Predictive Value of Tumor Inltrating Lymphocytes in Neoadjuvant Chemotherapy Treated Breast Cancer: A Meta-Analysis of 8052 Patients

Background: We conducted a meta-analysis to determine the prognostic value of Tumor inltrating lymphocytes (TILs) for patients with breast cancer on Neoadjuvant Chemotherapy, to explore the prognostic value of different TILs threshold in terms of pathological complete response (PCR). Methods: A systematic search of PubMed, EMBASE and Web of Science electronic databases was conducted to identify eligible articles published before September 2020. Data from studies were analyzed by using Review Manager 5.3 and Stata 15.0 Results: A total of 22 published studies (including 8 052 patients) were eligible. Patients with high TILs level showed a higher rate of PCR to treatment (OR=3.182, 95 %CI, 2.549-3.973) compared to breast cancer patients with low TILs level. Although the association of TILs with response to neoadjuvant chemotherapy was similar across most breast cancer subtypes, there were a few differences ER negative or ER positive breast cancer. In studies (Type of breast cancer not clearly classied in the literature) where the cut-off value for TILs was ≥ 10 %, higher levels of total TILs predicted a higher PCR rate of Neoadjuvant Chemotherapy. However, for HER2-positive breast cancer patients, when a cut-off valve of TILs ≥ 30 % was used, the OR was 2.631 (95 % CI, 1.739-3.982, P = 0.000). TILs also were related to better DFS (HR=0.95, 95 %CI, 0.92-0.98, P=0.000) and overall survival (OS) (HR=0.90, 95 %CI, 0.85-0.95, P<0.0001) after Neoadjuvant Chemotherapy. Conclusions: TILs can be used as predictors of patients with breast cancer on Neoadjuvant Chemotherapy. TILs threshold with the greatest prognostic signicance of PCR is as yet unknown, but a TILs threshold of at least 30 % is associated with the most powerful outcome prognostication of PCR. incomplete outcome data; and selective reporting and other bias. These domains categorized having a high, low, or unclear risk of bias. qualied case-control studies assessed the Newcastle-Ottawa scale (NOS) [27]. contains eight comparability, and (case control studies). The quality to and studies as high quality. there any disagreement in the results of the literature quality it resolved by consensus, and a third author (Yang) intervene if necessary.


Introduction
Breast cancer is one of the malignant tumors to threaten women's health around the world. In 2018, an estimated 2.1 million women were newly diagnosed with breast cancer in the world [1], 626,679 women with breast cancer died. Neoadjuvant chemotherapy (NACT) has been demonstrated to be of standard therapeutic strategy in locally advanced and inoperable breast cancer. It can convert a previous locally advanced and inoperable breast tumor into an operable tumor [2][3], and in largely operable tumors, downstaging results in a small increase (7%-12%) in breast conservation rates [4][5][6]. The MD Anderson Cancer Center and the International Breast Group de ned unanimously Pathological complete response (PCR) as the complete disappearance of invasive cancer cells in primary breast tumor [7][8], studies have shown that PCR is an important predictor of prognosis for evaluating breast cancer patients with NACT. Patients who achieve PCR have favorable prognosis and longer survival period [9][10][11]. However, a small group of patients with NACT achieved PCR still relapse or die, whereas some patients without PCR have good prognosis. Thus, it is essential to nd more biomarkers to evaluate the e cacy and prognosis of breast cancer patients for NACT.
Tumor in ltrating lymphocytes (TILs) are one of the signi cant components in tumor microenvironment, which re ect the intensity of the immune response within the tumor bed [12][13][14]. Studies have shown that TILs can predict a favorable outcome for neoadjuvant chemotherapy in breast cancer, but there is still controversy. A meta-analysis reported TILs cannot predict hormone receptor-negative breast cancer patients [15], but there were also reports of the opposite results [16]. High levels of speci c phenotypes of TILs (CD8 + TIL, CD4 + TIL and Foxp3 + ), were associated with a better prognosis of breast cancer patient after NACT [17][18][19], while a review shows a worse prognosis [15]. Studies have suggested [20][21] that compared with the lower density of intratumoral tumor in ltrating lymphocytes (iTIL), stromal Tumor in ltrating lymphocytes (sTIL) was the predominant location of TILs in breast cancer, which are most suitable as a biomarker. Even so, considering the International TILs Working Group recommendations, no formal recommendation for a clinically cut-off value of TILs was established up to now [22]. Previous researches agreed that cut-off value more than 50% TILs were de ned as higher TILs, that is, lymphocyte predominant breast cancer (LPBC) [23]. However, it is still controversial to de ne the cut-off value of LPBC as 50% or 60% because of lower proportion of LPBC in clinical practice.
To address these controversies, we conducted a meta-analysis aimed to evaluate TILs, location or subtype of TILs as a potential prognostic marker for patients with breast cancer on Neoadjuvant Therapy, to explore the optimal thresholds and to determine the relationship between TILs and several clinicopathological features.

Literature search
Eligible studies were identi ed by a systematic literature search of the PubMed, EMBASE and Web of Science databases, with date restriction up to August 2020. The search strategy was carried out using the following keywords: 'breast cancer', 'tumor in ltrating lymphocytes', 'tumor associated lymphocytes', 'CD3-positiveT-lymphocytes', 'CD4-positiveT-lymphocytes', 'CD8-positiveT-lymphocytes', 'Foxp3-positiveT-lymphocytes' and 'neoadjuvant'. Independently searched by two authors (Li and Zhang), and any discrepancies were solved by discussion with a third author (Yang).

Inclusion and exclusion criteria
The studies included in the meta-analysis were either randomized controlled studies (RCTs) or case-control studies that evaluated the association between TILs and NACT for breast cancer. The inclusion criteria were as follows: (a) Study population or study subgroup consisted of breast cancer patient for NACT.
(b) Investigated the predictive value for short-term prognosis or long-term prognosis of TILs and its subtypes CD3 + , CD4 + , CD8 + , CXCL13 and Foxp3 + lymphocytes. (c) TILs were clearly de ned, of which cut-off value chooses greater or equal to 10%. (d) The published data contains articles with relative risk (RR), odds ratio (OR) or hazard ratio (HR) and 95% con dence interval (95%CI). (e) Original research articles published in English with full-text.
The exclusion criteria were following: (a) TILs were not clearly de ned with using the median, quartiles or various scores and related statistics. (b) Lacking of key information such as, odds ratio (OR), hazard ratio (HR), 95%CI and P value. (c) Non-English language. (d) In vitro and animal study. (e) Reviews, commentaries, editorials, protocols, case reports, qualitative research, or letters.

Data extraction
Two authors (Li and Zhang) independently extracted the data to ensure the reliability of the data. Any disagreements are resolved by consensus, a third party (Yang) will intervene if necessary. All data and information were recorded in pre-designed tables. We extracted the following data for this review: rst author's name, publication date and country, study design information, number of participants, neoadjuvant chemotherapy program, median follow-up time, breast cancer type, TILs types, cut-off value, clinicopathological characteristics, outcome indicators of long-term prognosis and short-term prognosis.
The cut-off value of TILs varied in different studies, and we used prede ned cut-off value to distinguish high and low TIL subgroups. 'High' TILs were de ned according to the cut-off ≥ 10%. Pathological complete response (PCR), de ned as the absence of all invasive disease cells and lymph node metastasis [24].
Overall survival (OS), de ned as the from the date of breast cancer diagnosis to the time of with any events, disease-free survival (DFS), de ned as the period from the start of treatment to the rst recurrence, or to death without any reason [25]. TILs can also be classi ed as stromal (sTILs) or intraepithelial (iTILs) according to different locations. To avoid bias from studies contributing very long-term follow-up data compared with other studies, both OS and DFS rates were standardized.
Assessment of study quality Two authors (Li and Zhang) independently evaluated the risk of bias in 11 RCTs using the Cochrane Collaboration's risk of bias assessment tool [26], including the following seven modules: random sequence generation; allocation concealment; blinding (participants, personnel, and outcome assessment); incomplete outcome data; and selective reporting and other bias. These domains are categorized as having a high, low, or unclear risk of bias. 11 quali ed case-control studies were assessed according to the Newcastle-Ottawa scale (NOS) [27]. The NOS contains eight items, which are categorized into the three dimensions of selection, comparability, and exposure (case control studies). The quality scores in NOS ranged from 0 to 9 and studies with scores 6 or more are rated as high quality. If there is any disagreement in the results of the literature quality evaluation, it is resolved by consensus, and a third author (Yang) will intervene if necessary.

Statistical analysis
TILs were analyzed as continuous variable (per 10% increment) and binary variable (cut-off value ≥ 10%). For continuous variable, HR and 95%CI were calculated for the effect of TILs in terms of DFS, OS and PCR. For binary variable, we evaluated the overall OR and 95%CI of eligible data for the predictive value of TILs in PCR to NACT. An OR > 1 indicates higher PCR in breast cancer patients with higher TILs group, an OR < 1 indicates lower PCR in breast cancer patients with higher TILs group. HR < 1 indicates that TILs can predict DFS and OS in breast cancer patients with NACT, HR > 1 is the opposite. The pooled OR and HR are considered statistically signi cant if the 95%CI did not include 1.0, with a P value of ≤ 0.05.
The Mantel-Haenzel method was used to obtain xed effects model of the pooled ORs [28], and standard checks of the homogeneity assumption were performed [29]. In the presence of signi cant heterogeneity among the trials, estimate the ORs using the random effects model [30], and use sensitivity analyses. To obtain a quantitative measure of the degree of inconsistency in the results of the studies, the Higgins I 2 index was computed [31]. The likelihood of publication bias was assessed by visual inspection of funnel plot for study size against treatment effect [32]. The STATA software version 15.0 or Revman5.3 were used for all statistical analyses.

Literature quality evaluation results
We evaluated the risk of bias for all included prospective studies (n = 11). The main sources of bias were related to blinded (participants and personnel). The risk of bias assessments for each cohort and evaluations for each domain across full reported studies are shown in Supplementary Fig. 1a and Fig. 1b.

Correlation of TILs with clinicopathological parameters
The correlations between clinicopathologic characteristics and TILs were analyzed in Table 2. Total TILs levels were not associated with tumor grade, tumor size and patients age, but there was association with shorter DFS in node-positive breast cancer patients (pooled HR = 3.340, 95%CI, 2.280-4.890). Unfortunately, we had not analyzed the relationship of Ki-67, hormone receptor or HER2 status due to the limited data.

Cut-off value of TILs and PCR
We used prede ned cut-off points to distinguish high and low TIL subgroups. In studies (types of breast cancer not clearly classi ed in the literature) and where the cut-off value for TILs was ≥ 10%, the high level of TILs can predict PCR rate better. This cut-off value of TILs applied to the TNBC patients, either. However, for HER2-positive breast cancer patients, subgroup analysis showed in 2 studies where the cut-off value for TILs was 10%, the pooled OR for PCR was 0.683 (95% CI, 0.029-16.014, P = 0.813). If we grouped studies with a cut-off of TILs ≥ 30%, the OR was 2.631 (95% CI, 1.739-3.982, P = 0.000) and no signi cant heterogeneity (P = 0.104, I 2 = 45.2%).

Publication bias
Funnel plot analysis which was performed to assess the publication bias of the selected studies for the pooled PCR rate ( Supplementary Fig. 2). Visual inspection of analysis indicated some evidence of asymmetry, but Egger's tests indicated that there was no signi cant publication bias, P > 0.5. Limited data for OS outcome indicators, so we did not do the funnel plot.

Sensitivity analysis
To assess the impact of each included study, a sensitivity analysis was performed. After excluding each study, similar results were observed.

Discussion
It is well known that TILs of tumor microenvironment modulate the cancer cell killing effect of NACT. TILs can not only effectively re ect the interaction between the immune microenvironment of the body and tumor cells, but also predict the outcome and treatment effect from the side, so as to make certain guidance for the formulation and adjustment of clinical treatment plan for tumor patients. However, concerning the effect of TILs on NACT treatment outcome in breast tumors, the results are sometimes discordant. Obviously, this meta-analysis is the rst study to evaluate systematically the relationship between TILs, a new biomarker for breast cancer, and the outcome of NACT.

Correlation of TILs with clinicopathological parameters
We analyzed the relationship between TILs and clinicopathological characteristics. As reported in most literature, total TILs are not associated with age or tumor size in breast cancer patients [33,35,39,43,50]. Earlier published showed that there was major correlation between the expression of TILs and lymph node status [49,52], which was similar with our research. TILs highly expressed in axillary lymph node status positive breast cancer patients, has a high risk of recurrence and metastasis. Denkert C [41] showed that TIL levels of breast cancer differed by tumor grade, but our study did not achieve a similar result. Therefore, it would be signi cative to further explore the relationships between clinicopathologic characteristics and TILs.

Impact of TILs on PCR prognosis
In clinical application, PCR has always been regarded as an indicator of the short-term prognosis of NACT. In recent years, the impact of TILs on the PCR rate for patients with breast cancer on Neoadjuvant Therapy, has attracted much attention. We observed a consistent positive association of increased TILs level with increased PCR, especially in HER2-positive breast cancer and TNBC [21], two molecular subtypes that have obvious response to NACT treatment. Several literature reports [21,54] that the effect of TILs for PCR had the opposite effect in HER2-negative breast cancer, compared with TNBC and HER2-positive breast cancer. By contrast, our meta-analysis showed TILs to be a prognostic marker that can also better predict PCR for HER2-negative breast cancer patients receiving NACT. However, due to the limited literature included in this study, careful analysis is needed. Patients with high-level TILs were con rmed to have a higher chance of achieving PCR, but there were a few differences in Estrogen Receptor (ER) subtypes. We found the high-level TILs not correlated with higher PCR rate in ER negative or ER positive breast cancer, which is slightly different from the previous reports [15][16]. In summary, the predictive value of TILs levels on NACT response is different in all molecular subtypes. This supports the hypothesis that the cellular composition of immune in ltration in tumors is different among each breast cancer type, which determines different clinical outcomes and NACT response [55].
Patients con rmed to have HER2 ampli cation further receive weekly trastuzumab or lapatinib infusions. For HER2-positive patients received trastuzumab, the prognostic role of TILs has not been de nitely ascertained. A study reports [37] that high levels of TILs may further enhance the e cacy of an anti-HER2 therapy by trastuzumab compared with lapatinib therapy. However, our analysis showed that high levels of TILs may not enhance the e cacy of an anti-HER2 therapy by trastuzumab and lapatinib, and cannot predict the PCR prognostic effect of these drugs. Further studies should be conducted to determine whether HER2-positive patients with higher TILs level can obtain bene t from trastuzumab and lapatinib therapy, due to limited data.

Cut-off value of TILs and PCR
In our meta-analysis, we used prede ned cut-off value to distinguish high and low groups with different TIL levels, but the actual distribution of TILs suggests that these were arti cial cut-off. The number of TILs was a continuous variable, which could reach any proportion (0-100%). Nevertheless, the categorization into low and high levels of TILs might be relevant for future clinical applications, since strati cation in clinical trials can be more easily done on the basis of categorical variables that divide patients into different groups.
The TILs threshold with the greatest prognostic signi cance is as yet unknown. Our result showed when a cut-off value of TILs ≥ 10% was used, high levels of TILs predict a better PCR rate of NACT. However, for HER2-positive breast cancer, a cut-off ≥ 30% is associated with a greater rate of PCR compared with lower expression levels. In order to determine the optimal threshold of TILs in future clinical trials and improve the accuracy of pathological assessment and prognosis prediction, a TILs threshold of at least 30% is associated with the most powerful outcome prognostication of PCR.

Impact of TILs on long-term prognosis
In general, DFS and OS have been regarded as long-term prognostic indicators of NACT. Previous studies [15][16] had shown that higher level of TILs suggested a better prognosis of patients with breast cancer on Neoadjuvant Therapy. Our study also indicated per 10% increase TILs means an extension of DFS and OS, speci cally in patients belonging to TNBC groups. One possible explanation of the differences between HER2 breast cancer and TNBC could be the contribution of different immune cell types. TNBC are considered subtypes with high immunoreactivity. In TNBC, the presence of many immune cell subtypes, including B cells, T cells, and macrophages, were linked to improved survival. Unfortunately, data from the HER2-negative breast cancer was not enough available for TIL analysis.
In actual clinical, the number of iTILs were correlated with the number of sTILs, but typically had a much lower density and therefore were less suitable as a biomarker. But our meta-analysis showed that value of TILs for prognostic implication was not affected by TILs location, both iTILs and sTILs contributed to better prognosis and favorable survival.

Prognostic value of TILs subtypes
It is worth noting that TILs contain a variety of subgroups, re ecting that lymphocytes in tumor microenvironment can affect the balance of immune response, leading to different outcomes. Some studies reported that Foxp3 + TILs can suppress antitumor immune response and lead to escape immune clearance [56][57][58]. Therefore, patients with high expression levels of Foxp3 + TILs have few chances of obtaining survival bene t from NACT. On the contrary, recruitment of activated CD8 + cytotoxic T cell of NACT is associated with a better outcome [59][60]. However, our meta-analysis results showed the higher PCR rate of breast cancer patients with high expression levels of Foxp3 + TILs or CD8 + TILs, but signi cant association between FOXP3 + TILs and DFS were not observed.
Besides, an increasing body of evidence suggested that humoral immunity, is important too. CXCL13, formerly termed B cell-attracting chemokine 1, is a cytokine that belongs to the CXC chemokine family. Previously, the expression of immune-related genes in pre-NACT biopsies obtained from TNBC patients was investigated and the expression of the CXCL13 mRNA transcripts. Therefore, CXCL13 also was a predictor of PCR [61]. Gu-Trantien [62] reported that the in ltration of CXCL13-producing CD4 + follicular helper T cells and a CXCL13 gene expression signature were related to a longer disease-free survival in breast cancer patients. Similarly, our pooled analysis results showed CXCL13 gene expression characteristics can prolong the disease-free survival time of breast cancer patients.

Limitations
In spite of this, there still exist some limitations in this meta-analysis. Some of individual results had large heterogeneity, which may be related to differences in the age and types of breast cancer, as well as clinical research methodology included in the study. This study only included published literature in English and limited quantity literature, there might exist language bias and study heterogeneity. Besides, some of the studies we included are retrospective studies, correctness of the results depended on the accuracy of the original literature research, we therefore formulated the strict inclusion and exclusion standard.

Conclusions
In conclusion, our ndings indicate that total TILs can be used as predictors of patients with breast cancer on Neoadjuvant Therapy. The higher the in ltration degree of TILs predicted the better prognosis of neoadjuvant chemotherapy, except for ER-negative and ER-positive breast cancer subtypes. Notably, some subtypes like Foxp3 + TILs show a worse long-term prognosis.
A TILs threshold of at least 30% is associated with the most powerful outcome prognostication of PCR. However, a large number of prospective trials are still needed.