Putative anoikis-resistant subpopulations in colorectal carcinoma: a marker of adverse prognosis

TJ. Putative anoikis-resistant subpopulations in colorectal carcinoma: a marker of adverse prognosis. APMIS 2020; 128: 390 – 400. Anoikis is a form of apoptosis induced when a cell loses contact with the extracellular matrix (ECM). Anoikis resistance is essential for metastasis formation, yet only detectable by in vitro experiments. We present a method for quantitation of putative anoikis-resistant (AR) subpopulations in colorectal carcinoma (CRC) and evaluate their prog- nostic signiﬁcance. We studied 137 CRC cases and identiﬁed cell subpopulations with and without stromal or extracellular matrix (ECM) contact with hematoxylin-and-eosin-stained sections and immunohistochemistry for laminin and type IV collagen. Suprabasal cells of micropapillary structures and inner cells of cribriform and solid structures lacked both stromal contact and contact with ECM proteins. Apoptosis rate (M30) was lower in these subpopulations than in the other carcinoma cells, consistent with putative AR subpopulation. We determined the areal density of these subpopulations (number/mm 2 tumor tissue), and their high areal density independently indicates low cancer-speciﬁc sur- vival. In conclusion, we show evidence that subpopulations of carcinoma cells in micropapillary, cribriform, and solid structures are resistant to anoikis as shown by lack of ECM contact and low apoptosis rate. Abundance of these sub- populations is a new independent indicator of poor prognosis in CRC, consistent with the importance of anoikis resistance in the formation of metastasis. survival analyses, we used receiver operating characteris- tics (ROC) analysis to determine optimal cutoﬀ values for areal densities of subpopulations by using the Youden index. For univariate survival analysis, we generated Kaplan – Meier curves for cancer-speciﬁc survival (CSS) and disease-free survival (DFS) and Log-rank test estimated diﬀerences. Cox regression models veriﬁed the inde- pendent prognostic eﬀects of the subpopulation counts on CSS when adjusted for covariates. Due to the low number of cases in multivariate analyses, we estimated the models for one covariate at a time, as described previously (15,16). We considered two-tailed, exact p-value < 0.05 as statistically signiﬁcant for all the tests.


*Equal contribution.
Colorectal carcinoma (CRC) is the fourth most common reason of cancer-associated deaths (1). About 90% of patients with local disease survive for 5 years, but with distant organ metastasis, the survival rate is only 10% (2). To determine an optimal treatment, prognostic and predictive factors are crucial (3).
Anoikis refers to apoptosis induced by the loss of contact with the extracellular matrix (ECM) (4,5). Anoikis resistance is essential for cancer cell survival during the formation of metastases (6).
There is only indirect evidence suggesting that anoikis resistance is a prognostic factor (7)(8)(9)(10). So far, the assessment of anoikis resistance has been limited to in vitro cell culture-based tests (4) as there are no methods to evaluate the presence and level of anoikis resistance using routine pathological specimens. However, based on the essential role of anoikis resistance in metastasis, the adoption of this concept to histopathological analysis might provide useful information.
We have shown that in carcinomas, anoikis resistance allows cells to form multicellular clusters where the inner cells lose their contact with the ECM, but still survive (11,12). We have previously shown that in CRC, in micropapillary structures (MIPs), defined as luminal extensions formed by piled-up carcinoma cells (11), suprabasal cells form a tumor cell population without ECM contact and still with low apoptosis rate, suggesting that these cells represent one putative anoikis-resistant (AR) tumor cell population (11). In the present study, we have used a similar approach to identify other types of multicellular carcinoma cell clusters without ECM contact and without evidence of increased apoptosis rate in CRC. We quantitated the structures composed of such subpopulations of carcinoma cells in a series of CRC and evaluated their prognostic significance.

Patients
We studied a prospective series of 149 CRC patients (13) operated in Oulu University Hospital between 2006 and 2010 (Table 1). From the series, 12 (8%) cases were excluded due to insufficiency of sample material, leaving a total of 137 cases. Clinical and follow-up data were obtained from the clinical records and Statistics Finland. Cancer-specific survival (CSS) was defined as time from operation to cancer-related death. The Ethics Committee of Oulu University Hospital approved this research project (58/2005, 184/2009).

Histology and immunohistochemistry
Basic histological assessments including grading and stage determination were based on conventional histopathological sections (13). For other analyses, we used previously (13) constructed tissue microarrays (TMAs). For the TMAs, we used H&E stained slides to select optimal sample locations, and depending on the size of the tumor, manually sampled 1-4 (median 3) cores of 3.0 mm diameter. Of the cores, 1-3 were taken from the invasive front and 1-2 from random locations in the tumor bulk, only avoiding necrotic areas. The current analysis for putative AR populations was only focused on the cores representing the bulk.

Detection of putative anoikis-resistant subpopulations
To identify putative AR subpopulations, we first identified carcinoma cells without contact with the ECM and then assessed apoptosis rate in cells with and without ECM contact. We identified subpopulations of tumor cells with and without contact with the ECM or stromal areas using H&E staining (Fig. 1) and immunohistochemical stainings for ECM proteins laminin and type IV collagen (Fig. 2).
In these analyses, we assessed all tumor cells present in the whole area of each core. Carcinoma cells without contact with the ECM or with mesenchymal areas of the carcinomas were found to be within the following three types of structures: (i) MIPs; (ii) cribriform structures; (iii) solid structures (Fig. 1). We defined MIPs as cells piled up at the luminal side of the glandular structures, the minimum thickness of this pile being two cells, and the lateral extent at minimum two cells. Cribriform structures were comprised of groups of cells at least four cells in diameter, and containing scattered, empty spaces without cells. Solid structures consisted of groups of cells at least four cells in diameter forming solid sheets. Suprabasal cells of MIPs and inner cells in cribriform or solid structures were devoid of ECM contact (Fig. 1). The cells in contact with ECM comprised cells of a single cell layer thick columnar epithelium and the outermost cell layer of multilayered epithelium, including the basal layer of MIPs (11) and the outermost layer of cribriform and solid structures (Fig. 1).
The proportions of cells positive for M30 and Ki-67 were separately determined in carcinoma cell subpopulations with and without ECM or stromal contact as defined above (Fig. 2). For the determination of positivity rates in subpopulations without ECM or stromal contact, a representative series of each type of putative AR structures was identified in each TMA tissue core representing tumor bulk. Depending on the occurrence of each structure type, 5-20 of each type were assessed. For the determination of positivity rates in cells with ECM contact, a minimum 10 separate groups of about 10-50 adjacent cells were assessed around cells without ECM contact and in the cells representing unilayered columnar epithelium. In cases with more than one core, mean indexes were calculated and used for the case. One investigator (MP) manually assessed M30 stainings and two investigators (TM and MP) assessed Ki-67, both by using a digital image analysis platform (QuPath, version 0.1.2) (14). For all stainings, cases with inconclusive staining patterns were additionally analyzed by an experienced pathologist (TJK). For assessment of interobserver agreement of the recognition of different subpopulations and cell counts, 10 cases were independently studied by three observers (TM, MP, TJK).

Quantitation of putative anoikis-resistant structures
For each case, the area (mm 2 ) occupied by tumor was determined in all TMA cores representing tumor bulk ( Fig. 3) by using virtual images of H&E stained sections (Aperio ImageScope; Leica Biosystems). All occurrences of each putative AR subpopulation type within the core ( Fig. 3) were recorded to determine their areal density (structures/mm 2 tumor tissue). In 58 (42%) cases, more than one core representing tumor bulk was available. In these cases, mean areal densities were calculated.

Statistical analysis
IBM SPSS Statistics, version 22 (IBM Corp., Armonk, NY, USA), was used for statistical analyses. Reproducibility of the assessments was evaluated by using interclass correlation coefficient (ICC) based on a mean rating (k = 2), absolute-agreement, 2-way mixed-effects model. As areal densities of different subpopulations showed skewed distributions, we assessed their association with categorical clinicopathological features with Mann-Whitney or Kruskal-Wallis tests, and correlations with Spearman rank correlation. Similarly, we used Wilcoxon matched-pair test for comparing apoptosis and proliferation rates in different tumor cell subpopulations. For survival analyses, we used receiver operating characteristics (ROC) analysis to determine optimal cutoff values for areal densities of subpopulations by using the Youden index. For univariate survival analysis, we generated Kaplan-Meier curves for cancer-specific survival (CSS) and disease-free survival (DFS) and Log-rank test estimated differences. Cox regression models verified the independent prognostic effects of the subpopulation counts on CSS when adjusted for covariates. Due to the low number of cases in multivariate analyses, we estimated the models for one covariate at a time, as described previously (15,16). We considered two-tailed, exact p-value < 0.05 as statistically significant for all the tests.

Recognition of putative anoikis-resistant subpopulations
Three types of multicellular carcinoma cell islands were identified, including MIPs, cribriform, and solid structures (Fig. 1). Laminin and type IV collagen stainings showed discontinuous staining on the stromal interface of carcinoma cell islands of all types. However, no extracellular positivity was present within the MIPs, cribriform (Fig. 2), or solid structures, indicating absence of organized ECM within these structures. Apoptosis rate as shown by the proportion of M30 positive cells was higher in the cells with ECM contact than in those without ECM contact (p < 0.001; Figs 2 and 4). This finding supported the concept that the latter subpopulations, including suprabasal cells in MIPs and inner cells in cribriform and solid structures, are resistant to anoikis and can be classified as putative AR populations. Proliferation rates were mostly similar in different subpopulations, and only MIPs showed lower rates than the other subpopulations (Fig. 4).

Quantitation of putative anoikis-resistant subpopulations
We determined areal densities (structures/mm 2 tumor tissue) for structures containing putative AR cell populations (Fig. 3) and calculated the sum of areal densities of all three subtypes. Areal densities of each subpopulation type correlated with the total areal density sum (MIP, q = 0.337; cribriform, q = 0.560; solid, q = 0.558; all p < 0.001). Among the putative AR subpopulations, the areal densities of cribriform and solid structures correlated (q = 0.204; p = 0.017) whereas MIP areal density showed an inverse correlation to that of solid structures (q = 0.234; p = 0.006). There were 58 cases with more than one core available for determination of putative AR areal density. The areal densities correlated moderately between the cores (Pearson c. 0.338, p = 0.010; ICC 0.412, p = 0.025). To assess interrater reproducibility, three investigators (MP, TM, TJK) annotated subpopulations and determined Ki67 indexes in a series of 10 unselected cases. In this series, ICC for Ki-67 based on all subpopulation types was 0.814 for average measures (p < 0.001) and 0.686 for single measurements (p < 0.001), indicating moderate to good reliability applicable for both the identification of subpopulation type and Ki-67 measurement.

Amount of putative anoikis-resistant subpopulations and clinicopathological features of carcinomas
We then analyzed relationships between areal densities of putative AR subpopulations and clinicopathological features of CRC. We found a significant association between low tumor grade and high areal density of MIPs (p = 0.005) and between high tumor grade and high areal density of solid islands (p = 0.029; Table 2). Although BRAF and KRAS mutations did not show any association with the overall amount of AR structures, KRAS mutation was associated with low abundance of solid subpopulation (p = 0.048; Table 2), but no association was found with the tumor stage or carcinoma type (serrated or conventional; Table 2).

Putative anoikis-resistant subpopulation areal density and carcinoma prognosis
To determine the prognostic value of putative anoikis-resistant subpopulations, we dichotomized areal density to high and low by using ROC curve analysis. Determination of cutoff by ROC curve was difficult for MIPs as the area under curve was low (0.527). However, at a closer look, MIPs were rare in grade 3 tumors (median 0.74/mm 2 ; range 0-3.9) compared to grade 1 and 2 tumors (1.4/mm 2 ; 0-16.1; p = 0.005) and correlated negatively with the number of solid structures (see above). Such associations are plausible based on the definition of MIPs, as these structures are extensions of the columnar epithelium and therefore rare in highgrade tumors, which, by definition (17), have an abundance of solid structures and only rare occurrence of gland-like structures with columnar epithelium (Fig. 1). Thus, for MIPs, we limited the ROC curve analysis to grade 1 and 2 cases with a resulting cutoff of 2.83/mm 2 . The cutoffs for areal densities of cribriform and solid subpopulations were 4.1/mm 2 and 0.49/mm 2 , respectively.  In univariate analysis, high areal density of MIPs (30/137; 22%) showed no association with prognosis (Fig. 5), but in grade 1 and 2 tumors, their high areal density (n = 27/117; 23%) tended to associate with worse CSS (p = 0.058). High areal density of the cribriform subpopulation (12/137; 9%) associated with worse survival (CSS; p = 0.004, Fig. 5), and a similar association was seen for the solid subpopulation (80/137; 58%; p = 0.017; Fig. 2). High areal densities associated with low DFS, but the association was significant only for the cribriform subpopulation (p = 0.015, data not shown).
High total areal density based on the sum of areal densities of all three putative AR subpopulations (>6.86/mm 2 ; 27/137; 20%) associated with worse CSS (p = 0.001; Fig. 5). In subgroup analyses, similar differences were found in stage III-IV tumors (p = 0.043), but not in stage I-II tumors (p = 0.645). Finally, high sum of areal densities of cribriform and solid subpopulations (>2.95/mm 2 : 60/137; 44%) associated with worse prognosis (p = 0.036). Although total areal density of putative AR subpopulations did not associate with DFS (p = 0.103), the areal density of cribriform Due to low number of cases, quartiles could not be determined and range (minimum-maximum) is shown instead.
structures showed a significant association (p = 0.015, data not shown). The total putative AR areal density was an independent prognostic factor of CSS estimated with Cox regression model, when adjusted for features including the presence or absence of nodal metastases (N), tumor stage (T1-2 vs T3-4), presence or absence of distant metastasis, lymphatic invasion, blood vessel invasion, tumor grade, infiltrating border, tumor location, and age (Table 3). Independent association for CSS was similarly evident for the areal densities of cribriform and solid structures (data not shown).

DISCUSSION
We show here that subrabasal cells in MIPs (11) and the inner cells in cribriform and solid structures are devoid of contact with mesenchymal areas or basement membrane proteins, and yet show a decreased apoptosis rate compared to tumor cells that are in contact with mesenchymal areas or basement membrane proteins at the stromal-epithelial interface of the tumors. These findings suggest that these subpopulations, all comprised of inner cells of multicellular carcinoma cell groups, represent tumor cells with anoikis resistance in CRC. By definition, anoikis is a form of apoptosis induced by loss of ECM contact (4)(5)11), and resistance to anoikis is considered essential for the survival of tumor cells during their travel to metastatic loci (5,6). Our analysis of 137 CRC patients indicated that high areal density of each of these putative AR subpopulation types associates with adverse prognosis, and for cribriform and solid subpopulations, the association is independent in multivariate analysis. However, the total areal density of all three AR subpopulation types was the most powerful prognostic factor, supporting generalizability of the concept of putative AR subpopulations and their prognostic value. Our observations suggest that the concept of anoikis resistance can be adopted to histopathological analysis of CRC and provide prognostic information.
Anoikis resistance has been mostly analyzed with in vitro cell cultures in anchorage-independent conditions (4). Such an approach allows mechanistic insight into anoikis resistance, but does not serve as a tool for routine analysis of anoikis resistance in clinical tumor specimens since fresh tumor specimens are needed. Although the prognostic and predictive significance of anoikis resistance in human malignancies is unknown, reports of prognostic biomarkers functionally linked to anoikis resistance, such as tyrosine kinase receptor B and HCRP-1, support the importance of anoikis resistance as a prognostic factor (7,18). Our study is the first one to use the concept of histopathological recognition of anoikis resistance and to utilize it to get prognostic information. Confirmation of the biological equivalence of in vitro anoikis resistance and anoikis resistance as detected by the presence of putative AR subpopulations in histological analysis would require studies showing correlation between in vitro detection of anoikis resistance with that based on histopathology. So far, we lack such biological confirmation of the current concept of anoikis resistance recognition, but this does not depreciate the prognostic value of our concept. However, confirmation of the prognostic value with an independent case series is still warranted.
The mechanisms causing the formation of the putative AR structures remain speculative. The ability of a cell to resist anoikis becomes evident when it loses its contact with the ECM. The mechanisms leading to loss of such contact include extrusion caused by cellular crowding (19), mechanical detachment, or regulatory aberration leading to loss of integrin function (20)(21)(22). To survive in such conditions, cells must respond either by inhibiting pro-apoptotic signaling or by activating anti-apoptotic mechanisms (5,6). It is likely that, in addition to anoikis resistance, other aberrations, such as suppression of apical polarity signaling (23), contribute to formation of the putative AR structures. Supporting the concept of multicellular groups without ECM contact indicating anoikis resistance, we have recently shown that transfection of Caco-2 cells with mutated KRAS or BRAF gene induces anoikis resistance (12) and, simultaneously, changes 3D growth from simple unilayered cysts to either focal piling of the cells reminiscent of MIP structures or solid growth reminiscent of solid structures, respectively (12).
The recognition of AR subpopulations was relatively reproducible. Tangentially sectioned glandular structures with a single-layered epithelium may mimic solid structures or MIPs. However, determination of areal densities of these putative AR subpopulations likely diminishes this bias, since high densities are an unlikely consequence of section Table 3. Univariate and multivariate Cox regression model for 5-year cancer-specific survival presenting independence of the prognostic significance of high total putative AR subpopulation areal density effects. We used tissue microarrays (TMAs) for the quantitation of putative AR subpopulations. Supporting the validity of this approach in terms of structural heterogeneity in carcinomas, we saw a moderate correlation between different cores. However, more studies are needed to address the withintumor variation of the putative AR structures and the clinical significance of such variation.
Assessment of anoikis resistance in routine diagnostic work by the methods described in the present study does not require any special staining techniques as it can be performed by using conventional H&E stained sections. In addition, although we used digital images and image analysis programs for determination of areal densities of the putative AR populations, it would also be possible by using conventional microscopy; for example, by counting their occurrences in a certain number of visual fields with a known area. We did not measure the time consumption for the present analyses, but a crude estimate would be 3-30 min for each case, depending on the experience of the assessor and the quantity of putative AR structures. Other options for the quantitation of putative AR structures include classical morphometrical methods (24).
We observed an association of high areal density of cribriform structures with poor prognosis. This finding is in agreement with reports of poor prognosis in cribriform-comedo type adenocarcinoma, a rare CRC subtype (25)(26)(27). However, the prognostic significance of the overall extent of cribriform pattern, as shown in our study, has been unknown in CRC. High areal density of solid structures associated with high grade, as expected (17). Tumor grade is a rather weak prognostic factor (28), and the reproducibility of grading in CRC is suboptimal (29). The recently described grading system for CRC is based on the amount of so-called poorly differentiated clusters (30), composed of at minimum five cells without any gland formation. Based on this definition (30), larger poorly differentiated clusters may contain putative AR tumor cells, comparable to those in solid structures in our study. These analogies suggest that the current concept of morphological expression of anoikis resistance may present a plausible biological explanation for both the traditional grading criteria of CRC and the grading based on poorly differentiated clusters.
We used tumor bulk specimens for evaluation of putative AR subpopulations. In the tumor front, budding is present in about 37% of the cases and indicates poor prognosis (31,32). Buds are single cells or groups with a maximum extent of four cells, with all cells in contact with the mesenchymal areas and the local ECM (32). According to our current criteria, buds do not qualify as putative AR structures and hence, the quantitation of the putative AR structures at the front likely does not provide prognostic information. Bulk and front show other biological and structural differences (33). Taken together, we suggest that tumor bulk may be the part of the tumor where biological conditions in the carcinoma cells and the microenvironment allow the emergence of features that mark potential of the cells to resist anoikis. Hence, we quantitated the putative AR structures only in the tumor bulk.
In conclusion, it seems possible to recognize and quantitate specific subpopulations of CRC tumor cells which likely indicate anoikis resistance, as shown by the absence of ECM contact and decreased apoptosis indexes in the cells. Importantly, abundance of such putative AR populations associated clearly with an adverse prognosis. Further studies showing correlation between histopathologically determined anoikis resistance with that based on conventional in vitro assays are necessary to confirm the current concept, and studies with larger case series are advised to confirm the prognostic value of our assay.