Analytical validity of CD20 as a biomarker in breast cancer patients using tissue microarrays and whole slide sections

ABSTRACT

Background

Clinical utility of biomarker-based treatment depends on both analytical and clinical validity. CD20+ B-lymphocytes often form nodular lymphoid structures within tumors, presenting potential analytical issues if using tissue microarrays (TMA). We compared CD20 expression in TMAs and whole-slide sections (WS) using digital image analysis.

Materials and methods

Immunohistochemical CD20+ expression was assessed on 1 TMA core/patient and corresponding WS from 221 breast cancer patients. CD20+ area fractions were quantified within four spatial tissue compartments: (1) tumor epithelium; (2) tumor margin; (3) tumor-related stroma; (4) total tumor area. Bland-Altman analysis and t-test compared CD20+ area fractions of WS and TMA.

Results

CD20+ cells were predominantly located in the stroma and often formed nodules. Across all compartments, the absolute area fraction of CD20+ was significantly lower in TMAs than in WS. Mean ratios of CD20+ area fractions (WS vs. TMA) ranged from 2.5 (95% confidence interval: 1.7–3.6; p < 0.001) to 6.4 (4.7–8.9; p < 0.001) within the four spatial regions.

Conclusions

These findings indicate that TMAs may underestimate the presence of CD20+ cells in breast cancer, presumably due to the formation of nodules. This heterogeneity in distribution is important to consider when exploring the use of CD20 as a potential clinical biomarker.

1. Introduction

Tumor-infiltrating lymphocytes (TILs) are currently being extensively studied as prognostic and predictive markers in numerous tumor types, including breast cancer (BC) [1]. Level 1 evidence supports that high TILs levels in the primary tumor are associated with improved disease-free survival in the more aggressive BC subtypes (HER2-positive and triple-negative (TN) BC) [2–4]. TILs have also been found to be predictive of pathological complete response (pCR) following neoadjuvant chemotherapy in all BC subtypes [2,5].

TILs comprise a diverse population of various immune cells, predominantly T-lymphocytes but also natural killer cells, B-lymphocytes, and plasma cells [6]. The immune cell composition likely influences their prognostic and predictive value, with most studies focusing on CD8+ T-lymphocytes, which have been associated with improved survival and systemic treatment response in BC [7–11]. Studies have also reported a positive association between the presence of tumor-infiltrating B-lymphocytes (TIL-B) and improved prognosis in terms of disease-free survival and overall survival, with the strongest association in TNBC and HER2-positive tumors [12–15]. Memory B-lymphocytes, in particular, have been associated with positive prognostic outcomes [13,16,17]. On the contrary, some studies have reported negative or no association between high plasma cell levels and prognosis [16,18,19]. In the neoadjuvant setting, a positive correlation between TIL-B levels and pCR has also been described [20–23].

The presence of TIL-B has been described in various cancer types [24–28], and their spatial distribution has been described as highly heterogeneous. TIL-B may be diffusely scattered throughout the tumor, form dense lymphoid aggregates, or may be part of tertiary lymphoid structures (TLS) [16,17], which are well-organized structures, with a central B-lymphocyte follicle surrounded by a T-lymphocyte zone [13]. Conversely, lymphoid aggregates contain randomly distributed immune cells, with no organization. High levels of B-cell infiltration are strongly associated with the presence of lymphoid aggregates, suggesting that B-cells preferentially localize within organized immune structures rather than being randomly distributed throughout the tumor microenvironment [29].

Despite increasing interest in TIL-B as prognostic and predictive markers in breast cancer, the analytical validity of CD20 assessment remains insufficiently explored, particularly when tissue microarrays (TMAs) are used instead of whole-slide sections (WS). Given the spatial heterogeneity of TIL-B, sampling strategy is likely to influence their quantification. Previous findings have demonstrated that limited tissue sampling may underestimate immune-related effect sizes for TIL-B, likely due to their spatial clustering and the inability of small tissue samples to capture such heterogeneity [30].

The clinical utility of a biomarker-based treatment demands not only clinical validity showing significant results from clinical studies, but also robust analytical validity, proving robustness and reproducibility of the biomarker [31]. It is, as such, of great importance to secure analytical validity during identification of new potential biomarkers.

This study aimed to construct a methodological report, testing the use of a single 1 mm central TMA core compared to WS in the evaluation of CD20 (biomarker for B-lymphocyte) in BC using digital image analysis (DIA) and to examine possible variations according to spatial locations of the cells.

2. Materials and methods

2.1. Patients

The material originated from the Danish Breast Cancer Group (DBCG) 82bc cohort diagnosed with early BC in the period 1982–1990. Patients were treated with mastectomy and partial axillary dissection, followed by systemic therapy, and randomized to post-mastectomy radiotherapy or no radiotherapy. A detailed description of the cohort has previously been published [32–34].

Formalin-fixed, paraffin-embedded (FFPE) tumor blocks from the original surgical specimens were, as part of previous studies, collected from 1078 DBCG82bc patients [34–36]. TMAs with a single, central 1.00 mm tumor-containing core from each FFPE block had already been constructed for these previous studies, preceded by verification of carcinoma content [35–37].

Tumor-containing paraffin blocks and corresponding TMA cores of 221 patients were selected for the present study. The cohort encompassed 124 patients previously included in a study of interobserver variation in evaluation of TILs and are representative for patients included in the original biomarker studies of the DCG82bc cohort [38]. However, since only 10% of breast cancer patients are expected to have high levels of TILs, the study cohort was further supplemented with 97 patients with high TILs (≥30%) to ensure that the study population encompassed tumor samples with differing levels of TILs. The 97 patients were identified from a previous study of TILs in the DBCG82bc cohort [34] (Supplementary Figure 1, Consort diagram).

2.2. Multiplex immunohistochemical analysis

Three µm-thick sections were made from the TMAs and the corresponding whole FFPE blocks from each patient. The sections were prepared for multiplex immunohistochemistry (IHC) and staining procedures were performed on the Ventana Discovery Ultra Autostainer (Roche Diagnostics International AG, Switzerland). Detailed descriptions of procedures for tissue handling and staining protocols have been described previously [39].

As part of a larger study exploring the different immune cell subsets in the DBCG82bc material [40], multiplex IHC was performed including markers for CD20 (clone L26, RTU, Ventana) as well as markers for tumor epithelium (pan-cytokeratin, clone PG-M1, dilution 1:100, Dako) and myoepithelial cells (p63, clone 4A4, RTU, Ventana) without hematoxylin counterstaining. External control tissue samples were included on each slide to verify that the multiplex IHC staining protocol performed as intended. Stainings for CD66b (neutrophils) and CD68 (macrophages) were also carried out on the same sections, but results are reported elsewhere.

2.3. Automated digital image analysis and area fraction calculations

Stained sections were digitally scanned by Nanozoomer 2.0HT; Hamamatsu Photonics KK, Japan. IHC analysis for estrogen receptor (ER) status (cut off 1%) was available from previous studies [36]. Automated DIA and alignment of multiplex slides with p63 stainings in order to differentiate between invasive carcinoma and in situ lesions was performed in Visiopharm version 2023.09.5.15777 (Visiopharm A/S, Hørsholm, Denmark) on WS and TMA cores. DIA allowed for quantitative estimation of each marker expression together with information on the spatial localization of the immune cells (intraepithelial, stromal, etc.). Using deep learning with manual pixel-level annotations, a convolutional neural network (U-net) was trained on 50 independent TMA cores to recognize and annotate areas corresponding to the IHC staining colors. Concurrent validation was performed on 60 independent TMA cores on 3 slides (per-class F1 score > 0.99). Supplementary Tables 1–2 provide details on network architecture, training parameters and annotation data, and Supplementary Table 3 displays the network’s per-class segmentation performance for 40 subimages (512×512 pixels) from 20 TMA cores and 20 WS images.

Regions of interest (ROIs) that encompassed the entire tumor area while excluding DCIS were automatically delineated (Supplementary Figure 2); subsequently, artifacts and necrosis were manually excluded. Artifacts were defined as tissue folds, tearing, air bubbles, debris and staining artifacts. Necrotic areas were identified morphologically by loss of tissue architecture substituted by avital areas of eosinophilic material without vital cell nuclei but with presence of cellular debris and neutrophilic granulocytes. ROIs were reviewed and manually adjusted if necessary. In both WS and TMA, areas (in µm²) of CD20+ cells were automatically quantified within four spatial tissue compartments. Region 1 (Tumor) was defined as tumor-epithelial islands identified by pancytokeratin-positive tumor epithelial cells, and CD20+ cells in this region were considered present if located within or in direct contact with tumor epithelium. Region 2 (Tumor margin) was defined as the interface between tumor epithelium and stroma, defined as stroma within 20 µm within proximity of the tumor epithelial islands. Based on experiences from previous studies [39,41,42], the margin of 20 µm was selected to capture the immediate stromal interface surrounding tumor epithelial islands, where immune cell-tumor epithelial cell interactions are most likely to occur and from the assumption that most immune cells measure approximately 7–20 µm. Region 3 (Stroma) included the interposed tumor-related stroma (Figure 1). Region 4 (Total tumor area) included the entire tumor area in TMA or WS, i.e., the sum of regions 1 (tumor) and 3 (stroma).

Figure 1.

Multiplex immunohistochemistry slide (A) with corresponding digital image annotation of the area inside the squared region (B). The area fraction of CD20+ cells was calculated for 4 tissue compartments: Region 1 (Tumor), defined as tumor-epithelial islands with CD20+ cells being present within or in direct contact with tumor epithelium; Region 2 (Tumor margin) defined as the interface between tumor epithelium and stroma equal to stroma within 20 µm within proximity of the tumor epithelial islands (marked in B as dark gray areas; Region 3 (Stroma) defined as the interposed tumor-related stroma; and Region 4 (Total tumor area) including the entire tumor area in TMA or WS (not highlighted in this figure). CD20+ cells are marked as turquoise, or as dark blue if inside the dark gray 20 µm margin. Tumor epithelial cells: yellow. CD68 macrophages: pink. CD66b neutrophilic granulocytes: brown. Magnification 23x.

Area fractions of CD20+ cells were calculated as absolute fractions for each region, where region X is region 1–4:

$$\mathrm{Area}\mathrm{fractions}\left(\mathrm{absolute}\right)=\frac{\mathrm{Area}\mathrm{of}\mathrm{CD}20+\mathrm{cells}\mathrm{in}\mathrm{region}\mathrm{X}}{\mathrm{Total}\mathrm{area}\mathrm{of}\mathrm{region}4\left(\mathrm{total}\mathrm{tumor}\mathrm{area}\right)}\mathrm{x}100\mathrm{\%}$$ (1)

Relative area fractions were calculated for each of the sub-regions, where region Y is region 1–3 as:

$$\mathrm{Area}\mathrm{fractions}\left(\mathrm{relative}\right)=\frac{\mathrm{Area}\mathrm{of}\mathrm{CD}20+\mathrm{cells}\mathrm{in}\mathrm{region}\mathrm{Y}}{\mathrm{Total}\mathrm{area}\mathrm{of}\mathrm{region}\mathrm{Y}}\mathrm{x}100\mathrm{\%}$$ (2)

Area fractions for all cells in regions 1–3 were calculated as:

$$\mathrm{Area}\mathrm{fractions}\left(\mathrm{region}\right)=\frac{\mathrm{Total}\mathrm{area}\mathrm{of}\mathrm{region}\mathrm{Y}}{\mathrm{Total}\mathrm{area}\mathrm{of}\mathrm{region}4\left(\mathrm{total}\mathrm{tumor}\mathrm{area}\right)}\mathrm{x}100\mathrm{\%}$$ (3)

2.4. Statistical analysis

Correlations were analyzed using scatter plots and Lin’s Concordance Correlation Coefficient (CCC) [43]. Area fractions of CD20 were log-transformed to reduce heteroscedasticity and achieve normality. Normality was checked before and after transformation by histograms, Q–Q plots and the Shapiro–Wilk test. After transformation, p-values of the Shapiro–Wilk tests were 0.91 for region 1 (tumor epithelium), 0.09 for region 2 (tumor margin), 0.83 for region 3 (stroma) and 0.88 for region 4 (total tumor area). Zero values (undefined under log-transformation) were excluded from statistical analyses but assigned a small constant for visualization in scatter plots. The agreements between the two methods were analyzed by Bland-Altman (BA) analysis, including limits of agreement (LoA) encompassing 95% of the differences between the two methods [44]. As data were log-transformed, ratios between CD20+ area fractions on WS and TMA were plotted against the geometric means of CD20+ area fractions in the two methods [45]. Paired sample t-test was used to compare the log-transformed CD20+ area fractions of WS and TMA and the average difference (mean ratio) is included in the BA plots with 95% confidence intervals (CI). CD20+ area fractions in the entire lesions (region 4) in WS and TMA were stratified according to ER status and illustrated by strip plots. Median values were tested using Wilcoxon signed-rank tests. Statistical analysis was performed using Stata-version 17.0 (StataCorp, College Station, TX, USA). p-values less than 0.05 were considered statistically significant. Power calculations were not performed as part of planning the study, since the available patient material originated from a previous cohort already collected.

2.5. Ethics

This retrospective biomarker study was performed using tissue samples and clinical data from the original prospective randomized DBCG82bc cohort. The study was approved by The Central Denmark Region Committees on Health Research Ethics (journal no. 1-10-72-45-17), The Danish Data Protection Agency (journal no. 1-16-02-698-17), and the DBCG. The reporting of data follows the Recommendation for Tumor Marker Prognostic studies (REMARK) criteria [46], and was conducted in accordance with the Declaration of Helsinki. Verbal informed consent was mandatory from all participants prior to inclusion in the original randomized DBCG82bc trials. The DBCG82bc cohort has been described previously [32–37,47,48].

3. Results

In general, CD20+ B-lymphocytes were heterogeneously distributed in the tumor and often formed lymphoid aggregates in the tumor periphery rather than being diffusely scattered within the tumor (Figure 2). Figure 2 also illustrates how information on the spatial context of CD20 expression in a TMA core is limited.

Figure 2.

Multiplex immunohistochemistry for whole slide (WS) overview illustrating CD20+ cells (turquoise) forming nodules/lymphoid aggregates within the tumor (yellow = tumor epithelial cells) (A). Corresponding tissue microarray (TMA) core from the same tissue section (B).

The median area fractions of CD20+ were significantly higher in WS compared to TMA cores, both when calculated as absolute area fractions (area of CD20+ relative to the total tumor area) and as relative area fractions (area of CD20+ relative to sub-regions). The median absolute area fraction for the total tumor area (region 4) was 0.44% in WS vs 0.037% in TMA (p < 0.001) (Table 1). Similar differences were observed when measuring absolute levels in tumor epithelial islands alone, stroma alone, and tumor margin alone (regions 1–3) (Table 1).

Table 1.

Expression of CD20+ as area fractions.

	Median CD20+ area fractions (absolute)					Median CD20+ area fractions (relative)			Median area fractions (regions)
	WS	No CD20+ n(%)	TMA	No CD20+ n(%)	p-value	WS	TMA	p-value	WS	TMA
Tumor epithelial islands (Region 1)	0.019%	5 (2%)	0.00080%	105 (48%)	<0.001	0.060%	0.0021%	<0.001	30%	32%
Tumor margin (Region 2)	0.10%	2 (1%)	0.015%	58 (26%)	<0.001	0.33%	0.049%	<0.001	35%	38%
Stroma (Region 3)	0.42%	0 (0%)	0.031%	49 (22%)	<0.001	0.66%	0.054%	<0.001	70%	68%
Total tumor area (Region 4)	0.44%	0 (0%)	0.037%	40 (18%)	<0.001

Absolute area fractions: area of CD20+ cells relative to total tumor area. Relative area fractions: area of CD20+ cells in sub-region relative to total area of sub-region. Area fraction regions: total area of sub-region relative to total area of total tumor area (region 4).

The majority of CD20+ cells were located in the stroma (WS: 0.42%, TMA: 0.031%), of which a fraction was located in the near margin around tumor epithelial islands (WS: 0.10%, TMA: 0.015%). CD20+ cells were detected in all WS, but 18% of the TMA cores did not contain any CD20+ cells, especially in the tumor epithelium (48% without CD20+ cells) (Table 1). The relative median area fractions of CD20+ were significantly lower in TMA cores, but in both WS and TMA, the relative median area fraction was at least 10-fold higher in stroma vs tumor epithelium (WS tumor epithelium: 0.060%, WS stroma: 0.66%; TMA tumor epithelium: 0.0021%, TMA stroma: 0.054%). The WS sections and TMA cores contained similar median areas of tumor epithelial cells vs stroma (WS tumor area: 30%, TMA tumor area: 32%) (Table 1).

We next examined whether CD20+ infiltration differed by ER status. Among the 221 included patients, 134 (61%) patients had ER-positive tumors and 87 (39%) had ER-negative tumors. When stratified by ER status, significantly higher CD20+ total area fractions were observed in ER-negative tumors compared to ER-positive tumors, both for WS (median 1.23% vs 0.19% in negative vs positive) and TMA (median 0.14% vs 0.010% in negative vs positive) (Figure 3).

Figure 3.

CD20+ area fractions on total tumor areas stratified by estrogen receptor (ER) status measured on (A) whole slides (WS) and (B) tissue microarrays (TMA). Number of samples without detectable CD20 expression in TMAs is shown at the bottom (size of the circle is not to scale).

Correlations and agreements between measurements of the same sample in WS and TMA are shown in Figure 4. In the total tumor area (region 4), the average absolute CD20+ area fractions were 6.4-fold higher in WS than in TMA (Figure 4H, Table 2). There was a tendency toward better agreements with higher CD20+ area fractions, but the limits of agreements were wide (0.088-fold to 468-fold). Similar correlations and agreements were observed for absolute CD20+ area fractions in tumor epithelium, stroma, and tumor margin (Figure 4, Table 2). The Bland-Altman plots further illustrated that most data points were positioned above the y = 1 line, indicating a predominance of higher values obtained from WS (Figure 4B, D, F, H). The 95% confidence intervals of the mean difference did not include y = 1, suggesting that the discrepancies observed between evaluation on TMA and WS are not the result of random error but attributable to a systematic bias and a tendency to measure higher CD20+ area fractions on WS than in TMA. This was further supported by correlation plots and Lin’s CCC < 0.50 for all regions (Figure 4A, C, E, G).

Figure 4.

Correlations and agreements (Bland-Altman plots) between CD20+ area fractions measured on whole slide sections (WS) and tissue microarrays (TMA), showing a systematic underestimation of CD20+ area fractions when using one centrally sampled TMA core compared to WS. Correlation plots (A, C, E, G) include Lin’s concordance correlation coefficients (CCC), ρ_C. Samples without CD20+ cells (shown in red) were excluded from statistical analyses due to undefined values under log-transformation, but were assigned a small constant for visualization purposes only and are not included in ρ_C. Bland-Altman plots (B, D, F, H) show ratios between CD20+ area fractions of WS relative to TMA against the geometric means of CD20+ area fractions. Average differences (mean ratio) and corresponding 95% confidence intervals (CI) are included (short dashed lines) as well as 95% limits of agreements (LoA) (long dashed lines). Correlation and agreement plots are for the absolute area fraction of CD20+ cells in the total tumor area (G, H), and for the absolute area fractions of CD20+ cells in tumor epithelial islands (A, B), tumor margin (C, D), and stroma (E, F) relative to the total tumor area.

Table 2.

Correlations and agreements between absolute CD20+ area fractions measured on whole slide (WS) and tissue microarrays (TMA).

	Lin’s CCCa (ρC)	Average difference		LoAc
		Mean ratiob w. 95% CI	p-value
Tumor epithelial islands (Region 1)	0.39	2.5 (1.7–3.6)	<0.001	0.047–130
Tumor margin (Region 2)	0.45	3.7 (2.7–5.0)	<0.001	0.083–165
Stroma (Region 3)	0.36	6.1 (4.4–8.4)	<0.001	0.087–424
Total tumor area (Region 4)	0.39	6.4 (4.7–8.9)	<0.001	0.088–468

^aLin’s concordance correlation coefficient.

^bRatio of WS vs TMA.

^cLoA: limits of agreement.

The lowest mean ratio CD20+ area fractions was observed in the region directly contacting tumor epithelium, indicating a superior agreement between TMA and WS in the tumor epithelial islands, as compared to the other regions (Table 1). In general, the majority of TIL-B was found in the stroma and lowest in direct contact with tumor epithelium, highlighted by a 10-fold higher relative median CD20+ area fraction in stroma vs tumor epithelium (Figure 4 and Table 1).

4. Discussion

This study addresses methodological challenges in assessing CD20 expression in BC when evaluating its potential as a clinical biomarker, especially with a focus on the use of TMAs. The present results showed that CD20+ area fractions were consistently higher when quantified in WS compared with a single central TMA core across all spatial tissue compartments (tumor, stroma, tumor margin, and total tumor area), with poor agreement between the two methods. The observed discrepancy between TMA and WS assessment of CD20+ cells likely reflects the spatial heterogeneity of TIL-B infiltration within breast tumors. In contrast to some immune cell populations that may be more diffusely distributed, TIL-B often occur in nodular aggregates or TLS, which are often located in the tumor periphery. Such localized immune cell clusters may therefore be missed when using TMAs, especially central cores as available for this study population. This biological distribution pattern likely contributes to the lower CD20+ area fractions observed in TMAs compared with WS in the present study. Moreover, this discrepancy may be further accentuated by the overall lower level of TIL-B in direct contact with tumor epithelium compared to the surrounding stroma.

These findings suggest that TMAs may systematically underestimate CD20+ infiltration, thereby compromising the accuracy and representativeness of TIL-B quantification. On the other hand, an overexpression of TIL-B could have been obtained if the TMA core were taken from an area including nodular aggregates or TLS. These findings highlight the importance of considering spatial localization and tissue sampling strategies when evaluating TIL-B related biomarkers in breast cancer.

We acknowledge that intratumoral heterogeneity may influence biomarker assessment when using single-core sampling; however, the results from these previous studies demonstrated that, despite intratumoral heterogeneity, central TMA cores provide overall representative estimates for a variety of biomarkers [36,37].

While performing immunohistochemical analysis on WS can be efficient and clinically relevant in routine pathological practice, it may be expensive and time-consuming in a research setting examining large patient cohorts. Therefore, the use of TMA has gained popularity as a more cost- and time-effective alternative in large sample biomarker studies, including studies investigating the role of immune cells. However, the International Immuno-Oncology Biomarker Working Group recommends evaluating TILs on WS, citing concerns about the representativeness of TMAs due to tumor heterogeneity [49]. Therefore, use of TMAs requires validation before being applied for reliable assessment of immune markers, given the inherent heterogeneity of cancer tissue and the tumor microenvironment (TME).

These results highlight the importance of using WS for reliable CD20 assessment in BC tissue, emphasizing that not all biomarkers may be applicable for TMA analysis. The size, number, and location of the cores required to represent WS can potentially be modified to overcome the challenges of tumor heterogeneity and may be biomarker-specific. In our study, the available TMA material included only one core of 1.00 mm from each patient and FFPE block. This contrasts with the few previous studies using TMA for TIL-B assessment, utilizing two to four cores from tumor-rich areas and with diameters ranging from 0.6 mm to 2.0 mm (reviewed in [50]).

Two previous studies have addressed the issue of TIL-B representativeness on TMA as compared to WS in BC. Both studies used manual counting of CD20+ cells, in contrast to our study using digital image analysis based on area fractions. Mahmoud et al. [12] compared CD20+ on TMA and corresponding WS in a pilot study involving samples from 30 BC patients using two 0.6 mm cores. Similar to our findings, they observed that CD20 assessed on WS exhibited both diffuse and aggregate patterns, with higher levels of CD20+ generally observed in the stroma rather than within the tumor epithelium. Nevertheless, they reported that the primary pattern was diffusely scattered TIL-B, with a less frequent presence of TIL-B aggregates. This contrasted with our findings, which revealed that TIL-B was commonly found in nodules/aggregates. Notably, the authors reported absolute cell counts rather than relative values as area fractions. This methodological difference may influence the interpretation of their results.

In another pilot study, Mohammed et al. [18] compared the expression of CD20+ and CD138+ (plasma-cell marker) cells evaluated on three 0.6 mm TMA cores and corresponding WS from 40 BC patients. They reported that TMAs were representative of WS when evaluating B-lymphocytes, as evidenced by interclass correlation coefficients of 0.80 for both markers. However, they only included a reduced proportion of the WS (three fields of 20x magnification) in their comparisons of mean CD20 values. This may account for their conclusion of a good agreement between the two methods, which is inconsistent with our findings that showed poor agreement and a statistically significant difference between the two methods.

In contrast to the previously discussed studies, the present study is limited by having only one core available from each FFPE block. These TMAs were originally constructed for broader biomarker studies within the DBCG82bc cohort and not specifically for CD20 evaluation. As such, only 1 mm core/patient was included, originally selected from central tumor areas with the presence of carcinoma, but without consideration of immune cell infiltration. This limitation raises the concern that a possible clinical value of TIL-B may be false-positive or false-negative, depending on the random inclusion or exclusion of regions with dense lymphoid aggregates or TLS in the core.

In our study, the primary objective was to evaluate the analytical validity of CD20, rather than to characterize structured immune features such as TLS, and information on TLS is therefore not available. However, Buisseret et al. [51] examined the spatial distribution of TIL-B and reported that the majority of TIL-B was located in TLS, often correlating with the presence of CD4+ T-follicular helper lymphocytes. Furthermore, TLS was identified in 60% of 125 BC tumors and was associated with higher levels of TILs and more aggressive BC subtypes.

In a study by Danenberg et al. [52], the spatial arrangement of specific cells in the TME was evaluated using imaging mass cytometry on samples from 693 BC patients. B-cells represented in aggregations were the most observed spatial distribution and were associated with distinct TME structures and were observed in both aggressive and less aggressive luminal tumors. The authors reported that ER status had the greatest influence on the TME. In our study, CD20+ total area fractions were stratified according to ER status to investigate the variations between the two groups. Consistent with prior findings, we discovered that higher median values of CD20+ area fractions were observed in ER-negative tumors, regardless of the assessment method (WS or TMA). These findings suggest that different underlying mechanisms contribute to B-cell aggregation and that the complex spatial interplay should be considered when evaluating cells in the tumor milieu.

The strength of the present study includes paired TMA and WS sections from each patient, the use of DIA and analysis of spatial compartments. Limitations include the use of one central 1-mm core from each FFPE block. The 221 patients included are not intended to be representative for the original DBCG82bc cohort in terms of clinicopathological variables or level of TILs, but were chosen to ensure a variety of samples with differing levels of TILs and deliberately intended to secure inclusion of samples with high levels of TILs. This selection strategy may have, however, introduced bias.

5. Conclusion

In conclusion, the present findings suggest that the presence of CD20+ cells in BC can be underestimated when using TMA, particularly with a single central core, presumably because they tend to form nodules, especially in the tumor periphery. This was supported by consistently higher CD20+ area fractions observed in WS compared with a single TMA core across all spatial compartments. Together, these findings highlight the impact of spatial heterogeneity and emphasize the importance of tissue sampling strategy when evaluating CD20+ TIL-B.

Funding Statement

This work was supported by the Kræftens Bekæmpelse [R146-A-9519 and R148-A10067-16-S46].

Article highlights

Introduction:

• Tumor-infiltrating B-lymphocytes (TIL-B) are emerging as important biomarkers in breast cancer.

• CD20 is widely used to assess B-cell infiltration.

• TIL-B exhibit spatial heterogeneity and often form nodular aggregates at the tumor periphery rendering the use of tissue microarrays (TMA) questionable.

Material and methods:

• In 221 breast cancer patients, CD20+ area fractions were compared between whole-slide sections (WS) and TMAs consisting of a single 1 mm central core per patient.

• TIL-B levels were assessed in four spatial compartments.

• Digital image analysis was applied to quantify CD20+ cells across spatial compartments.

Results:

• CD20+ area fractions were consistently higher in WS than in TMAs across all spatial compartments.

• Agreement between WS and TMA measurements was poor.

Conclusion:

• Single-core TMAs may underestimate CD20+ infiltration due to spatial heterogeneity.

Disclosure statement

Trine Tramm reports relationship with Pfizer and AstraZeneca that includes: speaking and lecture fees. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript, apart from those disclosed.

No AI-based tools and technologies were utilized in the production of this manuscript.

Writing disclosure

No writing assistance was utilized in the production of this manuscript.

Reviewer disclosure

Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Ethical conduct of research

The study was approved by The Central Denmark Region Committees on Health Research Ethics (journal no. 1-10-72-45-17), The Danish Data Protection Agency (journal no. 1-16-02-698-17), and the DBCG. The reporting of data follows the Recommendation for Tumor Marker Prognostic studies (REMARK) criteria, and was conducted in accordance with the Declaration of Helsinki. Verbal informed consent was mandatory from all participants prior to inclusion in the original randomized DBCG82bc trials. The DBCG82bc cohort has been described previously.

Data availability statement

The data are not publicly available due to information that could compromise the privacy of the patients included in this study.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/17520363.2026.2714196