CD271 Expression on Patient Melanoma Cells Is Unstable and Unlinked to Tumorigenicity

The stability of markers that identify cancer cells that propagate disease is important to the outcomes of targeted therapy strategies. In human melanoma, con ﬂ icting data exist as to whether hierarchical expression of CD271/p75/NGFR (nerve growth factor receptor) marks cells with enriched tumorigenicity, which would compel their speci ﬁ c targeting in therapy. To test whether these discrepancies relate to differences among groups in assay approaches, we undertook side-by-side testing of published methods of patient-derived melanoma xenograft-ing (PDX), including comparisons of tissue digestion procedures or coinjected Matrigel formulations. We found that CD271 (cid:1) and CD271 þ melanoma cells from each of seven patients were similarly tumorigenic, regardless of assay variations. Surprisingly variable CD271 expression patterns were observed in the analyses of sibling PDX tumors ( n ¼ 68) grown in the same experiments from either CD271 (cid:1) or CD271 þ cells obtained from patients. This indicates unstable intratumoral lineage relationships between CD271 (cid:1) and CD271 þ melanoma cells that are inconsistent with classical, epigenetically based theories of disease progression, such as the cancer stem cell and plasticity models. SNP genotyping of pairs of sibling PDX tumors grown from phenotypically identical CD271 (cid:1) or CD271 þ cells showed large pairwise differences in copy number (28% – 48%). Differences were also apparent in the copy number pro ﬁ les of CD271 (cid:1) and CD271 þ cells puri ﬁ ed directly from each of the four melanomas (1.4% – 23%). Thus, CD271 expression in patient melanomas is unstable, not consistently linked to increased tumorigenicity and associated with genetic heterogeneity, undermining its use as a marker in clinical studies. Cancer Res; 76(13); 3965 – 77. (cid:1) 2016 AACR.


Introduction
Extensive effort has been made across a wide range of cancers to identify markers of cells that propagate malignant disease.Indeed, the effectiveness of targeted cancer therapies depends on identifying markers in or on cancer cells that are reproducibly associated with malignant behaviors.Some of the most useful markers are those generated by genetic mutation.In some melanomas, for example, mutated BRAF is a stable, ubiquitous marker (1) that dominantly drives disease progression (2).This necessitates the continued targeting of BRAF even in treatment approaches that seek to overcome mechanisms of BRAF inhibitor resistance (e.g., ClinicalTrials.govidentifier: NCT02159066).
Markers related to epigenetically driven malignant states are also potential therapy targets.For example, propagation of melanoma cell lines is abrogated by targeting the histone H3 lysine 4 demethylase JARID1B, which otherwise reversibly drives slowcycling, tumorigenic cell subpopulations (3).Markers expressed specifically on cancer stem cells (CSC) are particularly appealing targets in cancers that follow a CSC model (4).However, the usefulness of markers that define epigenetically driven mechanisms of disease progression depends on their stability, which is seldom tested in uncultured cancer cells.The targeting in patients of cancer markers that are not consistently expressed and/or that do not consistently define malignant states in cancer cells will not offer substantial clinical benefit.
In melanoma, cells distinguished by differences in expression of the neural crest stem cell marker CD271/p75/NGFR were shown by multiple groups (5,6) to have differing abilities for tumor formation in patient-derived xenograft (PDX) assays.In these studies, CD271 þ cells were more tumorigenic than CD271 À cells, and analyses of secondary tumors suggested hierarchical relationships between CD271 À and CD271 þ cells, consistent with a CSC model.Studies of cell lines also found enriched tumorigenicity among slow-cycling fractions of CD271 þ melanoma cells (7) and a functional role for CD271 in disease propagation (8).

Tumor cell preparation
Patient melanomas were obtained with consent under Peter MacCallum Cancer Centre Human Research Ethics Committee protocol 10/02.Tumors were mechanically dissociated with a McIIwain tissue chopper (Mickle Laboratory Engineering).Enzymatic tumor dissociations were performed according to published methods (5,6,9; see Supplementary Methods).Digestion of PDX melanomas and patient melanomas was identical.

Tumorigenesis assays
After sorting, cells were counted and resuspended in staining media with 25% high concentration (HC) Matrigel (BD Biosciences) unless otherwise indicated.Animal experiments were performed under Peter MacCallum Cancer Centre Animal Ethics and Experimentation Committee protocols E421 and E526.Subcutaneous injections were performed in each flank and in the interscapular region of NOD/CB17Prkdcscid Il2rgtm1 Wjl/SzJ (NOD/SCID Il2rg À/À , NSG) mice (Jackson Laboratories).Tumors were evaluated weekly by palpation and caliper measurement.

IHC
Primary antibodies were monoclonal mouse anti-CD133 (gift from Andreas Behren, Olivia Newton-John Cancer Research Institute, Australia), polyclonal rabbit anti-JARID1B (Novus Biologicals, NB100-97821), polyclonal rabbit anti-ABCB5 (Sigma, HPA026975), and monoclonal rabbit anti-ALDH1A1 (Abcam, ab52492), or isotype controls.Sections were heated at 60 C for 30 minutes and dewaxed using a Leica Jung autostainer XL.Antigen retrieval was performed in trisodium citrate buffer at 125 C for 3 minutes, followed by a wash in Tris-buffered saline and 0.1% Tween-20 (TBST).Sections were incubated in 3% peroxidase block for 20 minutes, washed in TBST, and then blocked with 1%-2% horse serum.Primary antibodies were incubated at room temperature for 60 minutes or at 4 C overnight.Sections were washed in TBST, incubated for 45 minutes in ImmPRESS HRP anti-mouse or anti-rabbit Ig (peroxidase) polymer detection (Vector Laboratories, MP-7402), and washed again.Signal detection was performed with 1-2 drops of AEC substrate-chromogen (Dako, K3464), followed by rinsing in water and counter staining with hematoxylin.Sections were cover slipped with Aquatek.Quantification of marker expression was performed by counting cells in random 20Â fields of 800-1,500 cells per field using the cell counter plugin of ImageJ (http://rsb.info.nih.gov/ij/).

DNA/RNA isolation and cDNA synthesis
DNA was extracted from pellets of cells (40,000-1,000,000 cells) purified by flow cytometry (Supplementary Fig. S3).DNA was extracted using DNeasy Blood and Tissue kits (Qiagen).Samples were eluted twice with 20-50 mL of AE buffer.DNA concentration and purity were measured on a NanoDrop2000 UV-Vis spectrophotometer (Thermo Scientific).RNA was extracted from cells using PureLink RNA Mini kits (Ambion).cDNA was synthesized using a single step protocol SuperScript VILO cDNA Synthesis Kit (Invitrogen), with input RNAs ranging from 6-8 ng per reaction.

qRT-PCR
qRT-PCR was performed using Fast SYBR Green Master Mix (Applied Biosystems) on a StepOne Plus Real Time PCR System (Applied Biosystems), using commercial primer kits (Hs_GAPDH_2_SG QuantiTect Primer Assay, Hs_NGFR_1_SG QuantiTect Primer Assay (Qiagen).Data were analyzed using StepOne Plus Real Time LinRegPCR software and normalized to GAPDH expression.

Statistical analysis
Tumor growth rates were determined by maximum tumor diameter (in mm) divided by time elapsed (in weeks) from the date tumors first became palpable.Differences between mean growth rates were compared using unpaired t tests.Non-normally distributed datasets were log-transformed prior to performance of t tests.Tumorigenic cell frequencies were calculated using ELDA (extreme limiting dilution analysis; ref. 15).To account for false negative tumor engraftments due to premature death of mice, mice were excluded that died without any tumors before the mean time from injection to palpability for all tumors in the same experiment.

SNP Genotyping
Genomic DNA (200 ng) was diluted to 50 ng/mL and genotyped on Illumina Beadchip HumanOmni2.5-8 ($2.5M SNP loci) or HumanOmniExpress ($700k SNP loci) arrays in the Australian Genome Research Facility.Data were analyzed using a novel method, the t statistic (see Supplementary Methods).

Detection of CD271 expression in melanoma cells is not affected by trypsin-based tumor dissociation
To investigate effects of tumor dissociation, we dissociated the same melanomas according to three different protocols [Q (9), C (6) and B (5)] used previously to test the CD271-based CSC model (Fig. 1A, Supplementary Fig. S1, and Supplementary Methods).Differences included use in the Q protocol of a brief 0.05% trypsin-EGTA incubation, which was proposed to explain differences between studies (6,14).Patient or PDX melanomas (Supplementary Fig. S2) were minced and evenly divided into three tubes prior to side-by-side dissociation (Fig. 1A).No consistent differences in cell yields were observed (Fig. 1B and C).
If use of trypsin in melanoma dissociation impairs binding to CD271 by anti-CD271 antibodies, as proposed (6,14), then the trypsin-containing Q protocol should yield fewer flow cytometrydetected CD271 þ cells than the C and B protocols (6).We thus labeled cells with antibodies to allow specific gating of human melanoma cells (Supplementary Fig. S3; ref. 10) and CD271expressing cells.Unlike previous studies (6), trypsin-based dissociation did not consistently reduce proportions of CD271 þ cells (Fig. 2A).Consistent with this, no differences in CD271 mRNA levels were detectable in CD271 À cells purified after dissociation via the Q, B, or C protocols (Fig. 2B).These data indicate that melanoma dissociation with low concentration trypsin does not affect the flow cytometry-based detection of CD271 on human melanoma cells.
Robust tumor formation from CD271 À and CD271 þ melanoma cells regardless of tumor dissociation method We next asked whether the method of melanoma dissociation could affect tumorigenic potential in unselected CD271 À or CD271 þ melanoma cells.Unfractionated Lin À cells, CD271 À cells, and CD271 þ cells were isolated by flow cytometry from seven patient melanomas that had been dissociated side-by-side in the Q, C, or B protocols.Sorted cells were mixed in 25% high concentration Matrigel and transplanted subcutaneously into NOD/SCID Il2rg À/À (NSG) mice (Fig. 3A).The frequencies of cells with tumorigenic potential in each isolated cell fraction were calculated using limit dilution analysis (15) (Fig. 3B and Supplementary Fig. S4).
In all experiments, tumor formation was routinely observed from injections of as few as ten Lin À , CD271 À , or CD271 þ cells, regardless of the method of tumor dissociation.Examples of this are shown in Fig. 3A.Across all experiments, the empirically determined frequency of tumorigenic Lin À cells was 1/63 when evaluated following dissociation in the Q protocol, compared with 1/113 and 1/127 for the C and B protocols, respectively (Fig. 3B; c 2 P > 0.05 for all comparisons).For both CD271 À and CD271 þ cells, tumors formed robustly and with similar regularity for all numbers of all phenotypes of injected cells.In some experiments, CD271 þ cells were more likely to form tumors than CD271 À cells (e.g., in melanoma 12-1036 dissociated via B method, tumorigenic cell frequencies were 1/273 in CD271 À cells and 1/149 in CD271 þ cells; c 2 P ¼ 0.5; Supplementary Fig. S4).However, the opposite was observed in other experiments (e.g., in melanoma 12-1036 dissociated via C method, tumorigenic cell frequencies were 1/67 in CD271 À cells and 1/121 in CD271 þ cells; c 2 P ¼ 0.2, Supplementary Fig. S4).Across all transplants, tumor formation was more efficient from cells dissociated by the Q and C methods (tumorigenic cell frequencies 1/71 and 1/119, respectively) than by the B method (tumorigenic cell frequency 1/170; c 2 P < 0.001).Consistent with previous data (10), no differences were observed at necropsy between rates of metastasis in NSG mice with tumors generated from CD271 À or CD271 þ cells (Fig. 3C).

Effects of Matrigel formulation on PDX tumor formation from purified melanoma cells
Matrigel is an extracellular matrix compound that enhances the tumorigenicity of human melanoma cells in NSG mice (9).It contains numerous structural proteins and growth factors that can impact cell biology assays (16)(17)(18).As published assays of CD271-associated melanoma tumorigenesis used different commercially available preparations of Matrigel [standard (5,6) or high concentration (HC; ref. 10)], we next tested the effects of Matrigel formulation on melanoma cell tumorigenicity in PDX assays.
Cells were dissociated from each of three melanomas according to the Q protocol.Flow cytometrically purified Lin À cells were evenly divided prior to side-by-side resuspension into 25% final concentrations of one of four different Matrigel formulations (HC, standard, HC growth factor-reduced, and HC phenol red-free) and injection into NSG mice.Although no differences in tumor formation were noted among cells transplanted with HC, standard, and HC phenol red-free Matrigel, HC growth factor-reduced Matrigel revealed tumorigenic potential in 2-to 3-fold more melanoma cells than the other formulations (Fig. 4A).
To test whether published differences in Matrigel formulation affect tumorigenicity associated with CD271 expression, additional aliquots of the same Lin À , CD271 À , and CD271 þ cells from two of the experiments in Fig. 3B (Q dissociation protocol) were injected after side-by-side resuspension in 30% standard Matrigel (5,6) or 25% high concentration Matrigel (9,10).Tumor formation and the growth rates of tumors were similar regardless of the type of Matrigel used or of the phenotype of injected cells (Fig. 4B; P > 0.2 for all comparisons).

Serial transplantability of CD271 À and CD271 þ cells from PDX melanomas
We also tested whether CD271 expression might be linked to the ability of melanoma cells to renew their malignant potential through serial transplantation.Secondary PDX tumors grown from CD271 À or CD271 þ patient melanoma cells were dissociated (Q method) and cells labeled with anti-CD271 antibody prior to analysis by flow cytometry.Notably and consistent with previous studies (7, 10), CD271 was heterogeneously expressed in secondary tumors regardless of the CD271 phenotype of cells that give rise to them (Fig. 5).
CD271 À and CD271 þ cells were purified from these secondary tumors and transplanted into tertiary recipients.For melanoma 12-1036, tertiary transplantation was unusually inefficient from a secondary tumor derived from CD271 À cells, whereas both CD271 À and CD271 þ cells from a CD271 þ cell-derived secondary tumor were readily passaged (Fig. 5B).However, in another melanoma (13-061), cells from secondary tumors were similarly and efficiently transplantable regardless of their CD271 content and regardless of the CD271 phenotype of cells that produced them.Thus, the serial transplantability of uncultured melanoma cells in these experiments was not consistently linked to CD271 expression.

CD271 reexpression is highly variable in secondary PDX melanomas
Evaluating marker reexpression in secondary tumors is essential for testing the CSC model.In the CD271-based CSC model of melanoma progression (5,6), CD271 þ cells sit atop the hierarchy,  such that CD271 þ cells generate progeny that recapitulate the CD271 heterogeneity of the parental tumor, whereas CD271 À cells produce no or very few CD271 þ cells.As published studies (6, 10) only evaluated CD271 reexpression in small numbers of PDX tumors derived from CD271 À and CD271 þ melanoma cells, with contrasting results, we wondered whether studying more tumors would reveal unexpected variation in CD271 reexpression patterns.
We thus evaluated CD271 expression in most (n ¼ 68) of the PDX tumors shown in Figs. 3 and 4 that were derived from transplantations of Lin À , CD271 À , and CD271 þ cells isolated from two patient melanomas (12-1036 and 12-1254).Although a small proportion of tumors derived from CD271 À and CD271 þ cells had CD271 expression patterns consistent with hierarchical relationships between these cells in the parental melanoma (e.g., tumors ST11 and ST14 in Supplementary Fig. S5), most did not.One tumor derived from CD271 þ cells and one from Lin À cells contained no CD271 þ cells.All other tumors displayed heterogeneity of CD271 expression, regardless of the phenotype of injected cells that generated them (Fig. 6B).
However, in contrast to studies that found stable, plastic equilibration of markers in melanoma (3,10), the proportions of  CD271 þ cells varied widely even among sibling secondary tumors (Fig. 6B).For example, separate injections of 100 CD271 À cells from melanoma 12-1036 formed tumors that contained 54.6% and 6.7% CD271 þ cells, respectively (Fig. 6A, tumors ST3 and ST4).Similarly, separate injections of 1,000 CD271 À cells from melanoma 12-1254 formed tumors that contained 24% and 11% CD271 þ cells, respectively (Supplementary Fig. S5, tumors ST12 and ST13).CD271 expression patterns were clearly different among tumors in each pair, despite tumors being grown from inoculates of the same numbers of cells taken from the same purified cell pools.By immunohistochemical staining, variation (median 6.5-fold, range 1.5-23.0)among sibling PDX tumors was also apparent in the expression of other markers previously linked to the CSC model in melanoma [ALDH (19,20), CD133 (21), and ABCB5 (22)].In contrast, the expression of JARID1B, which plastically regulates the propagation of melanoma cell lines (3), was stable (median difference 1.2-fold, range 1.1-1.4;Fig. 6C and Supplementary Fig. S6).These data confirm that intratumoral relationships between CD271 À and CD271 þ melanoma cell subpopulations are not hierarchical in a manner predicted by the CSC model.However, the striking variability of CD271 expression is not consistent with A B   classical models of cancer cell plasticity that predict reproducible distributions of marker expression.We thus wondered whether highly variable CD271 expression among sibling PDX melanomas could be associated with genetic variation (11,23).

Genetic differences among sibling PDX tumors grown from purified melanoma cells
To test genetic differences linked to CD271 expression in PDX melanomas, we genotyped Lin À cells sorted by flow cytometry from four pairs of sibling tumors (ST3/4, ST6/7, ST10/11, and ST12/13 in Fig. 6A; Supplementary Fig. S5).Tumors in each pair clearly differed in CD271 expression pattern despite being grown from cells obtained from the same pools of CD271-purified cells.Genomic DNA was extracted from Lin À cells from each tumor and subjected to SNP genotyping using Illumina Human OmniExpress 715K arrays.
To evaluate differences in genomic copy number alterations (CNA) between tumors in each pair, we developed a novel approach for comparative SNP array analysis (see Supplementary Methods) to address limitations in published algorithms (24), which we found inaccurately assigned copy number changes in tumor genomes with complex ploidy and heterogeneity characteristics.To compare copy number profile between tumor pairs, we calculated t statistics in 100 SNP  data for all injections of CD271 À , CD271 þ , and Lin À cells at the indicated numbers, following dissociation of patient melanomas  by the Q method and mixture in either 25% HC Matrigel or 30% S Matrigel, as per published studies (5,6,9).Frequencies of tumorigenic cells were not significantly different for any parameter (P > 0.2, c 2 ).Bottom, average fold-change (relative to HC CD271 À tumors) in growth rates of PDX melanomas in the top table (all ¼ Lin À ).Bars, means AE SD (all P values > 0.05, Student t test).
windows across the genome, independently for both log R ratio (LRR) and for a transformed B allele frequency (BAF).Regions of copy number difference were found by applying a min-runmax-gap algorithm (25).Genetic differences among tumors within a pair were summarized as the proportions of tumor genome affected by divergent CNAs.In three control technical replicate assays of DNAs extracted from PDX melanomas, the tstatistic algorithm detected copy number differences of 0%-0.01%(data not shown).
Tumors in each sibling PDX melanoma pair contained clear genetic differences (Fig. 7A and B).For example, tumors ST3/4 differed genetically by 28% and tumors ST12/13 differed genetically by 48% (Fig. 7B).Genetic differences were either CNAs present in one tumor but not detectable in the other [e.g., heterozygous loss of chromosome (chr) 11 in tumor ST3 but not ST4; Fig. 7A, double-headed arrows], or differences in the proportions of genetically distinct subclones present within heteroge-neous tumors (e.g., chr 8 in PDX tumor ST11 was $70% diploid and $30% monoploid, whereas in tumor ST10 chr 8 was $85% diploid and $15% monoploid; Supplementary Fig. S7, middle plots, double-headed arrows).Genetic differences among sibling PDX melanomas were not only due to release of evolutionary bottlenecks in parental tumors by transplantation of small numbers of cells, as they did not correlate with whether tumors were grown from the same or different numbers of inoculated cells (Fig. 7B, bottom).These data reveal a surprising degree of genetic divergence among PDX tumors grown from phenotypically identical patient melanoma cells.

Intratumoral genetic heterogeneity is linked to differences in CD271 expression in melanoma
We next tested whether CD271 À and CD271 þ cells within the same melanomas might be genetically distinct subclones.Genomic DNA from flow cytometrically purified CD271 cells from patient (n ¼ 1) and PDX (n ¼ 4) melanomas underwent SNP genotyping to identify copy number differences using the t statistic method (see above).In 4 of 5 (80%) melanomas, genetic differences (range 1.4%-23%) were apparent between CD271 À and CD271 þ cells in the same melanomas (Fig. 7C-G and Supplementary Fig. S8).For example, in patient melanoma 13-609, a subclone characterized by loss of heterozygosity (LOH) and focal deletions in chr 9 and a chr 19q amplification was identified in CD271 À cells, whereas the complementary CD271 þ population contained a mix of this subclone plus another subclone with LOH and no focal deletions on chr 9, LOH in chr 14, and chr 19 diploidy (Fig. 7C-F).In other melanomas, differences in proportions of subclones were evident between CD271 À and CD271 þ cells.For example, chr 12 in mel 1254 was approximately 8% tetraploid and approximately 92% triploid in CD271 þ cells and approximately 25% tetraploid and approximately 75% triploid in CD271 À cells (Supplementary Fig. S8, bottom right plots, blue arrows).
As the human CD271/p75/NGFR gene locus is on chromosome 17, we looked carefully in this region for copy number differences between CD271 À and CD271 þ cells, finding none.We also looked for divergent copy number changes affecting genes whose products were found to regulate expression of CD271, such as IFNG (26), IFNB1 (27), TNF (28), BMP2 (29), BMP7 (30), EGR2 (31), RUNX2 (32), VDR (33), and FOXO3A (34).None of these CD271-regulatory genes was directly affected by differential copy number between CD271 À and CD271 þ cells.Nonetheless, genetic differences were associated with differences in CD271    expression among uncultured cells or their progeny in most of the patient and PDX melanomas we studied.

Discussion
We found that human melanoma does not follow a CD271based CSC model, regardless of variations in PDX assays.In sideby-side comparisons of published methods for testing the CSC model in melanoma (5,6,9), we found no consistent effect of assay variation on tumorigenesis, which was routinely observed from low numbers of CD271 À and CD271 þ patient melanoma cells (Figs. 3 and 4).The contrasting conclusions by different groups regarding the CD271-based CSC model in melanoma are thus not explained by differences in the processing of cells prior to xenotransplantation, as has been proposed (6,10,14).Regardless of the method of cell preparation, uncultured human CD271 þ melanoma cells do not consistently have more tumorigenic potential than CD271 À melanoma cells.
A definitive explanation for differences among published studies remains obscure.The use of different strains of recipient mice is one potential reason, as CD271 À melanoma cells may be more suppressed or more efficiently cleared, compared with CD271 þ cells, in mice that are less immunocompromised than NSG mice (6).However, this is not relevant to the CSC model, which describes mechanisms for differences in cell-intrinsic tumorigenic potential among cancer cells, rather than differences in cellextrinsic regulation of cancer cell fate (11), including via xenogeneic immune mechanisms.
It is possible that some differences among studies relate to genomic instability in melanoma cells (Fig. 7; ref. 35), as the random acquisition of favorable or deleterious genotypes could promote or impede tumorigenicity in different experiments.Increased tumorigenicity might be linked to increased CD271 expression if a particular set of genetic changes caused both these features.In some experiments we observed, as others have (5,6), that CD271 þ melanoma cells were more tumorigenic than CD271 À cells (Fig. 4B).However, in other experiments the opposite was seen (Fig. 3B).This suggests that genetic changes that increase tumorigenic potential in melanoma can be associated with reduced CD271 expression.Genetic divergence linked to phenotypic and functional differences among malignant subclones has been recognized in other cancers (36)(37)(38).
Although variations in marker expression among sibling PDX tumors could relate to microenvironmental differences that drive varied epigenetic determinants of expression, we identified clear copy number differences (28%-48%) among sibling tumors with distinct CD271 expression profiles.CD271 À cells were genetically distinguishable from CD271 þ cells in 4 of 5 melanomas, containing different and complex mixtures of shared and, in some cases, distinct clones (Fig. 7C-G and Supplementary Fig. S8).Although these intratumoral copy number differences could not simply explain differences in CD271 protein expression as a consequence of varying doses of genes encoding CD271 or its known regulatory proteins, the expression of CD271 and its regulators are likely to be modulated by other factors affected by the dynamic genetic changes we observed to promote or inhibit CD271.These data are consistent with studies of phenotypically distinct breast cancer cells (39,40), in which only small genetic differences were concluded to exist among putative breast CSC populations.As far as we are aware, our study is the first to identify extensive intratumoral genetic differences among uncultured human cancer cells that are phenotypically distinct in a manner previously assumed to be driven by stable epigenetic mechanisms.This raises the likelihood that CD271 expression is modulated by competing, genetically unstable subclones, perhaps via cross-talk (41,42), resulting in widely variable expression patterns.The instability of CD271 expression in melanoma renders it an unattractive therapeutic target.
A key concept underpinning the CSC model is the hierarchical organization in tumors of cells that are phenotypically distinct (43).Consistent with some previous studies (10), but in contrast to others (6), our analyses of CD271 expression in secondary tumors revealed no evidence of hierarchical relationships between CD271 À and CD271 þ melanoma cells.However, in evaluating a large number of secondary PDX tumors, we also observed striking variation in CD271 reexpression patterns (Fig. 6B), even among tumors grown in the same experiments from the same numbers of phenotypically identical cells (Fig. 6A and Supplementary Fig. S5).Indeed, most secondary PDX melanomas had CD271 expression levels more than 50% different from parental tumors (Fig. 6B).Not only are these data inconsistent with a CSC model, they are also not consistent with a classical plasticity model of marker expression, in which reequilibration of marker-defined cell populations occurs in a predictable manner (11,44,45).In vivo marker expression patterns in tumors composed of uncultured cancer cells may be far more variable than they are in highly passaged cell lines or than they appear when only small numbers of PDX tumors are evaluated.
The CSC and plasticity models of cancer progression are most useful if marker expression states are stably linked to malignant potential.If such links are only transiently and variably present during disease progression, then targeting these markers will not substantially benefit patients.Indeed, although epigenetic changes drive a variety of phenotypes and functional states in cancer, if they do so on a profoundly unstable genomic background, their importance will likely be reduced over the course of a patient's disease as genetically distinct subclones, modulated by new epigenetic states, are selected.The presence of substantial intratumoral genomic variation, such as we observed in this study, decreases the usefulness of the CSC and plasticity models for understanding mechanisms of progression in genetically unstable cancers.
Including data from this study and from other studies (5, 6, 10), CD271 expression has been found to be independent of tumorigenicity in most patient melanomas evaluated.Although others found evidence for hierarchical cellular organization in melanoma with other markers (20,22), the genomic instability and variability in marker expression we identified indicate that extended testing of these markers is warranted before they, and the cells they mark, can be considered promising therapeutic targets.To identify reliable and useful markers of cancer cells that might be worth targeting in patients, comprehensive in vivo evaluation is required of the stability and functional associations of marker expression in uncultured cells.

Figure 7 .
Figure 7. Genetic heterogeneity linked to CD271 expression among and within melanomas.A, genomic copy number differences among sibling PDX tumors.Copy number alterations in sibling PDX tumors (ST3 and ST4) grown from separate 100-cell injections of CD271 þ cells from the same patient melanoma (see Fig. 6A).LogR ratio (LRR) data shown to left and B allele frequencies (BAF) to right.t-LRR and t-BAF values indicate differences between the raw LRR and BAF signals.Double-headed arrows, chr 11.Heatmap indicates genomic regions (gray) that differ in copy number between tumors.B, top, heatmaps indicating regions of copy number difference (gray) between three other sibling PDX tumor pairs (ST6 vs. ST7, ST10 vs. ST11, ST12 vs. ST13; see Fig. 6 and Supplementary Fig. 5A).Bottom, quantitation of genomic differences within each sibling PDX tumor pair.Histograms are colored to indicate the cell dose used to establish the tumors in each pair.Values indicate proportional genomic differences in copy number among tumors in a pair (range 28%-48%).C, genomic copy number differences among CD271 À and CD271 þ cells in a patient melanoma.Left, flow cytometry analyses of CD271 expression in patient melanoma 13-609 (top, sorting gates are shown) and in sorted (bottom histograms) CD271 À (gray) and CD271 þ (black) cells submitted for SNP genotyping (right).Right, LRR and BAF plots derived from SNP genotyping of sorted CD271 À and CD271 þ cells from patient melanoma 13-609.t-LRR and t-BAF values indicate differences between raw LRR and BAF signals from each cell population.D-F, LRR (left) and BAF (right) plots indicating copy number differences on chromosomes 9 (D), 14 (E), and 19 (F) between CD271 À and CD271 þ cells from patient melanoma 13-609.G, quantitation of genomic differences among CD271 À and CD271 þ cells in four PDX (404, 12-736, 12-1036, 12-1254) and one patient (13-609) melanoma.Values indicate proportional genomic differences in copy number among the two CD271-defined subpopulations in each melanoma [range (0% to 23%)].

Melanoma Matrigel Number of tumors/Number of injections Cells per injection Melanoma-initiating cell frequency (95% confidence interval) All Melanoma initiating cell frequency (95% confidence interval) Average fold change in growth rate (mm/week) Average fold change in growth rate (mm/week)
Ã , P < 0.05, c 2 ).Bottom left, average fold-change (relative to HC tumors) in growth rates of PDX melanomas in the top table, grouped according to the type of Matrigel use to establish them.Bars, means AE SD (all P values > 0.05, Student t test).Bottom right, tumor formation (within dotted lines) in NSG mice following transplantations of 10 Lin À melanoma cells from patient melanoma 806 after mixing in different types of Matrigel.B, top, summary tumorigenesis À and CD271 þ