Stem Cell Origin of Testicular Seminoma

Abstract

A major question regarding cancer is its cells of origin. The discovery of diverse tumors being derived from a hierarchy of stem cells may pave the way to prove the theory of a stem cell origin of cancers. To our knowledge, we have evidence for the first time in human samples that unique germ cell tumors are linked to particular germline stem cells on the basis of their molecular expression profile. We postulated that distinct malignant subtypes arise from certain stem cells in a stem cell hierarchy.

Introduction

A major question concerning cancer is its cells of origin. We hypothesized that distinct cancer subtypes arise from unique cancer-initiating cells. By performing a microarray meta-analysis of seminomas and spermatogonial stem cells, we investigated a putative cell of origin for seminoma.

Materials and Methods

We obtained published microarray data for 6 human adult germ cell lines, 16 embryonic stem cell lines, 3 normal testicular tissue samples, and 40 seminomas from the Gene Expression Omnibus database. By assessing correlations between various tissue microarrays, we determined the number of transitional events and the distance between seminomas and human spermatogonial stem cells.

Results

Our meta-analysis showed that spermatogonial stem cells correlated similarly with seminoma (95% CI of Spearman ρ, 0.33–0.44) and with normal somatic testicular tissue cells (95% CI, 0.39–0.40), which suggests parallel paths of cellular origins.

Conclusion

Analysis of our results suggests that a unique cancer subtype, namely seminoma, may have originated from an undifferentiated cell with stemness features rather than from a differentiated cell that acquired stemness features.

Introduction

According to the stem cell theory of cancer, distinct cancer types arise from unique cancer-initiating cells with variable “stemness” features in a stem cell hierarchy.^1,2 However, testing the hypotheses related to this theory in the clinics remains a challenge. Perhaps our best chance to validate the stem cell theory of cancer in solid tumors is to test it on germline stem cells and germ cell tumors. After all, germ cells are prototypical stem cells, the forebear of all stem cells. As stem cells, germline stem cells are better defined and easier to obtain than other putative stem cells in solid organs.³ Importantly, germ cell tumors and their corresponding cells of origin may offer us the best opportunity to prove the stem cell theory of cancer.

A prevailing theory envisions clonal evolution of germ cell tumor from seminoma to nonseminoma.⁴ Another theory envisions a malignant precursor that develops into either a seminoma or a nonseminoma.⁵ We propose an alternative model in which discrete precursor cells in a stem cell hierarchy give rise to either a mixed nonseminoma or a pure seminoma.¹ Because a nonseminoma originates from an earlier primordial germ cell (PGC) that resembles an embryonic stem cell (ESC), it tends to express a more heterogeneous phenotype than does a pure seminoma, which is derived from a later gonadal stem cell, or gonocyte.

A stem cell origin of cancer is almost self-evident in germ cell tumors. It is difficult to imagine that a somatic cell could be reprogrammed to become a definitive germ cell in real life; it makes more sense that an embryonic cell would differentiate into germ cells^6,7 and somatic cells. Hence, nonseminomas have a different cell of origin and do not arise from seminomas.^8,9 Nonseminomas may express embryonic (embryonal carcinoma), extraembryonic (choriocarcinoma, yolk sac tumor), somatic (teratoma), and germ cell (seminoma) components, whereas pure seminomas tend to be restricted, expressing only the germ cell phenotype.

We postulated that an aberrant PGC that remained pluripotent¹⁰ like an ESC develops into a nonseminomatous germ cell tumor and a defective gonocyte, similar to a spermatogonial stem cell (SSC), forms a seminoma. Accordingly, we predicted that the genetic signature or expression profile of nonseminomatous tumor resembles that of PGC, whereas that of seminomas ought to match that of SSC. The experimental results may provide direct evidence in relevant human samples that distinct cancer subtypes arise from specific stem cells in a stem cell hierarchy.

Gene expression analysis is a powerful tool for determining the ontology of normal cellular development and disease progression. Hence, from the mutation rate and intrinsic variability within gene expression, one can determine hypothetical relationships of ontologic origins.¹¹ One can also examine similarities between alleles to determine distances from common ancestral genes and allelic-based ontology in a stepwise mutation model.^12,13 To extend this methodology to gene expression data sets, one can assess distances from the correlation coefficient, specifically, defining the distance between genes as 1 − r² (see D’Haeseleer et al¹⁴). Here, we extended these concepts to evaluate the ontologic distances between cells through analysis of the similarities of their gene expression and to explore a possible cell of origin for testicular seminomas.

Materials and Methods

Microarray Meta-analysis

We characterized the number of forward transitional events based on the principle that, as a cell differentiates toward a more committed state, changes in gene expression occur so that the cell becomes more dissimilar from its pluripotent ancestor. This concurs with the increasingly different phenotypes cells express as they differentiate. The amount of similarity can be measured statistically by the correlation between different cell types’ overall gene expression. Because cells become more dissimilar as they differentiate, the correlation is inversely related to the number of forward transitional steps between cells in the hierarchy.

For our analyses, we first searched for data from tissue microarray samples in the Gene Expression Omnibus database by using the key words “seminoma” and “spermatogonial stem cells” filtered by “species” for Homo sapiens. We identified the GSE11350,¹⁵ GSE3921,¹⁶ and GSE8607¹⁷ data sets, which had the various samples that we needed plus a substantial number of matching probes. These data sets provided microarray data for 6 human adult germ cell lines or SCCs, 16 ESC lines, 3 normal testes, and 40 seminomas.

We made several comparisons: the 40 seminomas vs. 6 of the SSCs; the 3 normal testes vs. 6 of the SSCs; and the 3 normal testes vs. the 40 seminomas (Table 1). For each comparison, if the microarray platforms were different, we then compared the data by using the subset of probes that had the same gene accession identification numbers in the different platforms. Hence, we also performed control comparisons between 13 of the seminomas and the 16 ESCs, and between 6 of the SSCs and 3 of the ESCs (Table 1). To collate the data across platforms, we used Microsoft Access 2003 software (Microsoft Corp, Redmond, WA).

Table 1.

Summary of Data and Comparisons of Microarray Meta-Analysis

	Comparison
Variable	ESCs	SSCs	Seminomas	ESCs	Seminomas	NTTs	Seminomas	SSCs	NTTs	SSCs
Data seta	GSE11350 − GPL570		GSE3921 − GPL3038 + GPL2932	GSE921 − GPL1283	GSE8607 − GPL8300		GSE8607 − GPL8300	GSE11350 − GPL570	GSE8607 − GPL8300	GSE11350 − GPL570
No. specimens	3	6	9 (GPL3038) + 4 (GPL2932)	16	40	3	40	6	3	6
No. probes	54,675		41,088 (GPL3038) + 39,522 (GPL2932)	43,008	12,624		12,624	54,675	12,624	54,675
No. common probes	54,675		3,389		12,624		30,760		30,760
No. correlation coefficient comparisons	18		208		120		240		18
Mean (95% CI of Spearman correlation coefficient [ρ])	0.91 (0.89–0.93)		0.14 (30.05 to 0.33)		0.77 (0.71–0.83)		0.38 (0.33–0.44)		0.39 (0.387–0.40)

Abbreviations: ESC = embryonic stem cell; NTT = normal testicular tissue; SSC = spermatogonial stem cell.

^aGene Expression Omnibus data set identification number.

Gene Expression Comparisons by Using Archived Microarray Data (Gene Expression Omnibus Database)

For each comparison between cell types, we opted to find the genes that matched between sample arrays because these samples had been run on various platforms. We chose combinations of arrays on the basis of maximizing the number of genes tested. To match genes, we compared GenBank accession numbers in Microsoft Access and put the matched copies into a separate file.

We used the Spearman rho (ρ) nonparametric correlation coefficient testing (IBM SPSS statistics software version 19; SPSS Inc, Chicago, IL) for analyzing the comparisons because some data sets were normalized and others were not, which meant that direct comparisons of gene expression levels could not be done. To find a correlation, we compared paired samples in each comparison category. In each pair, the corresponding probes were ranked according to their relative expression within the sample. The 2 sets of ranked lists were then compared to see how similar they were, which resulted in correlation. Every possible pair for each comparison was then tested. Any null values in any of the data sets were set to zero for ranking. These constituted a small percentage of the total data available.

Seminomas vs. SSCs

We first compared seminomas with SSC samples. This involved matching genes between GSE8607-GPL8300¹⁷ and GSE11350-GPL570.¹⁵ The matched list contained 30,760 probes. We believe that this number, which is actually higher than one of the initial probe sets, is due to the replication that occurred in matching. We performed the Spearman ρ nonparametric correlation coefficient tests between each pair of samples in the 2 categories.

Seminomas vs. Normal Testes

A second comparison was made between seminomas and normal testes. Two data sets contained samples for comparison, but there were insufficient matching probes between the 2 subsets for comparison. Within the GSE8607 data set¹⁷ were 40 samples of seminoma and 3 samples of normal testis, with an initial probe set of 12,625. Because these were from the same data set, all the probes were used, and the Spearman ρ tests were performed between each pair of samples in the 2 categories.

Normal Testes vs. SSCs

Our third comparison, between normal testes and the SSC samples, involved matching genes between the GSE8607¹⁷ and GSE11350¹⁵ data sets. The matched list contained 30,760 probes. We believe that this number is also due to the replication that occurs in matching. The Spearman ρ tests were performed between each pair of samples in the 2 categories.

Seminomas vs. ESCs

To determine whether the relationship between seminomas and SSCs is the result of the greater differentiation of SSCs, we also compared seminomas with ESCs. Both sample types came from the same data set, GSE3921,¹⁶ although different microarray platforms had been used. Within the seminoma data set, GPL3038 had 9 samples of seminoma, with an initial probe set of 41,088, and GPL2932 had 4 samples of seminoma, with an initial probe set of 39,522. These platforms were first collated by matching probes and then were compared with the ESCs. The ESCs had been assessed on platform GPL1283, which had 16 samples of ESCs, with an initial probe set of 43,008. These platforms were then matched, which yielded a set of 3,389 probes. In total, 208 Spearman ρ tests were performed between each pair of samples in the 2 categories.

SSCs vs. ESCs

Our fifth comparison was between SSCs and ESCs. Both sets of samples came from the data set GSE11350.¹⁵ The comparison involved all 54,675 probes, for a total of 18 Spearman ρ tests.

Results

We used correlations as a guide for determining the number of transitional events and the distance between cells in a stem cell hierarchy.^11–14 We defined the similarity between gene expression of different cells as a marker of the number of transitional events between cells of interest in a stem cell hierarchy. The results of our various comparisons are summarized in Table 1. The correlations and distributions between the seminomas and SSCs and between the normal testes and SSCs are illustrated in Figure 1.

Figure 1.

Histograms, Illustrating the Correlations Between Tissue Types by Using Spearman ρ Correlation Coefficient Testing of Microarray Data. (A) Seminomas vs. Spermatogonial Stem Cells (SSC). (B) Normal Testicular Tissue vs. SSCs. Both are Relatively Tightly Distributed, but Note That the Correlation Between the Seminomas and SSCs (A) Is Skewed to the Left

Our analyses showed that seminomas correlated with SSCs: (95% CI of Spearman ρ, 0.33–0.44), which is skewed to the left (Figure 1A), and that normal testes correlated with SSCs: (95% CI, 0.39–0.40) (Figure 1B). In our control comparisons, seminomas did not correlate as well with ESCs: (95% CI, −0.05–0.33). Also, as expected, SSCs correlated with ESCs: (95% CI, 0.89–0.93), and seminomas correlated with normal testes: (95% CI, 0.71–0.83).

Discussion

A major question concerning cancer is its cells of origin. We hypothesized that distinct cancer subtypes arise from unique cancer-initiating cells with variable stemness features in a stem cell hierarchy.^1,2 Here we show evidence in relevant human samples that suggests that a unique cancer subtype, namely seminoma, originates from an undifferentiated cell with stemness features rather than from a differentiated cell that acquires stemness features.

According to the stem cell theory of cancers, tumors derived from distinct cells of origin are inherently different from one another and are inclined to pursue disparate clinical courses.^1,2 Hence, tumors derived from stem cells higher in the stem cell hierarchy tend to metastasize more readily and widely than do those from stem cells lower in the hierarchy. They also tend to express a more heterogeneous or mixed phenotype. This hypothesis accounts for the puzzling fact that certain tumors are intractably lethal, whereas others are surprisingly indolent.

Germ cell tumors offer many advantages in our quest to show that the tumorigenic potential of different cancer subtypes arises from their respective stem cells of origin. For instance, the biologic, histologic, and clinical aspects of the different germ cell tumor subtypes are already well known.¹⁸ The technical know-how for the identification and culture of various stem cell lineages in a germ cell hierarchy is also more established. Another advantage of using human germline stem cells, such as SSCs, is that these cells are more “benign” and less “challenged” than are those obtained by genetic manipulation or other means.

By using correlations as a guide for determining the number of transitional events and the distance between cells in a stem cell hierarchy, we have obtained evidence for the first time that shows that a distinct cancer subtype may arise from a specific stem cell. Because SSCs correlated well with ESCs (95% CI, 0.89–0.93), and seminomas correlated better with SSCs (95% CI, 0.33–0.44) than with ESCs (95% CI, −0.05–0.33), we reasoned that seminoma is more related to SSC than it is to ESC. An alternative explanation is that seminomas arise from normal testes, because the correlation between seminomas and normal testes is high (95% CI, 0.71–0.83). But if we consider that the total life span from spermatogonia to spermatozoa is short, ie, less than 3 months,¹⁹ then it seems rather improbable for seminomas to arise from these transient “differentiating” or “differentiated” germ cells. Analysis of our data suggests that it is more plausible for a tumor like seminoma to derive from an undifferentiated cell with stemness features than from a differentiated cell that becomes “de-differentiated.”

We emphasize that the cellular origin of a cancer does not necessarily correlate with its histologic features.²⁰ Just because the bulk of a tumor is composed of predominantly differentiated cells does not mean that the tumor is derived from those cells. Although our data demonstrate that seminomas have a close correlation with normal testes, a finding that appears to support the conventional wisdom that seminomas originate from normal testes, we propose that the correlation may actually support an alternative concept, that a late gonadocyte (such as SSC) could differentiate into either normal testis (95% CI, 0.39–0.44) or a malignant seminoma (95% CI, 0.33–0.44) in parallel pathways (Figure 2). In other words, the similar gene expression profiles of seminomas and normal testes could be attributed to their having a common cellular origin rather than to one tissue being derived from the other.

Figure 2.

Two Theories of Carcinogenesis. (Upper Panel) The Stem Cell Theory. (Lower Panel) The Conventional Multistep De-differentiation Theory

Abbreviations: ESC = embryonic stem cell; SSC = spermatogonial stem cell.

Furthermore, Zhao et al²¹ demonstrated that tumors derived from somatic cells do not form easily, if at all. Although there is abundant evidence that shows that induced pluripotent stem cells can be reprogrammed from somatic cells with stemness factors, induced pluripotent stem cells are still distinct and different from ESCs. Hence, when tumors acquire stem cell–like phenotypes and “de-differentiate,” such as those derived from induced pluripotent stem cells, they are immunologically recognized and rejected in syngeneic mice, ie, they undergo “spontaneous remission,” unlike those tumors derived from ESCs. We inferred from the experiments of Zhao et al²¹ that either tumors do not arise from somatic cells at all or, if they do, they constitute unique subtypes of cancer that are inherently more indolent and potentially more treatable.

There is evidence that indicates that SSC and seminoma (c-kit⁺, Sohlh1⁺) may be distinct and different from ESC and embryonal carcinoma (c-kit⁻, Sohlh1⁻).³ In other words, c-kit⁺ cells may be a marker of late germ cell stem cells. For example, in neonatal mice, gonocytes and prespermatogonia with an Oct4⁻EGFP⁺/c-kit⁻ phenotype had a greater repopulation capacity than Oct4⁻EGFP⁺/c-kit⁺ cell fractions had.²² Further, c-kit expression was the only difference between human ESCs (c-kit⁻) and adult gonadal stem cells (c-kit⁺).¹⁵ In addition, seminomas have been reported to express the gp230 glycoform of a tumor-specific glycoprotein that is associated with terminal differentiation, whereas nonseminomas express gp200, an embryonic version of the same glycoprotein.²³

An important caveat from this study: In testing any hypotheses about the origin of human cancers, it is necessary to use human tissue samples. Otherwise, the clinical relevance of the results is questionable. For instance, although the most common subtype of germ cell tumor is seminoma, very few pertinent cell lines or animal models are available for studying its clinical or biologic significance. This exemplifies a potential dilemma in cancer research: Malignant cells that are easy to grow in the laboratory may be derived from earlier stem cells and have a complex phenotype, whereas those that originate from later stem cells have a simpler phenotype but may not be amenable for research. Seminoma provides an unusual balance between complexity and feasibility, which makes this study not only unique but also pertinent.

We realize that several factors could have mitigated the results of this study. For instance, no specific or exclusive marker for SSCs is known, so SSC cultures (like their malignant counterparts) are likely to be mixed cell populations with various degrees of differentiation. It is plausible that another unique gonocyte (close to SSC in lineage but not SSC itself) could be the actual cell of origin for seminomas.^24,25 Similarly, we used ESCs to represent PGCs because both are pluripotent, and some ESCs have the potential to express germ cell phenotypes.^6,7 Nevertheless, we anticipate that future discoveries and the availability of more-specific gonadal germ cells would validate the clinical and biologic implications of this study.

Conclusion

We investigated a putative cell of origin for seminoma. Our meta-analysis showed that SSC correlated similarly with seminoma and with normal testis, which suggests parallel paths of cellular origins. Analysis of our results indicate that a unique cancer subtype, namely seminoma, may have originated from an undifferentiated cell with stemness features rather than from a differentiated cell that acquires stemness features.

Clinical Practice Points.

• Recently, the idea of a stem cell origin of cancers has been garnering increasing attention. Although the study of cancer stem cells is common, their relevance in the clinic still needs to be addressed. Therefore, testing any hypotheses related to the stem cell theory of cancer in human samples is both pertinent and imperative.

• We believe that germ cell tumors offer many advantages in our quest to show that the tumorigenic potential of different cancer subtypes arises from their respective cancer-initiating cells with stem cell features.

• Our work has shown that spermatogonial stem cells and seminomas have similar molecular signatures, which suggests that specific tumors are derived from their respective stem cells in a stem cell hierarchy. To our knowledge, this is the first evidence in relevant human samples that unique cancer subtypes arise from certain stem cells.