Dissemination of circulating tumor cell clusters occurs early in non-metastatic breast cancer patients


 Background: Metastatic spreading is promoted by cancer cell seeding from the primary tumor into the bloodstream. In patients with metastatic breast cancer (MBC), the clinical relevance of circulating tumor cell clusters (CTC-clusters) has been extensively reported, while their study in earlier stages is limited. Several methods, besides the FDA-cleared CellSearch®, limited to the detection of epithelial-enriched clusters, can be used for the detection of CTC-clusters. We hypothesize that resorting to marker-independent approaches can improve CTC-cluster detection. Methods: Blood samples collected from healthy donors and spiked-in with tumor mammospheres, or from BC patients, were processed for CTC-cluster detection with 3 technologies: CellSearch®, CellSieve™ filters, ScreenCell® filters. The number of CTC-clusters was compared among the technologies and analyzed in relation to patient characteristics and outcome. Results: In spiked-in samples, the 3 technologies showed similar capability of recover epithelial mammospheres, whereas, in a series of 19 clinical samples processed in parallel with the CellSearch® and CellSieve™ filters (that allow the detection of both epithelial and non-epithelial clusters), CTC-clusters were detected in 53% of samples with the CellSearch®, versus 79% and 84% with the CellSieve™, when considering only epithelial or both epithelial and non-epithelial clusters, respectively. Next, blood samples from 37 non-metastatic breast cancer (NMBC) and 23 MBC patients were processed using ScreenCell® filters for attaining both unbiased enrichment and marker-independent identification of clusters based on cytomorphological criteria. At baseline, CTC-clusters were detected in 70% of NMBC cases and in 20% of MBC patients (median number= 2, range 0–20, versus 0, range 0‑15, P =0.0015). Among NMBC patients, clusters were slightly higher in women with node-positive than node‑negative status (0 versus 3, P =0.1110 ) and were more frequently observed in women with luminal‑like and triple-negative tumors than in patients with HER2-positive disease (median CTC-cluster number =4, 5, and 0 for luminal‑like, triple-negative, and HER2-positive BC, respectively, P =0.0467). Conclusions: We demonstrated that CTC-cluster detection can be improved by a marker-independent enrichment and identification, and we reported that CTC-clusters are more frequently detected in NMBC than in MBC patients, suggesting that dissemination of CTC-clusters is an early event in BC natural history.

Background 81 Metastatic spreading is the main cause for death in patients diagnosed with cancer. This process is 82 promoted in its initial steps by cancer cell seeding from the primary tumor into the blood stream. 83 Accordingly, a large amount of data has been collected across different tumor types linking the 84 dissemination of circulating tumor cells (CTCs) with both poor prognosis and treatment 85 failure/resistance [1]. 86 Nonetheless, single CTCs are inefficient in sustaining metastatic dissemination as, to be able to 87 colonize new sites, they must overcome numerous obstacles such as avoid anoikis, escape 88 immunological control by circulating immune cells, resist to sharing stress due to fluid circulation, 89 resulting in the fact that most CTCs do not survive long in the circulation [2, 3]. Therefore, being 90 able to interact with other CTCs or with other cells by generating homo-or heterotypic CTC-clusters 91 appears a biologically reasonable solution for increasing the metastatic potential of CTCs once they 92 are facing the hostile blood environment. 93 Functional studies employing animal models and patient-derived data [4][5][6][7] definitely support a role 94 of CTC-clusters in tumor dissemination and metastasis formation in breast cancer (BC). Such studies 95 also offer hints on the biology of clusters revealing the mechanistic basis for their association with 96 poor outcome and suggesting possible targets for specific treatments aiming at interfere with CTC-97 clusters formation and metastatic dissemination. 98 It is well known that metastatic dissemination occurs at early stages and is followed by a prolonged 99 dormant status of these early disseminated cells [8][9][10]. This observation is supported by data 100 demonstrating that enumeration of single CTCs predicts progression-free survival (PFS) and overall 101 survival (OS) also in non-metastatic breast cancer (NMBC) patients (women with no evidence for 102 distant metastases), both prior [11,12] or after [13] breast surgery. Therefore, addressing the 103 presence of CTC-clusters in BC patients without clinically overt metastases holds promise to gain 104 important hints about the dissemination process. 105 However, this issue has not yet been addressed and, in BC, most studies evaluating the clinical 106 relevance of CTC-clusters have been limited to patients with metastatic or advanced disease [14][15][16][17][18][19][20]. 107 Overall, these studies suggest a direct association between detection of CTC-clusters and poor 108 clinical outcome, although the heterogeneous patient case series, technical issues in CTC-cluster 109 enumeration and variable definitions of CTC-clusters must be taken into account as possible 110 limitations and confounding factors. 111 Noteworthy, all the mentioned studies used the CellSearch® for CTC-cluster detection, which is 112 possibly not the ideal method for CTC-cluster identification. The CellSearch® is a platform 113 specifically developed for assuring high detection of single CTCs with epithelial features and for 114 attaining standardization of their enumeration [21]. No data are instead available on its performance 115 for CTC-cluster detection both in terms of recovery and of the integrity of isolated clusters. The 116 CellSearch® approach includes a CTC-enrichment step employing ferrofluid nanoparticles with 117 antibodies targeting EpCAM, which operates a selection in favor of clusters with exquisite epithelial 118 features and possibly excludes larger CTC-clusters [22], that could result into an underestimation in 119 CTC-cluster enumeration. Moreover, epithelial-to-mesenchymal transition (EMT) is recognized as an 120 important driver of tumor invasion and metastatic dissemination [23], and literature data supported an 121 increasing detection of mesenchymal markers in CTC-clusters compared to single CTCs in breast 122 cancer patients [24]. Thus, investigating the use of epitope-independent methods, compared to the 123 CellSearch®, for CTC-cluster detection is urgently needed to be able to fully appraise the actual 124 clinical value of CTC-cluster in BC. 125 To tackle technical issues in CTC-cluster enumeration we compared, in a series of spiked-in and 126 clinical samples, the number of CTC-clusters recovered using the CellSearch® platform and two 127 size-exclusion methods based on a short-time filtration that allows for the detection of both epithelial 128 and non-epithelial CTC-clusters. Thereafter, we implemented the recovery of CTC-clusters by 129

CTC-cluster enumeration by ScreenCell® filters 206
Peripheral blood samples (9 mL), collected into K2EDTA BD Vacutainer tubes (BD) using a 21G 207 needle, were stored at 4 °C in the dark and processed within 2.5 hours for CTC-cluster enrichment 208 using the ScreenCell® Cyto kit (ScreenCell, Sarcelles, France) [29] according to the manufacturer's 209 instructions, with slight modifications with respect to what previously described [30,31]. Briefly, 210 three aliquots of 3.0 mL of whole blood per sample were separately mixed with 4 mL of a proprietary 211 red blood cell lysis and fixation buffer (ScreenCell® FC2 filtration buffer) and incubated for 8 212 minutes at room temperature. Each aliquot was filtered to isolate CTC-clusters using ScreenCell® 213 Cyto isolation supports (ISs), consisting in a microporous membrane with pores of 6.5 µm diameter. 214 After rinsing with PBS, ISs were air-dried and stained with Hematoxyilin Solution S (Merck, 215 Darmstadt, Germany) for 1 min and Shandon Eosin Y Aqueous Solution (Thermo Fisher Scientific  216 Inc., Waltham, MA, USA) for 30 seconds, at room temperature; or with May Grünwald (Merck 217 Millipore, Burlington, MA, USA; incubation for 2.5 min followed by a second incubation for 2.5 min 218 in May Grünwald diluted 1:2 with water) and Giemsa (Merck Millipore; diluted 1:10 with water, 10 219 min incubation) at room temperature. The stained ISs were sent to ScreenCell for evaluation by a 220 certified pathologist according to published criteria [32]. CTC-clusters were defined as clusters of ≥2 221 CTCs showing the criteria of malignancy: nuclear size ≥20 μm, nuclear-to-cytoplasmic ratio ≥0.75, 222 irregular nuclear contours and nuclear hyperchromatism. In case the cytoplasm edges were not 223 clearly visible inside the cluster (preventing nuclear-to-cytoplasmic ratio evaluation), malignancy 224 identification was mainly based on nuclei appearance: nuclei scattered irregularly through the cluster 225 and anisokaryosis (i.e. nuclei of variable sizes and shapes), in addition to nuclear size ≥20 μm and 226 irregular nuclear membrane. Detailed guidelines for ScreenCell filter interpretation are described 227 Samples showing poor quality of cytology were excluded from the analysis. The total 228 number of CTC-clusters for each sample was obtained by summing the CTC-clusters identified in the 229 3 ISs (corresponding to 9 mL of blood). 230

Statistical Analysis 231
Clinical and pathological variables were reported through descriptive analyses. Categorical variables 232 were reported as frequency distribution, whereas continuous variables were described according to 233

Comparison of different strategies for CTC-clusters identification 245
Technical validation of approaches used for CTC-cluster detection 246 To explore technical limitations of standard (CellSearch®) and filtration-based methods for 247 CTC-cluster detection, spiking experiments were performed comparing size-exclusion approaches 248 with the CellSearch® method (the currently most frequently used method in CTC-cluster studies). In 249 particular, we compared 3 technologies: the CellSearch®, the CellSieve™ filters and the 250 ScreenCell® filters. The latter two are very similar for the enrichment strategy (based on short-time 251 filtration through a membrane with pores of 7 and 6.5 µm, respectively), but differ for the criteria 252  Table 1.  Another possible concern about using filtration devices for CTC-cluster identification, is the Once assessed that the ability of enriching clusters for the 3 technologies was similar, we next aimed 286 at evaluating whether phenotypic heterogeneity of CTC-clusters in clinical samples (i.e. the presence 287 of both epithelial and non-epithelial clusters) could have an impact on CTC-cluster detection by the 288 epithelial marker-based CellSearch® platform, compared to the marker-independent and size-based 289 approaches. In that respect, 19 blood samples collected from 16 patients with MBC were processed 290 in parallel with CellSearch® and CellSieve™ filters (Fig. 1A). For this analysis, CellSieve™ was 291 used as the representative among the 2 filtration methods, since its enrichment strategy is the same of 292 ScreenCell filters (based on size), but its identification criteria are based on the detection of epithelial 293 markers, and therefore allow for the distinction between epithelial and non-epithelial clusters (not 294 possible with ScreenCell® filters). 295 Blood samples were collected from clinically selected patients with highly aggressive disease and 296 during disease progression to increase the probability of CTC-cluster presence (in Additional file 1, 297 patients' clinico-pathological characteristics are reported). For samples processed with the 298 CellSearch®, only CTC-clusters expressing CK (CK pos CTC-clusters, defined as groups of 2 or more 299 cells showing CK pos and CD45 neg staining, Fig. 1B) could be detected, whereas for samples processed 300 with CellSieve™ filters it was possible to identify both CK pos and CK neg clusters ( Fig. 1C and 1D, 301 respectively). CK neg clusters were defined as groups of 2 or more cells showing a CK neg , CD45 neg and 302 CD31 neg staining (the latter marker allowing for the exclusion of endothelial cell clusters). CD31 303 expression was unexpectedly observed also in a few CK pos CTC-clusters (Additional file 2). These 304 clusters, being CK pos and CD45 neg were included in the analysis. 305 We detected ≥1 CK pos CTC-clusters in 10 samples by using the CellSearch® and in 15 samples by 306 using CellSieve™ filters (Additional file 3). Moreover, in the samples processed by filtration, CK neg 307 clusters were observed in 12 out of 18 evaluable samples, in 1 case alone and in 11 cases together 308 with CK pos CTC-clusters. Overall, by considering 1 cluster per sample as positivity threshold, the 309 positivity rate increased by using the size-based approach, going from 53 % (using the CellSearch®) 310 to 79 % considering only CK pos and 84 % considering both CK pos and CK neg clusters identified with 311 CellSieve™ filters (Fig. 1E). Moreover, the absolute numbers of detected clusters were higher in 312 samples processed with CellSieve™ filters than with the CellSearch® (Fig. 1F; Additional file 3). In 313 samples processed with the CellSearch®, a median of 1 CK pos CTC-cluster (interquantile range, 314 IQR = 0-2; range 0-108) was identified, compared to a median of 3 CK pos CTC-clusters (IQR 1-6; 315 range 0-112) for samples processed with CellSieve™ filters (P = 0.0293). The increase in cluster 316 counts for samples processed with CellSieve™ filters was even higher when considering CK pos and 317 CK neg clusters together (median = 7, IQR 1-11; range 0-112, P = 0.0038). 318 These results suggest that by using a size-based and marker-independent approach it is possible to 319 detect a higher number of clusters, allowing to identify them also in patients considered CTC-cluster 320 negative by the CellSearch®. However, the observed phenotypic heterogeneity of clusters in BC 321 patient samples, and in particular the presence of CK neg clusters, highlighted an important limitation 322 of CellSieve™ technology, which was able to enrich this type of clusters, but did not allow to 323 reliably assess their malignancy (since they were only DAPI pos ). On the other hand, ScreenCell® 324 technology had the same ability of enriching CK neg clusters (since it is size-based as well), but its 325 identification was based on cytomorphological evaluation and was therefore not dependent on the 326 expression of any specific tumor markers. We therefore decided to use ScreenCell® filters to 327 investigate the presence of CTC-clusters in both MBC and NMBC patients. 328

CTC-clusters in patients with metastatic and non-metastatic breast cancer 351
To investigate the presence of CTC-clusters in our cohort of patients with MBC and NMBC, blood 352 samples collected before starting systemic treatment underwent CTC-cluster enrichment by filtration, 353 followed by a marker-independent CTC-cluster identification based on cytomorphological criteria 354 using ScreenCell® filters ( Fig. 2A). This simplified identification strategy requires only H&E 355 staining rather than immunofluorescence, and it gives reliable results regarding cell malignancy, 356 independently from the expression of specific markers. At baseline, in NMBC patients, 1 or more 357 CTC-clusters were detected in 26/37 cases (70 %), with a median of 2 clusters per sample (range 358 0-20) (Fig. 2B). Among the 23 baseline samples collected from MBC patients, 3 samples were from 359 pre-treated patients and one was not evaluable for CTC-cluster identification; CTC-clusters were 360 detected in 4 of the 19 remaining samples (21 %), with a median of 0 CTC-clusters per sample (range 361 0-15). CTC-clusters were therefore more frequent and more abundant in patients with NMBC than 362 MBC (P = 0.0015). In particular, patients with stage II BC showed a higher CTC-cluster count than 363 patients with stage III and IV BC (Additional file 4A). Among patients with NMBC, a slightly higher 364 number of CTC-clusters was detected in patients with node-positive status (Fig. 2C), although this 365 difference was not statistically significant (median CTC-cluster number = 0 vs. 3 for node-negative 366 vs. node-positive patients, P = 0.1110). CTC-clusters were more frequently observed in patients with 367 luminal-like and triple negative BC than in patients with HER2-positive disease (median CTC-cluster 368 number = 4, 5, and 0 for luminal-like, triple-negative, and HER2-positive BC respectively, P = 369 0.0467) (Fig. 2D). For 25 patients for whom a primary tumor tissue sample was available, the 370 presence of CTC-clusters was analyzed with respect to the presence of tumor-infiltrating 371 lymphocytes (TILs) at the primary tumor site but no difference in CTC-cluster counts was observed 372 between patients presenting a high or low level of TILs (median CTC-cluster number = 3 vs. 2 for 373 patients with < 12 % vs. ≥ 12 % TILs, P = 0.5392) (Additional file 4B). 374 These results indicate that CTC-clusters are present in early stages in BC patients and are more 375 frequent than in MBC patients. Among NMBC patients, CTC-clusters are more abundant in the 376 blood of patients with HER2-negative disease. 377

Longitudinal evaluation of CTC-clusters during neoadjuvant therapy 378
To further investigate the clinical relevance of CTC-clusters in NMBC patients, longitudinal blood 379 samples collected at baseline (N = 37), during (N = 30), at the end (N = 14) of NAC and after surgery 380 (N = 18) were analyzed (Fig.3A). The median number of detected CTC-clusters at baseline was 2 381 (range 0-20), during treatment (DT) was 1 (range 0-97), and at the end of treatment (EOT) was 3 382 (range 0-116). Thus, CTC-clusters did not decrease during NAC, but instead increased in some 383 patients. Overall, no significant differences were observed in DT and EOT with respect to baseline. 384 On the other hand, a significant decrease was observed from DT to surgery (P = 0.0448) and EOT to 385 surgery (P = 0.0208). Only a slight decrease was instead observed between baseline and surgery (P = 386 0.0678). The median number of CTC-clusters after surgery was 0 (range 0-20). 387 At baseline, numerically less clusters were observed in NAC-responders, i.e. patients with complete 388 disappearance or a reduction of primary tumor volume of at least 50% after NAC, as compared to 389 non-responders, i.e. patients with stable disease after NAC: 1 cluster (range 0-20) versus 4 clusters 390 (range 0-12), respectively (P = 0.58). The presence of CTC-clusters at baseline was not significantly 391 associated with pCR (Additional file 5A). However, patients without clusters at baseline reported a 392 numerically higher pCR rate as compared with those presenting with clusters, 27% versus 23%, 393 respectively. Moreover, after surgery, a significantly lower number of clusters was observed in 394 patients with pathological complete or partial response versus stable disease (P = 0.0208) (Additional 395 file 5B). As of May 15, 2020, a total of 10 out of 37 NMBC patients relapsed. No difference in 396 baseline or post treatment distribution of clusters was reported among patients with or without a 397 relapse. At the same date, 4 out of 19 evaluable MBC patients had died, notably the negative 398 predictive value of clusters at baseline in this case was as high as 86%, but the data is merely 399 explorative due to the small sample size. 400 We present two examples of patients who responded to NAC but did not achieve pCR, illustrating 401 the cluster's dynamics during treatment. 402 Patient A (Fig. 3B)  In the current study, we have challenged the most frequently used technical approach for CTC-cluster 432 detection in BC, the CellSearch®, by comparing it with methods based on size exclusion. We report 433 that filtration allowed to detect a higher number of clusters in the blood of BC patients. Moreover, by 434 using a filtration-based approach to analyze blood samples prospectively collected from NMBC and 435 MBC patients, we observed that CTC-clusters were more frequently detected in NMBC than in MBC 436 patients, and that molecular subtypes affected their presence in NMBC. Finally, the presence of 437 clusters before starting neoadjuvant treatment did not associate with pCR and their numbers 438 increased during treatment, but dropped after surgery. 439 To the best of our knowledge, this is the first study specifically comparing CTC-cluster detection by 440 Conversely, the observed increased detection of epithelial clusters (CK pos ) is an unexpected finding. 453 A possible explanation is that CK pos CTC-clusters can also include cells undergoing EMT and 454 therefore expressing a mixed phenotype rather than a frankly epithelial one. Overall we report that, in baseline samples collected at the beginning of NAC, the detection of at 474 least 1 CTC-cluster occurred at least 3-times more frequently in women with early breast cancer than 475 in women beginning first line treatment for MBC (a result that we also observed in our previous pilot 476 study, which was comparing ScreenCell® with AdnaTest technology [30]). Although, due to the 477 small case series, we have not done a formal analysis to exclude a bias due to different distribution of 478 molecular subtypes between the two groups, molecular subtype linked effects would have impacted 479 the data in opposite direction than observed. Thus our findings support the concept that dissemination 480 of CTC-clusters is an early event in NMBC patients, rather than an event occurring during metastatic 481 progression, as might have been expected by the high metastatic potential of clusters [4]. Since 482 dissemination is proven to occur early in breast cancer [8,9], and indeed single CTCs hold 483 prognostic value also in NMBC women [11][12][13], the more frequent presence of CTC-clusters and the 484 higher number of clusters seen in early rather than later steps of the disease is intriguing. 485 Nevertheless, many questions on clinical and biological aspects still remain to be answered. We 486 observed that molecular subtypes affect the prevalence of CTC-clusters. In particular CTC-clusters 487 were found to be significantly more frequent in women bearing HER2-negative tumors, a result that 488 may appear as contra-intuitive since HER2-positive tumors are more aggressive and are frequently 489 associated with stemness markers [40]. Moreover, we have noticed that luminal-like tumors release 490 high number of clusters, a finding possibly linked to their late relapse-pattern and to a more efficient 491 promotion of dormancy within the clusters from patients with ER+ tumors [41]. Overall, this 492 suggests that clusters should be studied in molecularly homogeneous populations, although this could 493 not be done in this study, due to the limited number of patients. 494 In our cohort of NMBC patients, the detection of clusters did not correlate with the likelihood of 495 achieving pCR, a finding already reported in the literature for CTCs [12,38]. Moreover, during the 496 course of treatment a trend towards an increase in CTC-clusters rather than a decrease was observed, 497 as also described in another study using ScreenCell® filters [38]. Indeed, only after surgery we 498 actually observed a decrease in the number of clusters. Although a significantly higher number of 499 clusters persisting after surgery was detected in patients with a pathologically non-responding disease 500 (median 3, IQR 1 -11.5 vs 0, IQR 0-1 for non-responders and responders, respectively). 501 Thus, it may be speculated that in NMBC, clusters formation is related to the presence and 502 characteristics of primary tumor, and the neoadjuvant treatment has a different effect on the primary 503 tumor and on clusters. Moreover, despite this study is not properly powered to detect differences in 504 disease-free survival and no association was observed between relapse and CTC-clusters at baseline, 505 it is intriguing to think of potential applications of cluster enumeration after surgery as a completion 506 of pathological staging to assess the overall combined response to systemic and locoregional 507 treatments. 508 Notably, a discrepancy between cluster dynamics and imaging was observed. As consistently shown 509 by the index cases, clusters generally increased during NAC notwithstanding the concomitant 510 radiological and metabolic response. On the other hand, patients that did not show response to NAC 511 had a significantly higher number of clusters after surgery. This suggests a more nuanced role of 512 clusters in NMBC with respect to that of epithelial clusters in the metastatic setting. 513 A crucial question raised by our results deals with the phenotype and the actual composition of 514 clusters. Having used a size-based approach for cluster enrichments and morphological criteria for 515 the detection, we were confident about the malignancy of the clusters, but we lost the information 516 regarding their epithelial/non-epithelial phenotype. However, since in the numerous studies run with 517 the CellSearch® in women with early disease, massive presence of clusters has not been reported, we 518 speculate that clusters detected in the current study are not frankly epithelial, but rather with a 519 mesenchymal or with a mixed phenotype. 520 Regarding the cluster composition, the role of inflammatory cells remains to be addressed. Indeed, 521 cooperation and crosstalk with other blood cells play a relevant role in increasing the metastasis-522 promoting efficiency of cluster [7, 42-45]. However, in the current study, we did not find an 523 association between TILs evaluated on the primary tumor and CTC-clusters, thus the possible 524 interaction between inflammatory cells and CTC-clusters warrants further studies. 525 Finally, the observation that clusters do not disappear with the neoadjuvant treatment (and thus 526 possibly also persist after adjuvant treatment), support the need to develop treatment strategies 527 specifically designed at interfering with clusters [6, 46]. Such strategies would be promising 528 especially if CTC-clusters isolated in NMBC patients would prove to hold metastatic potential, a still 529 unanswered question worth to be addressed in the future. 530 We are aware of the study limits due to the small size and heterogeneity of the case series, although 531 its strength may be linked to the fact that these represent real-world patients, prospectively collected 532 within the daily clinical practice. 533 534

535
This study represents a small snapshot on CTC-cluster detection methods and on the prevalence of 536 clusters in BC patients at different disease stages. Nonetheless, it highlights the possible bias linked 537 to inadequate methods for cluster detection, a technical bias that is worth to be considered in future 538 translational studies. In addition, we report a new observation on the fact that CTC-clusters are 539 frequent in women with NMBC. This represents a provocative finding that needs to be addressed in 540 future studies on larger series of cancer patients, homogeneous with respect to molecular subtype. 541 Finally, the observation that CTC-clusters do not disappear during neoadjuvant treatment, foster the 542 importance of developing treatments specifically aimed at interfering with them. 543 The datasets used and/or analysed during the current study are available from the corresponding 566 author on reasonable request. 567

Competing interests 568
All authors declare no competing interests. 569 We are really grateful to all the patients who donated their blood for this study.