Association of CD4⁺ Radiation-Induced Lymphocyte Apoptosis with Fibrosis and Telangiectasia after Radiotherapy in 272 Breast Cancer Patients with >10-Year Follow-up

Radiotherapy (RT), in combination with surgery, chemotherapy, and immunotherapy, is a fundamental part of modern multimodal antitumor therapy. More than 60% of all cancer patients treated with curative intent receive RT as part of their treatment (1). Earlier tumor detection and technological advances in treatment planning and delivery have resulted in an increasing number of long-term survivors of breast cancer (2, 3). Late normal-tissue reactions limit the dose of RT given to cancer patients and affect the quality of life of many patients. Thus identification of relatively radiosensitive or -resistant patients would enable improved personalized treatment, e.g., by modifying the dose, fractionation scheme, or offering an alternative treatment.

It has been estimated that up to 80% of the observed variation in risk of radiation-induced late reaction may be ascribed to factors associated with the individual patient (4). Over the last decades, various assays have been tested in search for predictive markers for individual risk that may be used to personalize RT for cancer patients. A number of these assays have tried to identify relevant genetic variants such as SNPs and, more recently, factors based on other “omics” including epigenetics, the transcriptome, proteome, kinome, and metabolome, thereby covering the entire spectrum from the genome to protein function (reviewed in ref. 5). Despite a large number of studies, only few significant genes have been identified, which is likely to be due to the involvement of multiple genes with small effect sizes in the development of RT-induced toxicities, thus requiring very large cohorts to achieve significance (6).

An alternative approach is to use functional assays which investigate the association between cellular endpoints (e.g., DNA damage and repair, clonogenic cell survival, or apoptosis) and the risk of developing clinical normal-tissue reactions (5). A promising functional assay is the radiation-induced lymphocyte apoptosis (RILA) assay on lymphocyte subpopulations from peripheral blood (PB) which can be performed with standard flow cytometry equipment (7). A prospective study of 399 patients with mixed malignancies showed an inverse correlation between RILA values for CD4⁺ or CD8⁺ T lymphocyte populations after in vitro irradiation with 8 Gy and the risk of any late toxicity scored by the Common Toxicity Criteria (CTC v2.0) 6 to 24 months after RT (median dose 66 Gy in 1.8-2.0 Gy/fx; ref. 8). The correlation with late reactions was further supported by a small study of 26 breast cancer patients (9), whereas two other studies (n = 59 and n = 16, respectively) failed to detect a significant correlation with breast appearance (10, 11). However, recently, a significant inverse correlation between RILA in CD8⁺ T cells and radiation-induced fibrosis was confirmed in a study of 456 breast cancer patients with up to 36 months of follow-up (12). Furthermore, similar correlations for CD8⁺ RILA have been found for late reactions in head and neck (n = 79) and cervical cancers (n = 94), as well as for genitourinary but not gastrointestinal late reaction in prostate patients (n = 198; refs. 13–15). Notably, most of the studies have used rather broad toxicity scores, often including several endpoints that may not be biologically related. Furthermore, although late reactions may continue to appear 5 to 10 years after RT, studies with long-term follow-up have not been published so far.

Because long-term follow-up data from prospective, clinical RILA studies may not be available soon, we validated the RILA assay in 272 PB samples from female early breast cancer patients who had been treated with breast conserving surgery (BCS) and RT and followed prospectively in the German ISE cohort study (16). The patients had a minimum follow-up of 10 years, and late reactions were scored according to the Late Effects in Normal Tissues-Subjective, Objective, Management and Analytic (LENT-SOMA) classification system (17) with well-defined criteria for the specific endpoints, subcutaneous fibrosis, and telangiectasia. Because the CD8⁺ lymphocyte population may include CD3⁻/CD8⁺ natural killer (NK) cells, three-color FACS was used to distinguish RILA of the CD4⁺, CD8⁺, and CD3⁻/CD8⁺ populations.

Four hundred and seventy-eight female patients with histologically confirmed early breast cancer or in situ carcinoma from the German ISE cohort (16) were treated with BCS during 1998–2001. The patients were recruited from the Rhine-Neckar region in Germany (Women's Clinic Heidelberg, St. Vincentius Clinic Karlsruhe, City Hospital Karlsruhe, and Universitätsmedizin Mannheim). In brief, breast cancer patients were eligible if they were treated unilaterally with RT (but without chemotherapy) after BCS. Adjuvant RT was given to the whole breast (50.0 Gy in 2.0 Gy/fraction or 50.4 Gy in 1.8 Gy/fraction) followed by a tumor bed boost of 6 to 16 Gy (median: 10 Gy, 2.0 Gy/fx; n = 224), or 56.0 Gy given in 2.0 Gy fractions without a boost (n = 28). Because of the higher dose to the whole breast in patients without boost, the difference in total dose was modest and did not result in a significant difference in fibrosis (P = 0.85) nor telangiectasia (P = 0.65). The equivalent dose in 2 Gy fractions (EQD2; calculated using the linear-quadratic model with α/β ratios of 2 Gy for fibrosis and 3 Gy for telangiectasia; ref. 18) was 60 Gy (47.9 Gy–70.0 Gy) for fibrosis and 60 Gy (48.4 Gy–70.0 Gy) for telangiectasia. Among the 409 patients alive at 10 years of follow-up, a total of 272 patients agreed to participate in the present study. Twenty patients were excluded: 6 due to mastectomy, contralateral breast cancer, or additional RT, 2 due to receiving high-dose brachytherapy, and 12 patients due to technical failing of the RILA assay. Thus, a total of 252 patients with a median follow-up of 11.6 years (range, 10.3–12.8 years) were available for data analysis (Table 1; Fig. 1). Subcutaneous fibrosis and telangiectasia of the irradiated breast were assessed by a single-study physician using the LENT SOMA scale (17). Furthermore, the location of fibrosis within or outside the surgical area was scored. Maximum early toxicity was scored dichotomously (grade, 0–2b vs. 2c and higher) using a modified CTCAE v2.0 scale, as described previously (16). Information on demographic factors, medical history, and lifestyle factors was gathered by means of self-reported questionnaires. The study was approved by the local ethics committee and was conducted according to Declaration of Helsinki principles. Written-informed consent was obtained from each patient.

At the last follow-up after a minimum of 10 years, 5 mL of PB was collected from each patient in heparin-coated syringes (Sarstedt) and stored overnight at room temperature to eliminate the potential influence of differences in storage times for the samples taken during the day and delivered to the lab in the evening. The next morning, PB was diluted 1:10 in DMEM supplemented with 20% FBS (Biochrom) and irradiated with 8 Gy of 6 MV X-rays or sham irradiated for baseline values (0 Gy). After irradiation, samples were incubated for 48 hours in a humidified cell culture incubator at 37°C under 5% CO₂. After incubation, the cells were harvested and 2 mL used for each staining. Samples were stained with CD3/CD4-FITC, CD8-PE, and appropriate IgG controls for FITC and PE (all BD-Pharmingen) for 15 minutes at room temperature and then red blood cells were lysed and fixed (FACS lysing solution, BD), according to the manufacturer's instructions. After washing with PBS, samples were resuspended and incubated for 10 minutes at room temperature in PBS supplemented with RNAse (100 μg/mL; Serva) and the DNA dye 7-Aminoactinomycin (7-AAD; 4 μg/mL; Thermo Fisher). After staining, cells were acquired on a FACSCanto II (BD) and analyzed using the FlowJo 7.6.4 software package (Tree Star). A live gate was applied during measurement to exclude granulocytes, as these are not of relevance for the assay and may be mistaken for CD4⁺ or CD8⁺ lymphocytes owing to their strong autofluorescence. A total of 20,000 live-gated events were acquired per sample. For each of the respective lymphocyte subpopulations (CD4, CD8, CD3⁻/CD8⁺), the 7-AAD⁺ population was gated and plotted as 7-AAD versus FSC. 7-AAD^dull-FCS^low cells were gated and defined as apoptotic, as these have a less than 2n DNA content and a reduced cell size (apoptotic phenotype; procedure modified from Menz and colleagues; ref. 19). RILA was defined as the percent apoptosis after 8 Gy minus the percent apoptosis at 0 Gy.

All in vitro irradiations were performed with 6 MV X-rays from a clinical linear accelerator (Versa HD, Elekta Synergy) at a dose rate of 6.67 Gy/min using a 40 × 40 cm² irradiation field. The samples were irradiated in 25 mL cell-culture flasks (Falcon, Corning B.V. Life Sciences) at 100 cm source-surface distance with 15 mm water-equivalent material for dose buildup and 8 cm for backscatter.

Patient samples were blinded to avoid bias and were unblinded for data analysis after completing the RILA measurements. Data are presented as median and range, unless otherwise noted. To determine the relationship between two parametric variables, univariate linear regression was performed. The nonparametric Wilcoxon/Kruskal–Wallis test was used for comparison of two groups and the Spearman rho test for correlations with toxicity grade. ROC curves were created using RILA as a univariate random variable (classifier) and quantified by the AUC. For the determination of other predictive parameters [sensitivity, specificity, positive-predictive value (PPV), and negative-predictive value (NPV)], the RILA values were split into tertiles as proposed previously (8) and the middle and upper tertiles combined versus the lower tertile to obtain a binary variable for univariate analysis of the dichotomized toxicity (fibrosis/telangiectasia grade 2+3 vs. 0+1) similar to a recent study (12). It was verified that the proportions of patients with fibrosis in the middle and upper RILA tertiles were not significantly different in univariate analysis (P = 0.08, Fisher exact test), whereas the proportions with fibrosis were significantly larger in the lower tertile compared with the middle (P < 0.0001) and the upper tertiles (P = 0.014). Furthermore, multivariate analysis was performed adjusting for established and putative clinical covariables (12, 16). The final multivariate model included: Age at surgery (years, continuous), body mass index (BMI, in kg/m², continuous), hypertension (yes vs. no), smoking status at 10-year follow-up (ever vs. never), and the EQD2 (18). The model on fibrosis was additionally adjusted for hormonal treatment (yes vs. no; according to ref. 12). It was confirmed that the middle tertile was not associated with an increased risk (OR) when compared with the upper tertile as reference, in contrast with the lower tertile (Supplementary Table S1). Although a lower OR was observed for the middle tertile, it was only marginally significant [OR, 0.43 (0.17–0.97); P = 0.04] and may be coincidental because no mechanism is known that would predict this finding and the significance would be lost after correcting for multiple testing. Univariate statistical analyses were performed using JMP11 statistical software (SAS Institute). Graphs were plotted using either JMP11 or SigmaPlot 11.0 (Systat Software GmbH). Multivariable logistic regression analysis and bootstrapping analysis were performed using SAS 9.4 software.

As surgery may affect the development of fibrosis, and the influence of wound healing and scar formation can be assumed to be minimal outside the surgical area, we also assessed fibrosis according to location within the surgical area only (group 1) or outside (group 2) using multinomial logistic regression. Patients without subcutaneous fibrosis ≥grade 2 were used as controls (group 0). Adjustment variables were the same as in the binary logistic regression models with overall fibrosis. To assess robustness of the multivariate model, a bootstrapping analysis was performed based on 1,000 replications. The reported c statistic was corrected for optimism (18).

In order to distinguish NK and different T-cell populations, triple-color FACS using 7-AAD for DNA staining was used instead of two-color FACS with propidium iodide (PI) as used in the original assay (8, 19). To verify that this did not influence the RILA values, pilot experiments with both fluorescent dyes were performed. A very strong correlation (r² = 0.953; P < 0.001) between the PI and 7-AAD RILA was observed for all three T-cell populations (Supplementary Fig. S1) confirming the validity of replacing PI with 7-AAD.

The distribution of RILA values for each lymphocyte population (CD4⁺, CD8⁺, and CD3⁻/CD8⁺) showed the largest range and highest median rate of radiation-induced apoptosis for NK cells (CD3⁻/8⁺), followed by CD8⁺ and CD4⁺ T cells (Fig. 2A; Supplementary Table S2). However, NK cells were much less abundant and more radiosensitive than CD8⁺ and CD4⁺. Thus based on 20,000 events, the median cell numbers were 419 (0 Gy) and 38 (8 Gy) for CD3⁻/CD8⁺, 2,441 (0 Gy) and 1,036 (8 Gy) for CD8⁺, and 4,222 (0 Gy) and 3,373 (8 Gy) for CD4⁺. RILA for the CD4⁺ and CD8⁺ lymphocyte populations revealed a significant correlation (r² = 0.603, P < 0.001; Supplementary Fig. S2), whereas only weak though significant correlations were found between CD4⁺ or CD8⁺ and the CD3⁻/8⁺ RILA values (r² = 0.082, P < 0.001 and r² = 0.116, P < 0.001, respectively). Notably, CD8⁺ RILA values appear to be “capped” at 60% to 65%, whereas this is not the case for CD4⁺ RILA. Thus, CD4 provides better discrimination than CD8 at high RILA values, whereas the variation between the two values dominates at low RILA values. As RILA was previously found to decrease with age (20), this was tested in the present cohort. No significant correlation with patient age was found in the range of 50.6 to 91.1 years (median, 70.7 years) for CD4⁺ [0.026% per year; 95% confidence interval (CI), −0.102%, 0.154%; r² < 0.001, P = 0.694; Fig. 2B] and CD3⁻/CD8⁺ RILA (0.031% per year, 95% CI, −0.319%, 0.382%; r² < 0.001, P = 0.860; Fig. 2C). By contrast, a very weak but significant increase of CD8⁺ RILA with age (0.267% per year; 95% CI, 0.034%–0.499%; r² = 0.016, P = 0.025) was observed (Fig. 2D). Thus, a decrease in RILA with increasing age was not observed in the age range (50.6–91.1 years) of this cohort.

Ten years after RT, 30.6% of the patients had developed subcutaneous fibrosis grade ≥2 and 9.5% telangiectasia grade ≥2, with 6.7% developing both (Table 1). Radiation-induced maximum early effects (grade ≥ 2c) were observed in 19.8% of the patients. RILA values were significantly lower in patients with RT-induced fibrosis compared with unaffected patients for CD4⁺ cells (P = 0.011) but not for CD8⁺ or CD3⁻/CD8⁺ cells (Fig. 3A). The difference in CD4⁺ RILA was even more pronounced for telangiectasia (P < 0.001) and showed a similar trend for CD8⁺ RILA (P = 0.060; Fig. 3B). The distribution of CD4⁺ RILA values seemed to be more skewed toward lower values, especially for telangiectasia. No association between basal apoptosis for any of the lymphocyte populations and the onset of fibrosis (p_CD4 = 0.84; p_CD8 = 0.67; p_NK = 0.78) or telangiectasia (p_CD4 = 0.56; p_CD8 = 0.79; p_NK = 0.61) was observed.

The correlation between the grade of late reactions and RILA values corroborated the association between the CD4⁺ RILA and the risk of developing fibrosis or telangiectasia (rho_fibr: −0.171, P = 0.008; rho_tel: −0.205, P = 0.002; Supplementary Fig. S3A). For CD8⁺ lymphocytes, no significant correlation of RILA was detected with grade of fibrosis (rho: −0.006; P = 0.920) nor telangiectasia (rho: −0.122; P = 0.053; Supplementary Fig. S3B). In addition, no significant association between RILA values and maximum early RT-induced toxicity (2c and higher vs. grade 0–2b) was observed for any of the lymphocyte populations (P > 0.6; Supplementary Fig. S4).

In multivariate analysis of RILA, including relevant clinical risk factors, the lower CD4⁺ RILA tertile was significantly associated with an increased OR for both fibrosis (OR, 3.48; 95% CI, 1.84–6.58) and telangiectasia (OR, 8.60; 95% CI, 2.71–27.3, Table 2A) compared with the middle/upper tertiles, whereas no association was observed for CD8⁺ RILA, neither with fibrosis nor with telangiectasia.

A subanalysis according to location of fibrosis with respect to the surgical area was performed, based on the assumption that fibrosis outside the surgical area was more likely to be related to RT with minimal influence of wound healing and scar formation after surgery (Supplementary Table S3). Using the nonparametric Wilcoxon/Kruskal–Wallis test, the distribution of CD4⁺ RILA values was not correlated with fibrosis limited to the surgical area (group 1, P = 0.17, n = 54) but significantly associated with fibrosis outside the surgical area (group 2, P = 0.004, n = 23; Supplementary Fig. S5). No significant differences were observed for CD8⁺ RILA (p_{within only} = 0.938; p_outside = 0.478). In multivariate logistic regression analysis, fibrosis outside the surgical area (Table 2B, n = 23) revealed higher ORs (4.33, 95% CI, 1.60–11.7) for low CD4⁺ RILA compared with overall fibrosis (Table 2A, OR = 3.48) or fibrosis limited to the surgical area (Table 2B, OR = 3.08). However, the difference in OR between fibrosis outside versus within the surgical area was not statistically significant (P = 0.5).

Predictive univariate ROC analysis yielded an AUC of 0.604 for predicting fibrosis from CD4⁺ RILA and 0.736 for telangiectasia, whereas the values for CD8⁺ were lower (0.511 and 0.617, respectively; Table 3). Notably, prediction of fibrosis outside the surgical area from CD4⁺ RILA was markedly better with AUC increased to 0.687. Comparing the lower tertile versus the upper two tertiles yielded 77% sensitivity and 71% specificity for CD4⁺ RILA prediction of telangiectasia ≥ grade 2 in the present cohort (Table 3). However, the NPV (97%) was much higher than the PPV (22%). Values for predictive modeling of fibrosis and for CD8⁺ RILA as a potential predictor are shown in Table 3. The NPV was increased (79%→93%) at the expense of the PPV (48%→22%) for CD4⁺ RILA and fibrosis outside the surgical area compared with any fibrosis (Table 3). With respect to the severity of late reaction, for CD4⁺ RILA, 53% (38/72) of patients with fibrosis (grade ≥ 2) and 69% (9/13) of the grade 3 cases were in the lower tertile of the samples. Stratifying by the location of ≥ grade 2 subcutaneous fibrosis yielded 51% (25/49) of all patients with fibrosis within and 61% (14/23) of all patients with fibrosis outside the surgical area in the lower tertile. For grade 3, this was observed in 60% (3/5) of the patients with fibrosis within and 100% (4/4) of the patients with fibrosis outside the surgical area. For telangiectasia ≥grade 2, the fractions of patients in the lower tertile were 77% (17/22) and 100% (7/7) for grade 3.

To estimate the predictive value of RILA in the presence of clinical risk factors, bootstrapping was performed. The results showed a moderate predictive value of CD4⁺ based on the corrected C statistics, which is equivalent to the AUC in an ROC analysis (Table 3), including CD4⁺ in the model improved discrimination (c statistics) from 0.62 to 0.68 for any breast fibrosis, from 0.62 to 0.66 for fibrosis within the surgical area only, from 0.61 to 0.69 for fibrosis outside the surgical area, and from 0.65 to 0.76 for telangiectasia. Addition of CD8⁺ RILA to the model did not improve prediction of fibrosis nor telangiectasia when adjusted for clinical risk factors (Table 3), implying that CD4⁺ RILA may be a more robust predictor of radiation-induced late reaction at 10 years of follow-up than CD8⁺ RILA.

In the present study, we tested the association between RILA and RT-induced late toxicities in 272 early breast cancer patients with long-term follow-up after RT (range, 10.3–12.8 years) in different lymphocyte subpopulations. RILA in CD4⁺ (T-helper) cells was strongly associated with telangiectasia after RT and, to a lesser extent, with subcutaneous fibrosis. The association was somewhat stronger for fibrosis outside the surgical area where the confounding influence of wound healing and scar formation was likely to be small. By contrast, RILA in the cytotoxic T-cell population (CD8⁺) did not significantly correlate with fibrosis or telangiectasia. In addition, there was no association of RILA for the highly radiosensitive NK-cell (CD3⁻/8⁺) population with late toxicity or between RILA for any of the three lymphocyte populations and early toxicity after RT. Multivariate analysis for CD4⁺ RILA accounting for clinical risk factors showed significantly increased ORs for both fibrosis and telangiectasia when comparing the lower tertile with the upper two RILA tertiles.

To our knowledge, this is the first study showing a strong association of T-cell RILA with telangiectasia as well as an association of CD4⁺ RILA with fibrosis in a large breast cancer patient cohort. The observation that all patients with telangiectasia or severe fibrosis outside the surgical area had CD4⁺ RILA values in the lower tertile (<7.9%) corroborates the association. Barber and colleagues (21) investigated the rate of apoptosis (using the terminal deoxynucleotidyl transferase–mediated dUTP nick end labeling assay) in total PB lymphocytes from 31 breast cancer patients, but did not observe any correlation between a reduction in apoptosis and breast fibrosis or telangiectasia 8 to 13 years after RT. In a different study, a correlation with reduced late reaction (mainly breast shrinkage and subcutaneous fibrosis) at long follow-up (mean 16.4 years) was found in a small cohort (n = 26) for a predictor combining high radiation-induced apoptosis and low double-strand break induction (9). It should be noted that in these studies, no labeling of lymphocyte subpopulations was performed, and apoptosis was measured by different methods which may have contributed to the conflicting results. Two further studies on breast cancer patients did not find a significant correlation with breast appearance after RT at median follow-up of 12.5 years (n = 59) and 5 years (n = 16), respectively (10, 11). However, the larger of these studies used lymphocytes which had been stored frozen as viable cells (10, 11), whereas all other studies have used fresh lymphocytes.

A prospective study by Ozsahin and colleagues found significant inverse correlations of CD4⁺ and CD8⁺ RILA with any moderate to severe late toxicity scored by CTCAE v2.0 in a large diverse patient cohort including 149 breast cancer patients (8). Based on a better prediction by CD8⁺ compared with CD4⁺ RILA in this study, Azria and colleagues recently conducted a prospective multicenter study on the association of CD8⁺ RILA with subcutaneous fibrosis (but not telangiectasia) in a cohort of 456 breast cancer patients (12). RILA was significantly lower (median 10.9%) for patients with moderate to severe breast fibrosis (CTCAE 3.0) within 3 years than for patients without reactions (median 15.6%; P < 0.001), thereby confirming the previous association of CD8⁺ RILA with broad late end points (8) for a specific end point in breast cancer patients. ROC analysis yielded an AUC of 0.62 (95% CI, 0.54–0.70). The present study, unexpectedly, found no correlation between CD8⁺ RILA and any fibrosis (AUC = 0.51), whereas univariate predictive modeling for CD4⁺ RILA yielded an AUC of 0.60. This AUC value, as well as the predictive parameter based on CD4⁺ RILA lower versus middle and upper tertiles, was comparable with the corresponding values for CD8⁺ in the study by Azria and colleagues (12), with the exception of the PPV, which was higher in our study (NPV was not shown in ref. 12). In the latter article, predictive parameters were also determined for the lower plus middle versus the upper CD8 RILA tertiles. Performing this analysis on our RILA data (see Supplementary Table S4) revealed a similar picture for both CD4⁺ and CD8⁺ RILA: whereas the sensitivity was slightly higher in the previous study (80% vs. 69% and 66.2%, respectively), the specificity was identical (34% vs. 35% and 33.1, respectively) and the NPV was higher (91% vs. 72% and 69%, respectively; PPV was not shown in ref. 12). Interestingly, confining the location of the fibrosis to outside the surgical area strengthened the association with CD4⁺ RILA (Supplementary Fig. S5) and adding CD4⁺ RILA improved the AUC to 0.687 (compared with 0.604 of any fibrosis; Table 3) despite the lower number of cases. The results from the multivariate analysis showed the largest improvement in c statistics when CD4⁺ RILA was included in the model for fibrosis outside (from 0.605 to 0.694) compared with any fibrosis (from 0.623 to 0.675) or fibrosis within the surgical scar area (from 0.621 to 0.661; Table 3). This suggests that wound healing is a confounding factor for fibrosis and that the risk of developing fibrosis within the scar area is determined mainly by clinical factors.

It is not clear why CD4⁺ RILA correlated better with normal-tissue reaction than CD8⁺ RILA in the present study with long follow-up. Although a correlation between CD4⁺ and CD8⁺ was observed, this was not strictly linear showing large relative variation at low values with decreasing correlation at high values. CD8⁺ RILA values in the present study were approximately 10 percent points higher (lower tertile < 22.0%, middle 22.0%–34.9%, upper ≥ 34.9%) than in the study by Azria and colleagues (<12%, 12%–24%, and ≥24%; ref. 12) but appeared to be “capped” at 60% to 65%. The different values for separating the tertiles in different studies may be related to minor differences in sample processing and technical aspects (RILA protocol and timing, serum used in the medium, gating strategy, patient malignancy, etc.). These factors are currently being investigated as part of an international multicenter study funded by the FP7 program (REQUITE; https://requite.eu/). The age of the patients at the time of surgery was similar in the present (median 59 years; range 39–79) and in the other study (median 56 years; range 29–88), but the present blood samples were taken 10 to 13 years after treatment. Although CD8⁺ RILA increased with age, the rate of change was too small (0.27% per year; upper 95% confidence limit: 0.50% per year) to explain the higher absolute values compared with the other study (12). Furthermore, CD8⁺ and CD4⁺ RILA values have previously been reported to decrease rather than increase with age (20), and RILA values similar to the CD8⁺ RILA in the present study have been published for unlabeled lymphocytes in a prospective study of breast cancer patients (9) with similar age distribution as in the study by Azria and colleagues (12). In addition, the present CD8⁺ RILA values still showed a wide range from 5% to 70%–75%. Therefore, a systematic change or loss of variation in CD8⁺ RILA with age does not explain the lack of association between late reaction and CD8⁺ RILA. Although differences in the FACS analysis may contribute to the difference in absolute values, it cannot be excluded that CD8⁺ RILA may change over time due to factors not related with the treatment and that CD4⁺ RILA may be more robust than CD8⁺ RILA.

Late reaction may continue to develop for several years after RT (22, 23). In the present study, 30.6% of the patients had developed ≥grade 2 subcutaneous fibrosis at 11.6-year median follow-up (range 10.3–12.7), whereas in the study by Azria and colleagues (12), 14.7% (64 of 434 patients with 3-year follow-up) had developed breast fibrosis up to 3 years (peak toxicity scored between 12 weeks to 36 months after RT). Although differences in treatments and clinical scoring (e.g., prevalence of tumor-bed boost; dose inhomogeneity; LENT-SOMA vs. CTCAE v3.0; different physicians) may play a role, the increased incidence at long-term follow-up is likely to be predominantly due to the continued risk of fibrosis (22, 23). If CD8⁺ RILA was predictive for breast fibrosis developing up to 3 years but not events developing later, this might lead to loss of the correlation with CD8⁺ RILA at long follow-up. A prospective study on 198 prostate cancer patients finding an association between CD4⁺ but not CD8⁺ RILA and genitourinary toxicity at 5.3-year median follow-up (15) might support this interpretation. A recent study on a breast cancer cohort (n = 105) with 6 years of median follow-up recorded only 5 cases of fibrosis as no boost was given (24, 25). Thus, although an association with CD8⁺ RILA was reported, the published data, and the difference in RT treatment, did not allow a more detailed comparison of RILA distributions.

An intriguing observation in the present study is that telangiectasia correlated more strongly with the RILA assay than fibrosis, suggesting that RILA may be a better predictor for late vascular effects than for fibrosis. This might potentially help explain the loss of correlation with CD8⁺ RILA with increasing follow-up time. Thus, if fibrotic events scored at early time points have a larger vascular component (e.g., edema) than at later time points, and if early vascular effects correlate with low CD8⁺ RILA, this could provide a mechanism for a change in predictive value of CD8⁺ as breast edema decreases and more breast fibrosis develops with time. To test this hypothesis, prospective studies with long-term follow-up, scoring both endpoints at regular intervals, will be required.

The association between low CD8⁺ RILA and increased risk of late reaction has been proposed to be caused by a slow apoptotic response and enhanced production of cytokines that attract inflammatory cells into the tissue (8, 26). We speculate that a low propensity for CD4⁺ apoptosis may favor polarization of Th cells toward an anti-inflammatory and fibrogenic response (Th₂ secreting IL4, IL5, IL13; regulatory T cells, T_reg, secreting IL10, TGFβ1). Thus, increased CD4⁺ apoptosis after RT might result in a less anti-inflammatory and fibrogenic microenvironment conferring resistance in a subgroup of patients consistent with the hypothesis of multiple subgroups characterized by different mechanisms of late reaction (5, 27). Indeed, the two mechanisms are not mutually exclusive because the recovery of normal tissue from radiation injury resembles the complex dynamics of wound-healing processes starting with an inflammatory response followed by an anti-inflammatory phase and later tissue remodeling. During the final preparation of this article, Florian Wirsdörfer and Verena Jendrossek [Institute of Cell Biology (Cancer Research), University Hospital Essen, 45122, Essen, Germany] reported that in a murine model of radiation-induced pneumopathy, CD4⁺ T lymphocytes are altered and contribute to profibrotic signaling and lung fibrosis [Oral presentation and personal communication at the Annual Conference of the German Association of Radiation Biology (Gesellschaft für Biologische Strahlenforschung) 2018, Frankfurt am Main, Germany]. These results corroborate our findings that low CD4⁺ RILA values are associated with enhanced risk for RT-related adverse late effects of the skin and support the hypothesis that a low propensity for apoptosis of CD4⁺ cells contributes to a fibrogenic response. An improved understanding of the role of immune cells and cytokines is likely to open new strategies for mitigating fibrosis (28) and possibly other adverse effects.

Although NK-cell RILA was included in an early study to determine the most suitable lymphocyte population for the RILA assay (7), the present study is the first to investigate its association with RT-induced toxicity in breast cancer patients. However, irradiation with 8 Gy almost completely eradicated NK cells making this population unsuitable for prediction at this dose level. This is consistent with a study on cervical cancer patients (13) but contrasts with a study on RT for prostate cancer (29). Both studies suggested that B cells might potentially be relevant. Thus, different lymphocyte populations, including NK cells at different dose levels, should be included in future studies and analyzed in relation to specific late endpoints.

The main limitation of the present study is that it is retrospective, although patients were recruited from the prospective ISE cohort. Thus, it is not possible to determine the true predictive values of the RILA assay. In particular, it is not possible to clarify if CD8⁺ RILA would have been predictive for fibrosis scored at earlier follow-up. On the other hand, the RILA assay is expected to measure constitutional apoptotic propensity, and no prospective studies with similar long-term follow-up have been performed so far. All patients were followed for more than 10 years irrespective of when the adverse event occurred, and the rate of late reaction was twice that of studies limited to 2 to 3 years of follow-up. Nevertheless, the present study supports the previous findings that RILA values determined on patients' PB lymphocytes are associated with the individual risk of developing late reaction after RT. However, it raises a number of important questions: (1) Do the predictive values of CD4⁺ and CD8⁺ RILA change with follow-up time? (2) Do any such changes depend on the specific late endpoint? (3) Is the predictive value related to a role of the immune system in late reaction? And, if so, (4) Can late reaction be mitigated by cytokines or antibodies against specific cell types during or after RT? Answering these questions will require further prospective and mechanistic studies. Although some will be addressed in the ongoing REQUITE study, the results and questions from the present study emphasize the need for very long follow-up in prospective studies.

In conclusion, a significant association between RILA of CD4⁺ T-cell populations and two specific RT-related late reactions (breast fibrosis and telangiectasia) in breast cancer patients after 10-year follow-up was shown for the first time. The association with telangiectasia was novel and was stronger than for fibrosis. For fibrosis, the association tended to be stronger outside the surgical area than within. Predictive modeling supported the use of the RILA functional assay for personalized RT and suggested that CD4⁺ RILA may be a more robust predictor over time than CD8⁺ RILA. Finally, the present study raises important questions regarding the relevant lymphocyte population and mechanistic aspects which should be addressed in future prospective studies.

E. Sperk reports receiving speakers bureau honoraria and commercial research support from Zeiss Meditec. F.A. Giordano reports receiving speakers bureau honoraria from Carl Zeiss Meditec AG, MSD Sharp & Dohme, NOXXON Pharma AG, and Roche Pharma AG, is a consultant/advisory board member for Carl Zeiss Meditec AG, Guebet SA, and NOXXON Pharma AG, and reports receiving commercial research grants from Carl Zeiss Meditec, ELEKTA AB, Guerbet SA, humediQ, and NOXXON Pharma AG. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M.R. Veldwijk, F. Wenz, J. Chang-Claude, C. Herskind

Development of methodology: M.R. Veldwijk, A. Botma, P. Seibold, S. Behrens

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.R. Veldwijk, P. Seibold, A. Botma, I. Helmbold, E. Sperk, N. Gürth, A. Kirchner

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.R. Veldwijk, P. Seibold, A. Botma, I. Helmbold, S. Behrens, F. Wenz, J. Chang-Claude, C. Herskind

Writing, review, and/or revision of the manuscript: M.R. Veldwijk, P. Seibold, A. Botma, I. Helmbold, E. Sperk, F.A. Giordano, N. Gürth, A. Kirchner, S. Behrens, F. Wenz, J. Chang-Claude, C. Herskind

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): I. Helmbold, F.A. Giordano, N. Gürth, A. Kirchner

Study supervision: M.R. Veldwijk, F. Wenz, J. Chang-Claude, C. Herskind

We thank the clinical members of the Department of Radiation Oncology (UMM, Mannheim) for their support, as well as C. Weiss, M.Sc. (Biometry and Statistics, Medical Faculty Mannheim, Ruprecht-Karls University Heidelberg, Mannheim, Germany) for help with the normalization of data, U. Eilber and C. Krieg (both DKFZ) for excellent data management, and Dr. A. Hüsing (DKFZ) for providing a SAS macro for the bootstrapping analysis. This study was supported in part by a grant of the “Dietmar Hopp Stiftung” (Grant number 23017006).

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