were chosen to have a minimum of 2% of both normalized energy and coherency,³⁷ where the energy and coherency values are calculated from a $2\times 2$ structure tensor that is determined for each pixel. The structure tensor is defined as

Once the binary image masks were obtained for both the forward and backward images, they were multiplied together to produce a final mask. The background-subtracted forward and backward images were divided to produce a raw F/B image, which was multiplied by the final mask. All pixels of value zero were excluded, and the average of this image gave an average F/B for a particular FOV.

2.5.

Image Analysis: Collagen Fiber Angle at Tumor Boundary

In addition to measuring F/B in each tumor region, we assessed the organization of collagen fibers at the tumor–host interface. Using the method of Provenzano et al.,¹³ the boundary between the tumor bulk and the interface was defined and then the angle of collagen fibers relative to the tangent of the boundary was measured every $30\mu \mathrm{m}$. This analysis was performed in three imaged locations per biopsy section.

2.6.

Calibration

For each imaging session, a background image of no sample and a reference F and B image of a dilute stock solution of isotropically emitting fluorescein isothiocyanate (FITC) were collected (to quantify day-to-day variations in sample alignment). Day-to-day variations in system alignment were normalized by dividing each F/B value by the F/B value measured for the FITC stock solution (expected F/B = 1) during that imaging session.

2.7.

Statistical Analysis

Logistic regression was used to assess the association between F/B and binary response variable “RCB class $>1$.” (This is a variable that can be either 0 or 1. A value of “0” is assigned when RCB class is 0 or I, and a value of “1” if RCB class is II or III.) A likelihood ratio test was used to calculate the $p$-value for the test of whether the regression coefficient for F/B was zero. All graphs were generated using GraphPad Prism 8 and statistics were assessed using JMP Pro 14.

3.1.

HER2+ Samples: Intensity Masks

When F/B is calculated from SHG images using masks that select collagen pixels and reject background based on pixel intensity, the average F/B from the tumor bulk was significantly different from that of the tumor–host interface [$\mathrm{F}/\mathrm{B}=8.84\pm 0.67$ (tumor bulk) or $15.35\pm 1.01$ (tumor–host interface), $p<0.0001$, data not shown]. Intensity mask-based F/B in each region type was subsequently stratified by the patient’s RCB class post-NACT after mastectomy. Samples were grouped as RCB-0/I or RCB-II/III in both the tumor bulk ($\mathrm{F}/\mathrm{B}=8.87\pm 0.88$ or $8.78\pm 1.05$, respectively) and the tumor–host interface ($\mathrm{F}/\mathrm{B}=16.51\pm 0.95$ or $12.32\pm 1.84$, respectively). When coded as a binary response (0/I or II/III), intensity mask-based F/B derived from the tumor–host interface was associated with RCB class ($p=0.0246$; Fig. 3). However, intensity mask-based F/B was not associated with RCB class when derived from images of the tumor bulk ($p=0.297$; Fig. 3).

Fig. 3

Intensity threshold-based F/B is correlated with RCB class in HER2+ biopsy samples. Intensity threshold-based F/B was calculated from SHG images of the tumor bulk and tumor–host interface in each HER2+ core needle biopsy ($n=29$). Each point is the average of three tumor bulk or tumor–host interface FOVs. Two categories of NACT response were determined using RCB class, grouped as RCB-0/I and RCB-II/III. F/B is correlated with RCB class in the tumor–host interface, but not in the bulk of the tumor. Error bars = standard error of the mean (SEM), $p=0.0246$.

3.2.

HER2+ Samples: Coherency Masks

The F/B was then derived using masks that select collagen pixels and reject background pixels based on their residence in linear structures (i.e., coherency). In each region type (tumor bulk and tumor–host interface), the F/B was coded as a binary response (0/I or II/III) post-NACT after mastectomy. Samples were again grouped as RCB-0/I or II/III in the tumor bulk ($\mathrm{F}/\mathrm{B}=33.03\pm 2.73$ or $35.12\pm 3.76$, respectively) and tumor–host interface ($\mathrm{F}/\mathrm{B}=39.47\pm 2.25$ or $35.66\pm 3.85$, respectively). Coherency mask-based F/B was not associated with RCB class when measured either in the tumor–host interface or tumor bulk ($p=0.643$ and 0.346, respectively; Fig. 4).

Fig. 4

Coherency-based F/B is not correlated with RCB class in HER2+ biopsy samples. Coherency-based F/B was calculated from tumor bulk and tumor–host interface in each HER2+ core needle biopsy ($n=29$) using OrientationJ. Coherency-based F/B is not correlated with RCB class in either region of HER2+ biopsy samples. Error bars = SEM.

3.3.

TNBC Samples: Intensity Masks

In TNBC needle biopsy samples, the average intensity mask-based F/B derived from the tumor bulk was significantly different from that of the tumor–host interface [$\mathrm{F}/\mathrm{B}=7.48\pm 0.55$ (tumor bulk) or $10.67\pm 0.69$ (tumor–host interface), $p=0.0008$, data not shown]. The F/B in each region was coded as a binary response according to a patient’s RCB class post-NACT after mastectomy. Intensity mask-based F/B was not associated with RCB class 0/I or II/III in the tumor bulk ($6.62\pm 0.75$ or $7.72\pm 0.72$, respectively; $p=0.329$; Fig. 5). Intensity mask-based F/B was also not associated with RCB class 0/I or II/III in the tumor–host interface ($10.81\pm 3.89$ or $10.31\pm 3.56$, respectively; $p=0.73$; Fig. 5).

Fig. 5

Intensity threshold-based F/B is not correlated with RCB class in TNBC biopsy samples. Intensity threshold-based F/B was calculated from both tumor bulk and tumor–host interface in each TNBC core needle biopsy ($n=27$). F/B is not correlated with RCB class in either region of TNBC biopsy samples. Error bars = SEM.

3.4.

TNBC Samples: Coherency Masks

Coherency mask-based F/B in each region type was coded as a binary response by RCB class post-NACT after mastectomy. Samples were again grouped as RCB-0/I or II/III in the tumor bulk ($\mathrm{F}/\mathrm{B}=29.98\pm 4.20$ or $34.18\pm 2.98$, respectively) and tumor–host interface ($\mathrm{F}/\mathrm{B}=35.99\pm 3.81$ or $35.23\pm 2.57$, respectively). Coherency mask-based F/B was not associated with RCB class when derived from either the tumor–host interface or tumor bulk ($p=0.400$ and 0.862, respectively; Fig. 6).

Fig. 6

Coherency-based F/B is not correlated with RCB class in TNBC biopsy samples. Coherency-based F/B was calculated from tumor bulk and tumor–host interface in each TNBC core needle biopsy ($n=27$) using OrientationJ, as previously described. Coherency-based F/B is not correlated with RCB class in either region of TNBC biopsy samples. Error bars = SEM.

3.5.

HER2+ and TNBC Samples: Fiber Angle at Tumor Boundary

The average collagen fiber angle along the tumor boundary for each patient was coded as a binary response by RCB class post-NACT after mastectomy. Samples were again grouped as RCB-0/I or II/III in both HER2+ and TNBC patient groups. Average fiber angle was not associated with RCB class when measured in either HER2+ ($p=0.298$) or TNBC samples ($p=0.258$; Fig. 7).

Fig. 7

Average fiber angle at the tumor boundary is not correlated with RCB class in HER2+ or TNBC biopsy samples. The collagen fiber angle relative to the tumor boundary was measured every $30\mu \mathrm{m}$ in HER2+ ($n=28$) and TNBC core needle biopsies ($n=24$). The average fiber angle of each sample is not correlated with RCB class. Error bars = SEM.