Evaluation of tumor-infiltrating lymphocytes and mammographic density as predictors of response to neoadjuvant systemic therapy in breast cancer

Abstract

Background

Response rates vary among breast cancer patients treated with neoadjuvant systemic therapy (NAST). Thus, there is a need for reliable treatment predictors. Evidence suggests tumor-infiltrating lymphocytes (TILs) predict NAST response. Still, TILs are seldom used clinically as a treatment determinant. Mammographic density (MD) is another potential marker for NAST benefit and its relationship with TILs is unknown. Our aims were to investigate TILs and MD as predictors of NAST response and to study the unexplored relationship between TILs and MD.

Material and methods

We studied 315 invasive breast carcinomas treated with NAST between 2013 and 2020. Clinicopathological data were retrieved from medical records. The endpoint was defined as pathological complete response (pCR) in the breast. TILs were evaluated in pre-treatment core biopsies and categorized as high (≥10%) or low (<10%). MD was scored (a–d) according to the breast imaging reporting and data system (BI-RADS) fifth edition. Binary logistic regression and Spearman’s test of correlation were performed using SPSS.

Results

Out of 315 carcinomas, 136 achieved pCR. 94 carcinomas had high TILs and 215 had low TILs. Six carcinomas had no available TIL data. The number of carcinomas in each BI-RADS category were 37, 122, 112, and 44 for a, b, c, and d, respectively. High TILs were independently associated with pCR (OR: 2.95; 95% CI: 1.59–5.46) compared to low TILs. In the univariable analysis, MD (BI-RADS d vs. a) showed a tendency of higher likelihood for pCR (OR: 2.43; 95% CI: 0.99–5.98). However, the association was non-significant, which is consistent with the result of the multivariable analysis (OR: 2.51; 95% CI: 0.78–8.04). We found no correlation between TILs and MD (0.02; p = .80).

Conclusion

TILs significantly predicted NAST response. We could not define MD as a significant predictor of NAST response. These findings should be further replicated.

Keywords:

• Breast cancer
• tumor-infiltrating lymphocytes
• mammographic density
• neoadjuvant therapy
• pathological complete response

Background

Modern treatment regimens for breast cancer are increasingly individualized based on patient and tumor characteristics. Today, neoadjuvant systemic therapy (NAST) followed by surgical resection is the first line of treatment for locally advanced breast cancer, and also human epidermal growth factor receptor 2 (HER2)-positive and triple negative breast cancers that are larger than 20 mm, and/or with node positive disease [1]. The long-term outcome of NAST is comparable with that of postoperative chemotherapy [2,3]. Furthermore, neoadjuvant regimens have advantages such as tumor-downsizing, thereby enabling breast-conserving surgery and improving operability for locally advanced tumors. NAST also allows for continuous evaluation of tumor response, thus ensuring tumor chemosensitivity [4,5]. Unfortunately, NAST response rates vary significantly between and within the different subtypes of breast cancer [6]. Therefore, there is an interest in new markers that can assist clinicians in identifying patients who will benefit most from NAST.

Stromal tumor-infiltrating lymphocytes (TILs) are an increasingly recognized biomarker for NAST response [7]. TILs are measured as the percentage of tumor stroma area infiltrated by mononuclear cells and are speculated to serve as a proxy marker for the tumor-associated immune response [7,8]. The prognostic and predictive roles of TILs are most pronounced in HER2-positive and triple negative breast cancer subtypes [8–10]. The biomarker has not yet been routinely implemented in clinical decision-making.

Mammographic density (MD) provides a radiological depiction of breast density and reflects the relative amounts of radiodense fibroglandular tissue and radiolucent fat tissue in the breast [11]. Women with high MD have a 4- to 6-fold increased risk of breast cancer [12]. The proposed biological explanations for this association include the large cell count (increased chance of acquiring mutations) and the stromal microenvironment [12,13]. Furthermore, high MD is responsible for the so-called masking effect (high radiodensity reduce the sensitivity of radiologic surveillance) [11]. MD has been investigated as a predictive marker for NAST response, but due to conflicting findings, the clinical importance remains unclear [14–20].

Previous research has indicated an association between high MD and an elevated immune cell presence in breast tissue samples obtained from healthy women [21,22]. Additionally, researchers have observed significant differences in radiological characteristics of breast tissue and tumor appearance when comparing groups with high and low levels of TILs, particularly in the context of mammographic radiomics [23,24] and ultrasound features [25]. However, the specific relationship between TILs and MD remains unexplored. The link between TILs and radiological features is of great interest, as it may unveil alternative non-invasive predictive markers. Furthermore, investigating the relationship between pre-NAST TILs and MD could improve understanding of the biological mechanisms playing into their respective predictive roles.

The aim of this study was to investigate whether pre-NAST stromal TILs and MD hold predictive value with regard to NAST efficacy in a population-based cohort. The secondary aim was to investigate the relationship between TILs and MD.

Material and methods

Study population

Using a database search at the Sahlgrenska University Hospital, the hospital that performs all breast cancer care in the Region Västra Götaland catchment area, we identified all breast carcinomas treated with NAST between 2013 and 2020 (N = 330). Women with available pre-NAST mammographic images, and pre-NAST core biopsies were eligible for this study. The flowchart in Figure 1 shows exclusion criteria. In total, 315 carcinomas were included in the final analysis. In patients with multiple, unilateral tumors, only the tumor characteristics of the largest tumor were included in the study (N = 38), i.e., tumor size, tumor grade, TIL level, hormone receptor status, HER2 status, and Ki-67. If a patient had bilateral tumors of different approximate subtypes, both tumors were included in the analysis as two different cases (N = 1). Exclusions were kept to a minimum given the limited number of cases and multiple objectives of this study. This study was performed per the Helsinki Declaration of 1964 and its later amendments and gained approval from the Regional Ethics Committee in Gothenburg Sweden in 2018 (reference number: 479-18). An extension request was approved in 2020 (reference number: 2020-04945). Informed consent was waived due to the retrospective nature of this study.

Figure 1. Flow chart of study participants.

Abbreviations: N: number of patients; TILs: tumor-infiltrating lymphocytes; MD: mammographic density; BI-RADS: breast imaging reporting and data system.

Neoadjuvant systemic therapy

All study participants received a neoadjuvant systemic treatment plan customized by a multidisciplinary team. The NAST consisted of 3–4 cycles of epirubicin cyclophosphamide and 9–12 weekly cycles of paclitaxel [26]. Selected triple negative tumors were recommended an extended regimen with 4 and 12 cycles of epirubicin cyclophosphamide and paclitaxel, respectively, and certain cases received an addition of carboplatin. HER2-positive tumors were treated with a combination of chemotherapy and HER2 blockade. Prior to 2015, the HER2 blockage consisted only of trastuzumab. From 2015 and onwards, the HER2 blockade included both trastuzumab and pertuzumab. NAST compliance was monitored by review of medical records. Patients who deviated from the NAST protocol were defined as having received ‘suboptimal treatment’ (N = 30). An arbitrary 75% threshold defined ‘suboptimal treatment’, indicating <75% of the intended treatment was administered prior to surgery. Reasons for deviation included severe side effects, toxicity, and disease progression. One patient only received endocrine therapy with an aromatase inhibitor due to heart failure and one patient declined chemotherapy and only received HER2 blockade and were included in the ‘suboptimal treatment’ category.

Definitions

Clinicopathologic data were retrieved from digital medical records. NAST response was defined as pathological complete response (pCR) in the breast tissue, i.e., no detectable malignant cells in the surgical breast specimen (but ductal cancer in situ may be present). Treatment response had previously been established by board-certified pathologists subspecialized in breast pathology and documented in patient medical records according to the Miller–Payne five-point grading system [27]. During the final months of 2020, a new classification system was introduced, and response was instead classified according to the Residual Cancer Burden system [28].

Patients over the age of 55 were classified as postmenopausal if the medical record lacked notation on the menopausal status. The tumor grade had been assessed according to the Bloom–Richardson–Elston (BRE) system, which scores the tubule formation, nuclear pleomorphism, and mitotic activity [29]. For the analysis, the BRE score was translated to a corresponding Nottingham Histological Grade (NHG; BRE score 3–5 = NHG 1; BRE score 6–7 = NHG 2; BRE score 8–9 = NHG 3) [30]. Only three patients had grade 1 tumors. Thus, tumor grades 1 and 2 were merged and analyzed in comparison to grade 3 tumors. The biomarker status (estrogen receptor [ER], progesterone receptor [PR], Ki-67, and HER2 expression) had been assessed by use of immunohistochemistry on core biopsies. Tumors were classified as ER/PR positive if ≥10% and ER/PR negative if <10%, as per Swedish National Breast Cancer Treatment Guidelines [30]. The Ki-67 percentage was analyzed as a continuous parameter and expressed per decile increase in the regression analysis. Participants with a HER2 score of 2+ were considered positive only if the additional testing with silver in situ hybridization indicated gene amplification. The approximate breast cancer subtype was retrospectively decided based on the 2013 St Gallen classification [31], the NHG [32], and the updated Ki-67 cut-point recommendation [33], as per current Swedish guidelines [30]. The subtypes used were as follows: luminal A-like, luminal B-like/HER2-negative, luminal B-like/HER2-positive, non-luminal/HER2-positive, and triple negative. There were no cases of pCR among patients with luminal A-like tumors. Hence, the luminal A-like and luminal B-like/HER2-negative tumors were merged into a luminal/HER2-negative group and set as referent for the regression analysis.

Tumor-infiltrating lymphocytes

Pre-NAST stromal TILs in hematoxylin and eosin-stained core biopsies were microscopically assessed and retrospectively recorded by a board-certified breast pathologist (AK). The assessment considered all accessible biopsy samples from each patient, without focusing on hot spots. Details on the biopsies are provided in the supplementary data. TILs were scored as a semicontinuous parameter, describing the percentage of tumor stromal area infiltrated by mononuclear cells, as suggested by the International TILs Working Group [34]. The categories used were as follows; no TILs (<1%), low TILs (1%–9%), intermediate TILs (10%–49%), high TILs (50%–74%), and very high TILs (≥75%). The semicontinuous categories were used in Spearman’s rank order correlation test. For the logistic regression analysis, the pre-NAST TIL categories were dichotomized, following descriptive analysis of the variable distribution, creating a low TIL group (<10%) and a high TIL group (≥10%) [35].

Mammographic density

Pre-NAST mammograms were retrospectively assessed by a trained medical student (AL), assisted by an experienced radiologist (EM), both blinded to patient outcome and tumor data. The breast density was scored according to the breast imaging reporting and data system (BI-RADS) fifth edition [36]. The BI-RADS assessment is qualitative and consists of four categories, a, b, c, and d. Category a correlate to breasts which are ‘almost entirely fatty’, b to breasts with ‘scattered areas of fibroglandular density’, c to ‘heterogeneously dense’ breasts and d describe breasts that are ‘extremely dense’ [36].

Statistical methods

The one-way analysis of variance or the t-test were used to test for differences between groups of MD or TILs and pCR, on continuous, normally distributed measures (i.e., age). The Kruskal–Wallis test or Mann–Whitney U test were used to test differences between MD or TILs and pCR, and numerical variables that were not normally distributed (i.e., BMI, Ki-67, and tumor size). The Fisher’s exact tests were used to test for differences in categorical variables. Correlations were estimated by use of Spearman’s rank order correlation test [37]. Uni- and multivariable binary logistic regression were performed to identify factors significantly associated with pCR. The low TIL group and BI-RADS a were set as referents for TILs and MD. The multivariable models were established through stepwise elimination of non-confounding variables with insignificant p values. If two variables were highly correlated (i.e., Spearman’s rank order correlation coefficient above 0.6), one was excluded from the multivariable analysis based on the clinical importance of the variables.

All statistical analyses were performed using SPSS (IBM Corp. Released 2021. IBM SPSS Statistics for Macintosh, Version 28.0. Armonk, NY: IBM Corp). Odds ratios (OR) and 95% confidence intervals (CIs) were used to present the results. A p value of <.05 was considered statistically significant.

Results

Patient and tumor characteristics

Baseline clinicopathological characteristics stratified by TIL score are presented in Table 1. The median age of the population was 50 years (interquartile range [IQR]: 42–59) and the median BMI was 25 kg/m² (IQR: 22–29). 128 (40.6%) cases were postmenopausal and of those 186 (59.4%) were classified based on their age at diagnosis. The number of carcinomas in each approximate breast cancer subtype were 20 (6.3%), 69 (21.9%), 80 (25.4%), 63 (20.0%), and 82 (26.0%) for luminal A-like, luminal B-like/HER2-negative, luminal B-like/HER2-positive, non-luminal/HER2-positive and triple negative tumors, respectively. One (0.3%) carcinoma had no ER, PR, Ki-67, or HER2 data. Out of the 315 carcinomas, 136 (43.2%) accomplished pCR. Table 2 presents baseline characteristics according to pCR status.

Table 1. Clinicopathological characteristics stratified by tumor-infiltrating lymphocytes and of the total cohort.

	Low TILs (n = 215)	High TILs (n = 94)	p Value^a	Unknown TILs (n = 6)	Total cohort (N = 315)
Age			.008
Median (IQR)	51 (42–60)	47 (40–55)		56 (49–63)	50 (42–59)
BMI			.588
Median (IQR)	25 (22–29)	25 (22–27)		27 (25–33)	25 (22–29)
Unknown, n (%)	5 (2.3%)	4 (4.3%)		0 (0.0%)	9 (2.9%)
Tumor size (mm)			.250
Median (IQR)	38 (25–50)	35 (22–50)		57 (33–70)	37 (25–50)
Unknown, n (%)	2 (0.9%)	0 (0.0%)		0 (0.0%)	2 (0.6%)
Menopausal status, n (%)			.059
Premenopausal	121 (56.3%)	64 (68.1%)		2 (33.3%)	187 (59.4%)
Postmenopausal	94 (43.7%)	30 (31.9%)		4 (66.7%)	128 (40.6%)
Mammographic density, n (%)			.434
BI-RADS a	24 (11.2%)	11 (11.7%)		2 (33.3%)	37 (11.7%)
BI-RADS b	83 (38.6%)	37 (39.4%)		2 (33.3%)	122 (38.7%)
BI-RADS c	82 (38.1%)	29 (30.9%)		1 (16.7%)	112 (35.6%)
BI-RADS d	26 (12.1%)	17 (18.1%)		1 (16.7%)	44 (14.0%)
Cancer stage^b, n (%)			.625
1	33 (15.3%)	19 (20.2%)		1 (16.7%)	53 (16.8%)
2	119 (55.3%)	48 (51.1%)		2 (33.3%)	169 (53.7%)
3	57 (26.5%)	23 (24.5%)		3 (50.0%)	83 (26.3%)
4	6 (2.8%)	4 (4.3%)		0 (0.0%)	10 (3.2%)
Tumor grade, n (%)			.109
1	2 (0.9%)	1 (1.1%)		0 (0.0%)	3 (1%)
2	131 (60.9%)	45 (47.9%)		2 (33.3%)	178 (56.5%)
3	82 (38.1%)	46 (48.9%)		4 (66.7%)	132 (41.9%)
Unknown	0 (0.0%)	2 (2.1%)		0 (0.0%)	2 (0.6%)
Histological type			.046
Ductal	195 (90.7%)	91 (96.8%)		3 (50%)	289 (91.7%)
Lobular	19 (8.8%)	2 (2.1%)		3 (50%)	24 (7.6%)
Unknown	1 (0.5%)	1 (1.1%)		0 (0.0%)	2 (0.6%)
ER status, n (%)			.013
Positive	126 (58.6%)	40 (42.6%)		3 (50.0%)	169 (53.7%)
Negative	89 (41.4%)	54 (57.4%)		2 (33.3%)	145 (46.0%)
Unknown	0 (0.0%)	0 (0.0%)		1 (16.7%)	1 (0.3%)
PR status, n (%)			.023
Positive	94 (43.7%)	28 (29.8%)		3 (50.0%)	125 (39.7%)
Negative	121 (56.3%)	66 (70.2%)		2 (33.3%)	189 (60.0%)
Unknown	0 (0.0%)	0 (0.0%)		1 (16.7%)	1 (0.3%)
Ki-67			.011
Median (IQR)	50 (29–75)	60 (37–81)		53 (21–85)	54 (31–80)
Unknown, n (%)	0 (0.0%)	0 (0.0%)		1 (16.7%)	1 (0.3%)
HER2 status, n (%)			.710
Positive	97 (45.1%)	45 (47.9%)		2 (33.3%)	144 (45.7%)
Negative	118 (54.9%)	49 (52.1%)		4 (66.7%)	171 (54.3%)
Approximate subtypes, n (%)			.068
Luminal A-like	16 (7.4%)	3 (3.2%)		1 (16.7%)	20 (6.3%)
Luminal B-like/HER2-negative	53 (24.7%)	14 (14.9%)		2 (33.3%)	69 (21.9%)
Luminal/HER2-positive	57 (26.5%)	23 (24.5%)		0 (0.0%)	80 (25.4%)
Non-luminal/HER2-positive	40 (18.6%)	22 (23.4%)		1 (16.7%)	63 (20.0%)
Triple negative	49 (22.8%)	32 (34.0%)		1 (16.7%)	82 (26.0%)
Unknown	0 (0.0%)	0 (0.0%)		1 (16.7%)	1 (0.3%)
NAST, n (%)			.531
Optimal treatment	196 (91.2%)	83 (88.3%)		6 (100%)	285 (90.5%)
Suboptimal treatment^c	19 (8.8%)	11 (11.7%)		0 (0.0%)	30 (9.5%)
pCR, n (%)			<.001
Yes	75 (34.9%)	58 (61.7%)		3 (50.0%)	136 (43.2%)
No	140 (65.1%)	36 (38.3%)		3 (50.0%)	179 (56.8%)

^aTests for differences between low and high TILs. Missing data were excluded.

^bCancer stage was determined by use of the TNM classification.

^cSuboptimal treatment’ was defined as having received <75% of the intended treatment prior to surgery.

Abbreviations: TILs: tumor-infiltrating lymphocytes; IQR: interquartile range; BMI: body mass index; BI-RADS: breast imaging reporting and data system; ER: estrogen receptor; PR: progesterone receptor; HER2: human epidermal growth factor receptor 2; NAST: neoadjuvant systemic therapy; pCR: pathological complete response.

Table 2. Clinicopathological characteristics stratified by pathological complete response.

	No pCR (n = 179)	pCR (n = 136)	p Value^a
Age			.052
Median (IQR)	52 (43–59)	47 (40–58)
BMI			.206
Median (IQR)	26 (22–29)	25 (22–27)
Unknown, n (%)	3 (1.7%)	6 (4.4%)
Tumor size (mm)			.061
Median (IQR)	40 (26–50)	35 (20–50)
Unknown, n (%)	1 (0.6%)	1 (0.7%)
Menopausal status, n (%)			.105
Premenopausal	99 (55.3%)	88 (64.7%)
Postmenopausal	80 (44.7%)	48 (35.3%)
Mammographic density, n (%)			.051
BI-RADS a	24 (13.4%)	13 (9.6%)
BI-RADS b	64 (35.8%)	58 (42.6%)
BI-RADS c	72 (40.2%)	40 (29.4%)
BI-RADS d	19 (10.6%)	25 (18.4%)
Cancer stage^b, n (%)			.075
1	22 (12.3%)	31 (22.8%)
2	98 (54.7%)	71 (52.2%)
3	53 (29.6%)	30 (22.1%)
4	6 (3.4%)	4 (2.9%)
Tumor grade, n (%)			.184
1	3 (1.7%)	0 (0%)
2	106 (59.2%)	72 (52.9%)
3	69 (38.5%)	63 (46.3%)
Unknown	1 (0.6%)	1 (0.7%)
Histological type			.005
Ductal	157 (87.7%)	132 (97.1%)
Lobular	20 (11.2%)	4 (2.9%)
Unknown	2 (1.1%)	0 (0.0%)
ER status, n (%)			<.001
Positive	121 (67.6%)	48 (35.3%)
Negative	58 (32.4%)	87 (64.0%)
Unknown	0 (0.0%)	1 (0.7%)
PR status, n (%)			<.001
Positive	95 (53.1%)	30 (22.1%)
Negative	84 (46.9%)	105 (77.2%)
Unknown	0 (0.0%)	1 (0.7%)
Ki-67			.001
Median (IQR)	48 (27–73)	60 (35–80)
Unknown, n (%)	0 (0.0%)	1 (0.7%)
HER2 status, n (%)			<.001
Positive	52 (29.1%)	92 (67.6%)
Negative	127 (70.9%)	44 (32.4%)
Approximate subtypes, n (%)			<.001
Luminal A-like	20 (11.2%)	0 (0.0%)
Luminal B-like/HER2-negative	62 (34.6%)	7 (5.1%)
Luminal/HER2-positive	39 (21.8%)	41 (30.1%)
Non-luminal/HER2-positive	13 (7.3%)	50 (36.8%)
Triple negative	45 (25.1%)	37 (27.2%)
Unknown	0 (0.0%)	1 (0.7%)
TILs, n (%)			<.001
High	36 (20.1%)	58 (42.6%)
Low	140 (78.2%)	75 (55.1%)
Unknown	3 (1.7%)	3 (2.2%)
NAST, n (%)			.002
Optimal treatment	154 (86.0%)	131 (96.3%)
Suboptimal treatment^c	25 (14.0%)	5 (3.7%)

^aTests for differences between no pCR and pCR. Missing data were excluded.

^bCancer stage was determined by use of the TNM classification.

^c‘Suboptimal treatment’ was defined as having received <75% of the intended treatment prior to surgery.

Abbreviations: pCR: pathological complete response; IQR: interquartile range; BMI: body mass index; BI-RADS: breast imaging reporting and data system; ER: estrogen receptor; PR: progesterone receptor; HER2: human epidermal growth factor receptor 2; TILs: tumor-infiltrating lymphocytes; NAST: neoadjuvant systemic therapy.

Tumor-infiltrating lymphocytes

The number of carcinomas within each semicontinuous TIL category were 1 (0.3%), 214 (67.9%), 87 (27.6%), 6 (1.9%) and 1 (0.3%) for no TILs (<1%), low TILs (1%–9%), intermediate TILs (10%–49%), high TILs (50%–74%) and very high TILs (≥75%), respectively. Dichotomization with a 10% cut-point resulted in 215 (68.3%) carcinomas in the low TIL group (<10%) and 94 (29.8%) in the high TIL group (≥10%). Six carcinomas (1.9%) had no available TIL data.

In the high TIL group, the median age was lower (47 years vs. 51 years, p = .008), the Ki-67 index was higher (60 vs. 50, p = .011) and tumors were more likely to be ductal and hormone receptor negative (p = .046 for histological type, p = .013 for ER and p = .023 for PR) than in the low TIL group (Table 1). The distribution of HER2-positive tumors was similar between the high TIL group and low TIL group (p = .710; Table 1). TILs were not correlated to any other clinicopathological variable (Figure 2).

Figure 2. Correlation heatmap with coefficients of clinicopathological variables and significance levels calculated by use of Spearman’s rank correlation test [37].

Age, BMI, tumor size, Ki-67, TTM, and TTS are included as continuous variables. Menopausal status is classified as pre- or postmenopausal. Tumor grade is classified as grade I, II, and III. The cancer stage was determined by use of the TNM staging system and comprises stages 1–4. MD refers to BI-RADS categories a–d. TILs are included as a semicontinuous variable. ER, PR, and HER2 status are classified as negative or positive. pCR is classified as no or yes. NAST status was defined as suboptimal or optimal treatment and ‘suboptimal treatment’ was defined as having received <75% of the intended treatment prior to surgery. *p < .05; **p < .01; ***p < .001.

Abbreviations: BMI: body mass index; MD: mammographic density; TILs: tumor-infiltrating lymphocytes; ER: estrogen receptor; PR: progesterone receptor; HER2: human epidermal growth factor receptor 2; pCR: pathological complete response; NAST: neoadjuvant systemic therapy; TTM: time to diagnosis from mammography; TTS: time to surgery from diagnosis; BI-RADS: breast imaging reporting and data system.

Mammographic density

Table 3 presents baseline characteristics according to MD. The number of carcinomas categorized as BI-RADS a, b, c, and d were 37 (11.7%), 122 (38.7%), 112 (35.6%) and 44 (14.0%), respectively. Patients with less dense breasts were more likely to be overweight, older, and postmenopausal (−0.34, −0.47, −0.47, respectively; p < .0001 for all three correlation coefficients; Figure 2). There were no correlations between the BI-RADS categories and tumor grade, cancer stage, tumor size, histological type, Ki-67, ER, or HER2 status (Figure 2).

Table 3. Clinicopathological characteristics stratified by mammographic density categories (BI-RADS fifth edition).

	BI-RADS a (n = 37)	BI-RADS b (n = 122)	BI-RADS c (n = 112)	BI-RADS d (n = 44)	p Value^a
Age					<.001
Median (IQR)	60 (54–64)	56 (47–63)	47 (40–53)	42 (37–48)
BMI					<.001
Median (IQR)	29 (26–34)	25 (23–29)	25 (22–27)	23 (20–24)
Unknown, n (%)	0 (0.0%)	4 (3.3%)	4 (3.6%)	1 (2.3%)
Tumor size (mm)					.869
Median (IQR)	36 (27–55)	40 (26–50)	35 (22–50)	40 (22–50)
Unknown, n (%)	0 (0.0%)	1 (0.8%)	1 (0.9%)	0 (0.0%)
Menopausal status, n (%)					<.001
Premenopausal	7 (18.9%)	53 (43.4%)	87 (77.7%)	40 (90.9%)
Postmenopausal	30 (81.1%)	69 (56.6%)	25 (22.3%)	4 (9.1%)
Cancer stage^b, n (%)					.927^d
1	4 (10.8%)	20 (16.4%)	22 (19.6%)	7 (15.9%)
2	21 (56.8%)	67 (54.9%)	56 (50.0%)	25 (56.8%)
3	12 (32.4%)	28 (23.0%)	31 (27.7%)	12 (27.3%)
4	0 (0.0%)	7 (5.7%)	3 (2.7%)	0 (0.0%)
Tumor grade, n (%)					.123^d
1	0 (0.0%)	1 (0.8%)	2 (1.8%)	0 (0.0%)
2	18 (48.6%)	62 (50.8%)	69 (61.6%)	29 (65.9%)
3	19 (51.4%)	58 (47.5%)	40 (35.7%)	15 (34.1%)
Unknown	0 (0.0%)	1 (0.8%)	1 (0.9%)	0 (0.0%)
Histological type					.464
Ductal	35 (94.6%)	114 (93.4%)	99 (88.4%)	41 (93.2%)
Lobular	1 (2.7%)	8 (6.6%)	12 (10.7%)	3 (6.8%)
Unknown	1 (2.7%)	0 (0.0%)	1 (0.9%)	0 (0.0%)
ER status, n (%)					.656
Positive	22 (59.5%)	60 (49.2%)	62 (55.4%)	25 (56.8%)
Negative	15 (40.5%)	61 (50.0%)	50 (44.6%)	19 (43.2%)
Unknown	0 (0.0%)	1 (0.8%)	0 (0.0%)	0 (0.0%)
PR status, n (%)					.199
Positive	18 (48.6%)	40 (32.8%)	50 (44.6%)	17 (38.6%)
Negative	19 (51.4%)	81 (66.4%)	62 (55.4%)	27 (61.4%)
Unknown	0 (0.0%)	1 (0.8%)	0 (0.0%)	0 (0.0%)
Ki-67					.477
Median (IQR)	60 (35–80)	60 (34–80)	50 (26–79)	49 (27–80)
Unknown, n (%)	0 (0.0%)	1 (0.8%)	0 (0.0%)	0 (0.0%)
HER2 status, n (%)					.340
Positive	13 (35.1%)	62 (50.8%)	48 (42.9%)	21 (47.7%)
Negative	24 (64.9%)	60 (49.2%)	64 (57.1%)	23 (52.3%)
Approximate subtypes, n (%)					.082^c
Luminal A-like	4 (10.8%)	5 (4.1%)	7 (6.3%)	4 (9.1%)
Luminal B-like/HER2-negative	12 (32.4%)	18 (14.8%)	31 (27.7%)	8 (18.2%)
Luminal/HER2-positive	6 (16.2%)	37 (30.3%)	24 (21.4%)	13 (29.5%)
Non-luminal/HER2-positive	7 (18.9%)	24 (19.7%)	24 (21.4%)	8 (18.2%)
Triple negative	8 (21.6%)	37 (30.3%)	26 (23.2%)	11 (25.0%)
Unknown	0 (0.0%)	1 (0.8%)	0 (0.0%)	0 (0.0%)
TILs, n (%)					.434
High	11 (29.7%)	37 (30.3%)	29 (25.9%)	17 (38.6%)
Low	24 (64.9%)	83 (68.0%)	82 (73.2%)	26 (59.1%)
Unknown	2 (5.4%)	2 (1.6%)	1 (0.9%)	1 (2.3%)
NAST, n (%)					.724
Optimal treatment	33 (89.2%)	109 (89.3%)	104 (92.9%)	39 (88.6%)
Suboptimal treatment^d	4 (10.8%)	13 (10.7%)	8 (7.1%)	5 (11.4%)
pCR, n (%)					.051
Yes	13 (35.1%)	58 (47.5%)	40 (35.7%)	25 (56.8%)
No	24 (64.9%)	64 (52.5%)	72 (64.3%)	19 (43.2%)

^aTests for differences between BI-RADS categories. Missing data were excluded.

^bCancer stage was determined by use of the TNM classification.

^cAlternative subgroups were used due to nonconvergence in Fisher’s exact test. Cancer stages 3 and 4 were merged and compared to stage 1 and stage 2, respectively. Tumor grades 1 and 2 were merged and compared to grade 3. The luminal/HER2-negative subtypes were compared to the combined Luminal B-like/HER2-positive and Non-luminal/HER2-positive group and to the triple negative subtype.

^d‘Suboptimal treatment’ was defined as having received <75% of the intended treatment prior to surgery.

Abbreviations: BI-RADS: breast imaging reporting and data system; IQR: interquartile range; BMI: body mass index; ER: estrogen receptor; PR: progesterone receptor; HER2: human epidermal growth factor receptor 2; TILs: tumor-infiltrating lymphocytes; NAST: neoadjuvant systemic therapy; pCR: pathological complete response.

Univariable analyses

The univariable logistic regression analysis is presented in Table 4 and was performed on the following variables: age, menopausal status, BMI, tumor size, tumor grade, cancer stage, histological type, MD, TILs, ER status, PR status, HER2 status, Ki-67, approximate subtype, NAST status (optimal completion of NAST or not), time in days between mammography and diagnosis, and time in days between diagnosis and surgery.

Table 4. Univariable and multivariable binary logistic regression analysis of clinicopathological variables associated with pathological complete response.

	Total cohort	Univariable analysis		Multivariable model 1^a		Multivariable model 2^b
Variable	N (N with pCR)	OR (95% CI)	p Value	OR (95% CI)	p Value	OR (95% CI)	p Value
Age	315 (136)	0.98 (0.96–1.00)	.053	1.00 (0.97–1.02)	.715
Mammographic density
BI-RADS a	37 (13)	1.00 (reference)		1.00 (reference)		1.00 (reference)
BI-RADS b	122 (58)	1.67 (0.78–3.59)	.186	1.29 (0.49–3.43)	.642	1.36 (0.52–3.61)	.532
BI-RADS c	112 (40)	1.03 (0.47–2.23)	.949	0.72 (0.31–2.27)	.723	0.90 (0.33–2.44)	.832
BI-RADS d	44 (25)	2.43 (0.99–5.98)	.053	2.48 (0.77–7.96)	.128	2.51 (0.78–8.04)	.121
Cancer stage^c
1	53 (31)	1.00 (reference)		1.00 (reference)
2	169 (71)	0.51 (0.28–0.96)	.037	0.52 (0.24–1.13)	.098
3	83 (30)	0.40 (0.20–0.81)	.011	0.42 (0.17–1.05)	.063
4	10 (4)	0.47 (0.12–1.88)	.287	0.54 (0.11–2.66)	.452
Histological type
Ductal	289 (132)	1.00 (reference)		1.00 (reference)
Lobular	20 (4)	0.24 (0.08–0.71)	.010	0.90 (0.21–3.81)	.884
ER status
Negative	145 (87)	1.00 (reference)		1.00 (reference)		1.00 (reference)
Positive	169 (48)	0.26 (0.17–0.42)	<.001	0.20 (0.11–0.38)	<.001	0.20 (0.11–0.38)	<.001
PR status
Negative	189 (105)	1.00 (reference)
Positive	125 (30)	0.25 (0.15–0.42)	<.001
Ki-67^d	314 (135)	1.16 (1.06–1.26)	.001	1.16 (1.03–1.30)	.015	1.17 (1.04–1.31)	.009
HER2 status
Negative	171 (44)	1.00 (reference)		1.00 (reference)		1.00 (reference)
Positive	144 (92)	5.11 (3.15–8.28)	<.001	9.54 (4.92–18.50)	<.001	9.04 (4.72–17.31)	<.001
Subtype
Luminal/HER2-negative	90 (8)	1.00 (reference)
Luminal/HER2-positive	78 (39)	12.32 (5.07–29.92)	<.001
Non-luminal/HER2-positive	64 (50)	45.06 (16.85–120.51)	<.001
Triple negative	82 (38)	9.63 (3.97–23.36)	<.001
TILs
Low	215 (75)	1.00 (reference)		1.00 (reference)		1.00 (reference)
High	94 (58)	3.01 (1.82–4.97)	<.001	2.98 (1.60–5.56)	<.001	2.95 (1.59–5.46)	<.001
NAST
Suboptimal treatment^e	30 (5)	1.00 (reference)		1.00 (reference)		1.00 (reference)
Optimal treatment	285 (131)	4.25 (1.58–11.42)	.004	3.83 (1.19–12.27)	.024	4.23 (1.34–13.37)	.014

^aModel 1 displays the unadjusted multivariable analysis.

^bModel 2 shows the fully adjusted multivariable analysis following stepwise elimination of highly correlated, non-confounding, and unsignificant variables.

^cCancer stage was determined by use of the TNM classification.

^dKi-67 is displayed per decile increase.

^e‘Suboptimal treatment’ was defined as having received <75% of the intended treatment prior to surgery.

Abbreviations: N: number of patients; pCR: pathological complete response; OR: odds ratio; CI: confidence interval; BI-RADS: breast imaging reporting and data system; ER: estrogen receptor; PR: progesterone receptor; HER2: human epidermal growth factor receptor 2; TILs: tumor-infiltrating lymphocytes; NAST: neoadjuvant systemic therapy.

The univariable analysis of TILs shows carcinomas in the high TIL group were more likely to achieve pCR compared to carcinomas with low TILs (OR: 3.01; 95% CI: 1.82–4.97; p < .001; Table 4).

The univariable analysis of MD indicated carcinomas in extremely dense breasts (BI-RADS d) were more likely to accomplish pCR compared to carcinomas in non-dense breasts (BI-RADS a; OR: 2.43; 95% CI: 0.99–5.98; p = .053). However, this association did not reach statistical significance. BI-RADS categories b and c were not associated with pCR; category b versus a (OR: 1.67; 95% CI: 0.78–3.59; p = .186) and category c versus a (OR: 1.03; 95% CI: 0.47–2.23; p = .949; Table 4).

Multivariable analyses

The multivariable models are presented in Table 4. TILs were a significant predictor of NAST response in the adjusted multivariable model (OR: 2.95; 95% CI: 1.59–5.46; p < .001). Conversely, the adjusted multivariable analysis rendered no statistically significant association between MD and pCR; category d versus a (OR: 2.51; 95% CI: 0.78–8.04; p = .121), category c versus a (OR: 0.90; 95% CI: 0.33–2.44; p = .832) and category b versus a (OR: 1.36; 95% CI: 0.52–3.61; p = .532; Table 4). Ki-67, NAST status, ER status, and HER2 status were all independent predictors of pCR in the adjusted multivariable model. The OR and 95% CIs were as follows: 1.17 (1.04–1.31) for Ki-67, 4.23 (1.34–13.37) for NAST status, 0.20 (0.11–0.38) for ER status and 9.04 (4.72–17.31) for HER2 status (Table 4).

Mammographic density and tumor-infiltrating lymphocytes correlation

No correlation was found between MD and TILs (0.02; p = .685; Figure 2).

Discussion

NAST is a well-established breast cancer treatment with long-term benefits comparable with those of post-surgical chemotherapy. However, clinicians lack tools to accurately predict NAST responders in the heterogenous group of breast cancer patients. We studied female breast cancer patients treated with NAST and chose pCR in surgical breast specimen as the primary endpoint. Our findings demonstrate that TILs significantly predict NAST response in both the univariable and multivariable analyses. In the univariable analysis, extremely dense breasts (BI-RADS d) showed a tendency of a higher likelihood for pCR following NAST compared with non-dense breasts (BI-RADS a). However, the association did not reach statistical significance and there was no gradual increment in the estimated pCR likelihood when moving from less dense breast to denser breast tissue. MD was not associated with pCR in the multivariable analysis. Ki-67, NAST status, ER status, and HER2 status were significant predictors in both the univariable and the multivariable analyses. There was no correlation between TILs and MD.

Tumor-infiltrating lymphocytes

High levels of pre-treatment stromal TILs have been associated with pCR following NAST [7–10]. The association is most pronounced in HER2-positive and triple negative breast cancers and is hypothesized to emanate from different levels and compositions of the lymphocyte phenotypes, compared to that of luminal/HER2-negative breast cancers [7,8]. For example, triple negative and HER2-positive breast cancers more frequently have high TIL levels [9,38]. Moreover, these subtypes have a greater infiltration of CD8+ cytotoxic T cells and FOXP3+ T regulatory cells, both of which predict a higher likelihood of NAST response [8,38].

Our study showed that TILs were predictive of NAST response, consistent with previous reports [7–10]. Due to a limited number of participants in our study, we were unable to conduct subanalyses based on breast cancer subtype. Regarding the descriptive analyses, the HER2 expression did not differ between tumors with high TILs and those with low TILs in this study. Age, Ki-67, and ER status differed significantly among the TIL groups, but there was no correlation between TILs and these variables. We attribute this lack of correlation to the composition of our cohort, which includes only aggressive tumors, specifically selected for NAST. Consequently, the spectrum of breast carcinomas is only partially represented. The lack of correlation may also relate to the 10% cut-point. While it demonstrated significant predictive value for TILs in this study, other studies suggest that a cut-point of 20% may be more favorable if available [8]. Furthermore, the spatial heterogeneity of lymphocytes in carcinomas and the potential inadequacy of the biopsy material may contribute to the absence of correlations [39,40]. Notably, this limitation has been particularly observed in HER2-positive carcinomas [39].

Future research should use larger cohorts and explore the immunologic subpopulations and the spatial organization of TILs. Such investigations could determine the extent to which these features impact the predictive value of TILs [7,8,41].

Mammographic density

This study investigated MD as a predictive factor for NAST response. Previous literature is conflicting regarding the association between MD and pCR. Some studies have failed to demonstrate an association [17–19] while others have found a higher likelihood of pCR in patients with non-dense breasts [14–16]. By contrast, Di Cosimo et al. found that dense breasts were associated with pCR [20]. There are multiple possible explanations for these conflicting findings. Early studies of NAST response predominantly investigated patients with locally advanced tumors. In 2017, the neoadjuvant treatment guidelines changed, enabling the observation of predictive parameters in more generalized breast cancer populations [42]. Di Cosimo et al. stratified the multivariable analysis by breast cancer subtypes [20]. They showed that the increased likelihood of pCR for BI-RADS categories b and c was present only in the triple negative and HER2-positive subtypes, not in the combined luminal A and B subtype-group. We hypothesize that the predictive value of MD differs based on breast cancer subtype, consequently displaying inconsistent predictive roles for said variable depending on the population composition. This needs to be explored further in well-powered studies.

Another reason for the discordant results across studies is the use of different classification systems and small cohorts. Elsamany and colleagues [14] used a binary and percentage-based classification of MD. Notable, the authors of this article were unable to detect an association between ER status and pCR. In the study by Skarping and colleagues [15] only 18.9% of the patients accomplished pCR leaving very few cases of pCR within each BI-RADS category. In the study by Cullinane et al. [19] there were no participants classified as BI-RADS a, and no available BMI data.

Our study confirms the well-established, inverse relationship between MD and age, BMI, and menopausal status [43]. Additionally, all four BI-RADS categories were represented and 43.2% of the participants achieved pCR, thus allowing for increased accuracy when analyzing the outcome for each BI-RADS category. Interestingly, our results are in line with Di Cosimo’s study, indicating a possibility of higher likelihood of pCR in extremely dense breasts compared with non-dense breasts in the univariable analysis. This association was not statistically significant. Still, the estimated value was relatively high and consistent, and thus we encourage further exploration of the predictive role of MD.

There is no clear biological explanation for a possible predictive role of MD. Previous reports have suggested MD is an important host factor [15,16,20], due to its reflection of the stromal environment [21], the level of growth factors [44,45], and the extent of immune cell infiltration [21,22]. We agree that these factors could influence the NAST response and additionally advocate the perhaps underrated importance of the mammary adipose tissue. Studies have shown that mammary adipocytes can promote chemoresistance [46,47] which could explain a lower likelihood of NAST response in fatty breasts compared to dense breasts. Further research is required to clarify the relationship between stroma, fat tissue, radiological features, and the impact of NAST.

The relationship between TILs and MD

Although studies have illustrated an association between immune cell presence and MD [21,22], we found no correlation between TILs and MD. However, we believe our cohort’s restricted breast carcinoma spectrum and the TILs threshold of 10% limit the chances of correlation identification, as described previously. Furthermore, the estimate of MD according to BI-RADS may be too imprecise and unable to unveil such a link. Hence, the TILs and MD correlation should be further investigated using finer radiological measurements and larger sample sizes.

Methodological considerations

This study has several limitations. It is a retrospective study with limited numbers of carcinomas within each approximate breast cancer subtype. Furthermore, the distribution of TILs can be heterogenous in the tumor stroma [39]. Therefore, a relatively small core biopsy material might not adequately represent TILs of the whole tumor [39,40,48]. Visual TIL scoring is also subject to inter-observer variability [49]. In this study, TILs were assessed by a single but board-certified and breast-specialized pathologist (AK), qualifications which enhance the reliability of the assessment [50]. Moreover, the BI-RADS fifth edition is a subjective scoring system that provides a gross estimate of breast density [51]. Therefore, we suggest future studies use objective, quantitative software to determine both TILs and breast density percentage. This minimizes subjective influences and allows TILs and MD to be analyzed as continuous variables.

Regarding the definition of ‘suboptimal treatment’ of NAST, our standpoint is that anticancer therapy is not a binary process but rather relies on the cumulative effect of the administered doses [52]. Thus, to minimize the exclusion of patients with a presumed comparable chance of achieving pCR, we opted for the arbitrary threshold of 75%.

We chose pCR in the surgical breast specimen as our surrogate endpoint for NAST response [53]. This decision considered the local assessment of MD and TILs and the fact that MD partially reflects the breast tissue biology. Studying the local response facilitates interpretation of our results and the subsequent explanatory, tissue biology-based hypotheses. However, the association between pCR and favorable long-term outcome is strongest when also including residual-free lymph nodes in the pCR definition, which we did not [53]. Therefore, careful consideration must precede the interpretation of our results regarding long-term outcomes.

Conclusions

In summary, this study demonstrates that core biopsy stromal TILs significantly predict pCR following NAST. We believe this biomarker should receive greater consideration at multidisciplinary conferences assessing NAST benefits. MD had no significant predictive value in this study. Additional research of extensive cohorts is needed to establish the true predictive value of MD. We found no correlation between MD and TILs.

Prior presentation

Preliminary results were presented at the Swedish conference ‘Kirurgveckan’ in Stockholm, Sweden on 25 August 2022.

Author contributions

AL, KC, EH, AST, and PK conceived the study. KC provided study participants. AL, KC, AK, and EM assembled and collected data. AL and EH analyzed the data. AL, EH, AST, and PK interpreted the results. AL drafted the manuscript. KC, EH, AST, and PK revised the manuscript. All authors read and approved the manuscript.

Acknowledgements

We thank the mammographic department at our institution for their assistance with the collection and evaluation of mammograms.

Disclosure statement

PK and EH report interests not related to the current study: contract with PFS Genomics/Exact Sciences regarding genomic profiling; co-inventor on patent applications; contract with Prelude Dx. AST reports interest not related to the current study: co-inventor on patent applications; contract with Prelude Dx. No disclosures were reported by the other authors.

Data availability statement

The data that support the findings of this study are available on request from the corresponding author.

Additional information

Funding

This work was supported by the Swedish Cancer Society [Grant No. 21 1889 S], the Swedish Research Council [Grant No. 2021-0138], and the Swedish governmental ALF agreement [Grant No. ALFGBG-965020].