Introduction

Mutations of the DNA polymerase ε gene (POLE) are an uncommon event in colorectal carcinomas, but their effects are, in many instances, profound and remarkable. Indeed, whole exome sequencing (WES) of 224 colorectal carcinomas by the TCGA project identified seven microsatellite-stable tumors that harbored POLE single nucleotide variants (3.1%) and carried a very high load of somatic gene mutations, in some cases even exceeding that of microsatellite-unstable tumors [1]. Based on these findings, the concept of the POLE mutated hypermutator-type of colorectal carcinoma as a separate molecular class was proposed, soon to be reinforced by similar observations reported for endometrial carcinoma [2].

In humans, in addition to the DNA POLE there is known the DNA polymerase δ which is encoded by the POLD1 gene. Both DNA polymerases are organized into two functionally distinct and evolutionary conserved domains, one that serves to incorporate nucleotides into the growing DNA strand (catalytic domain) and another that excises and substitutes correct nucleotides where infidelity in base-pairing has occurred (exonuclease domain). POLE single nucleotide variants associated with the hypermutator-type most often reside in the exonuclease domain (reviewed in [3]). However, a similar pattern so far has not been observed in sporadic colorectal carcinomas for POLD1.

Recognition of POLE mutated hypermutator-type colorectal and endometrial cancer has generated great interest among clinicians and researchers alike for two reasons. First, the high load of somatic mutations is expected to result in a wealth of neoantigens which, similiarly to microsatellite-unstable tumors, could make these tumors "immunoreactive" and, thus, amenable to immune-checkpoint inhibitor therapy, one of today's hot topics in oncology. Second, while in the TCGA series the POLE mutations appeared to be somatic events, only a year later a study by Palles et al. [4] of patients with early-onset colorectal carcinoma or intestinal polyposis demonstrated POLE or POLD1 germline mutations in some patients. This, of course, has considerable implications for screening strategies of younger patients afflicted with either of these conditions, for genetic counseling, and for clinical approaches to molecular diagnostics.

Although the hypermutator class of colorectal carcinoma is well established by now some issues remain with the concept because biochemical analyses of exonuclease activity of POLE and POLD1 single nucleotide variants as well as studies in model cell systems indicate a far less than straightforward mutation-function relationship [3]. This may mean that, in clinical specimens, the effects different POLE or POLD1 mutations have on tumors may vary, and it provides a valid incentive to continue systematic DNA polymerase gene sequencing and detailed study of mutated cases. Information on these tumors is still limited and each new case is of value if underpinned by a complete array of clinical, morphological, and molecular data, tumor mutational burden (TMB), in particular. Furthermore, while it is generally assumed that POLE or POLD1 needs to be compromised in one allele only to produce the hypermutator state, this has not been formally shown with clinical specimens because all published studies are based on tumor material that contained stroma alongside with tumor cells, which is a drawback to allelotyping analyses; and gene promoter methylation as a potential “second hit” has not at all systematically been assayed for.

This study was conceived to address the above issues. Specifically, we screened an unselected series of 271 colorectal carcinomas and investigated if the POLE and POLD1 mutations found by this approach (i) belonged to clinicopathologically distinct tumors, those with an “immunoreactive” phenotype, in particular; (ii) conferred the hypermutator status on their tumor of origin; and (iii) were accompanied by a “hit” on the second allele of the mutated gene.

Materials and methods

Case selection and morphological studies

Screening for POLE and POLD1 mutations was done on a series of 271 colorectal carcinomas selected from the Rostock branch of the North German Tumor Bank of Colorectal Cancer [5]. Patients’ written informed consent was obtained as part of the tumor banking procedures, and studies on the archived tumor material were approved by internal board review (Ethics Committee of Rostock University, ref. II HV 43/2004). Routine pathological work-up of the resection specimens and basic molecular studies of the tumors (Bethesda panel microsatellite analyses and methylation analyses) were done as previously detailed [6]. This cohort for POLE and POLD1 mutation screening included 129 women and 142 men (mean age 68.07 years; median age 69 years; range 21–98 years; 14 patients younger than 45 years); for ten patients Lynch syndrome was diagnosed eventually.

The archived slides of the POLE and POLD1 mutated tumors were reviewed to assess if there were morphological peculiarities that distinguished them from standard-type colorectal adenocarcinomas. Looking for the “immunoreactive” phenotype [7], CD8 immunohistochemistry (clone C8/144B, Dako, Glostrup, Denmark) was done, and densities of tumor infiltrating T cells were evaluated using QuPath [8] on digitized slides. For PD-L1 immunohistochemistry the E1L3N antibody (CST, Leiden, NL) was applied at 1:50 dilution after heat-induced antigen retrieval.

Clinical information on the patients was obtained by review of the charts of the Clinical Cancer Registry, University of Rostock.

POLE and POLD1 gene sequencing

DNA for POLE and POLD1 sequencing was obtained by stereomicroscopically controlled microdissection of areas with viable tumor (>50%) from deparaffinised sections using a sterile scalpel blade. DNA was extracted as described in Hühns et al. [9]. Alternatively, snap-frozen samples that had been stored at −80 °C were used and DNA was extracted as published [6]. If snap-frozen tumor tissue was used, cryostat sections were prepared to ascertain that the tumors were adequately represented by the sample (tumor content >50%). Normal DNA was from paraffin-blocks or snap-frozen samples of uninvolved colorectal mucosa at least 5 cm away from the tumors. DNA content was measured using the QuantiFluor® ONE dsDNA System (Promega, Madison, USA). Supplementary Table 1 lists which type of tissue (snap-frozen vs. paraffin-embedded) went into the different molecular studies.

Mutation screening of the exonuclease coding regions of POLE (exons 9–14) and POLD1 (exons 7–10, 13) was performed by PCR amplification and Sanger sequencing. The primer sequences and PCR conditions are detailed in the Supplement (Supplementary Table 2). The sequence data were compared with reference sequences (POLE: ENSG00000177084, POLD1: ENSG00000062822) using SeqScape Software v2.7 (Applied Biosystems, Darmstadt, Germany).

Whole exome sequencing, somatic mutation analysis, and the assessment of tumor mutational burden

WES was done when screening by Sanger sequencing demonstrated a POLE or POLD1 mutation. Except for tumor ID-T8, these studies were done with DNA extracted from tumor areas microdissected from paraffin sections as described above, and DNA was extracted using the GeneRead DNA FFPE Kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions. For tumor ID-T8, snap-frozen tumor tissue was used, after preparing cryostat scout sections as described above DNA was extracted as published [6].

WES was performed as described previously [10]. Briefly, the SureSelect Human All Exon kit (Agilent, Santa Clara, USA) was used for target enrichment, and a HiSeq 4000 (Illumina, San Diego, USA) instrument for high-throughput DNA sequencing. Raw sequencing reads were converted to standard fastq format using bcl2fastq software 2.17.1.14, and fed to an in-house developed pipeline for the analysis of WES data that is based on the 1000 Genomes Project data analysis pipeline and GATK best practice recommendations. The short-reads were aligned to the GRCh37 (hg19) build of the human reference genome with bwa-mem with an average coverage of 245x and 95% of bases covered with at least 20 reads. Variant calling was performed using three different variant callers (GATK HaplotypeCaller, freebayes, and samtools).

Somatic variants were called from WES vcf files using the Sentieon-Genomics pipeline (sentieon-genomics-201808.03) which identified on average 12,000 high confidence somatic variants per sample. TMB was then calculated from high confidence somatic mutations which were located in exons and had a tumor allele frequency of 5% or greater. TMB levels were divided into three groups: low (<12 mutations/Mbp), high (>12 mutations/Mbp), and hypermutated (>100 mutations/Mbp).

POLE and POLD1 resequencing and allelotyping on DNA from laser-capture microdissected tumor samples

For laser-capture microdissection, 5 µm of cryostat or paraffin sections were mounted on polyethylene naphthalate (PEN) membrane-coated slides (MembraneSlide 1.0 PEN; Carl Zeiss Microscopy GmbH, Jena, Germany), fixed for 2 min in 70% ethanol and stained in hemalum for 15 s. Subsequently, the slides were rinsed in water for 2 min and dehydrated in 70%, 96 and 100% ethanol, each for a minute. After air-drying, slides were stored at −20 °C until use. Laser-capture microdissection was performed on a Zeiss Axio Observer (Carl Zeiss) using the RoboLPC method as implemented with the device to microdissect and capture the areas of interest. Tumor samples were collected into 0.5 ml adhesive caps and DNA was isolated using the QIAamp DNA Micro Kit (Qiagen) according to the manufacturer’s instructions.

First, the DNA was used for resequencing the POLE or POLD1 mutations seen by screening. Second, allelic imbalances (AI) were assayed for with polymorphic microsatellite markers that map to chromosomal regions close to POLE (D12S357, D12S1638, D12S1723) or POLD1 (D19S246, D19S585, D19S601, D19S867, D19S904, D19S907) as described in Church et al. [11]. Tumor and nontumor DNA was amplified by PCR as follows: reaction were started at 95 °C for 1 min and this was followed by 30 cycles at 95 °C for 15 s, 52 °C (only for D12S1638, D12S1723) or 55 °C for 15 s, 72 °C for 10 s. PCR products were separated by capillary gel electrophoresis and detected on an automated ABI 3500 Genetic Analyzer (Applied Biosystems). Fragment length and fluorescence intensity were evaluated by Gen′eMapper software (Applied Biosystems). The GeneScan™-500LIZ™ size marker (Applied Biosystems) served as an internal standard. AI was scored positive if the tumor/normal ratio was <0.5 or >2.0.

Methylation analyses

These studies were done with DNA extracted from tumor that was microdissected from paraffin sections as described above. Bisulfite conversion was carried out using the Epitect Bisulfite Kit (Qiagen) according to the manufacturer’s recommendations. CpG promoter methylation analyses of POLE and POLD1 was performed with a Q24 Pyromark instrument (Qiagen). A novel Pyromark CpG assay was developed for the POLE promoter with the Pyromark Assay Design 2.0 software (Qiagen). The assay addressed methylation in six CpG sites within the POLE promoter (chr12: 132687510-132687530) by PCR as follow: forward primer: 5′-GGTAGGGATAGGGGAAAGTG-3′, reverse primer: 5′(5-biotin labeled)-ACCCACCCCACCTCAAACTAACC-3′, sequencing primer: 5′GGAAAAGTGTGTGGTAG-3′. The Pyromark Hs_POLD1_01_PM PyroMark CpG assay (Qiagen) was used for methylation analyses of the POLD1 promoter according to the manufacturer’s recommendations. Target CpGs were evaluated by Pyromark Q24 2.0.7 software (Qiagen), which converts the pyrograms to numerical values for peak heights and calculates the proportion of methylation. Methylation values <5% were considered as background. Hypermethylated DNA provided by the manufacturer was used as standard control.

Results

POLE and POLD1 mutations detected in this series

An unselected series of 271 primary colorectal carcinomas was screened by Sanger sequencing for POLE mutations in exons 9–14 and POLD1 mutations in exons 7–10 and exon 13, as detailed in “Materials and methods” section.

By this screening approach, we identified six tumors with nonsynonymous POLE variants (Table 1). Specifically, we found the single nucleotide variants p.P286R (tumors ID-T6 and ID-T2), p.D287E (tumor ID-T3), p.V411L (tumors ID-T5 and ID-T1), and p.A465T (tumor ID-T4), which are all on record in the COSMIC database or have been reported previously in the literature [12]. All mutations were located in the exonuclease domain of the polymerase (Fig. 1), and for all of them a significant effect on protein function was predicted by in silico algorithms (PolyPhen, MutationTaster, PROVEAN, SIFT; Table 2). The variants p.P286R, p.V411L, and p.A465T in the tumors were somatic observations, as assaying DNA extracted from normal mucosa demonstrated wildtype genotypes. However, the p.D287E variant was also found in the mucosa specimen and in nontumor liver tissue from a resection performed later on, and thus apparently was present in the patient’s germline. All these POLE variants were recovered by WES, which in four of the six tumor samples also yielded a total of seven additional single nucleotide variations outside the exonuclease domain (see Fig. 1 and Table 1).

Table 1 Complete list of clinical, morphological, and molecular data of colorectal carcinomas detected with POLE or POLD1 mutations.
Full size table
Fig. 1: “Lollipop” graphic of POLE and POLD1 missense variants.
figure 1

Variants found by Sanger sequencing are in red letters, variants found in addition by WES are in blue.

Full size image
Table 2 Results from in silico algorithm testing of the nonsynonomous POLE and POLD1 missense variants detected in this study.
Full size table

By Sanger sequencing of POLD1, three tumors with variants were found as follows (Fig. 1 and Table 1): a single nucleotide variant p.E279K (in tumor ID-T7), a variant p.L342M (in tumor ID-T8), and a variant p.R525Q (in tumor ID-T9; details in Table 2). All these variants were somatic and were located in the exonuclease domain (Fig. 1). However, in silico predictions as to their functional effects were mixed (Table 2): the p.E279K variant was classified as nonsignificant by all four algorithms, whereas a compromise of gene function was predicted for the variants p.L324M and p.R525Q by one and two of the four algorithms, respectively. WES of the cases with POLD1 variants confirmed the p.L342M in tumor ID-T8. However, WES of tumors ID-T7 and ID-T9 led to two unexpected observations: First, in tumor ID-T7, the variant p.E279K mutation was not recovered. To clarify this discrepancy, Sanger resequencing was done with the DNA used for the initial screen as well as with the DNA used for WES. Resequencing demonstrated the p.E279K in the initial screening sample, but not in the sample used for WES, which indicates a heterogeneous distribution of this variant within this tumor. Second, in tumor ID-T9, the POLD1 p.R525Q variant was not recovered by WES, which, however, demonstrated a POLE p.V411L variant that had not been picked up by Sanger sequencing. To exclude technical issues, Sanger resequencing was done with the DNA sample used for WES, and this clearly demonstrated the POLE p.V411L whereas the POLD1 p.R525Q was absent from this sample. Thus, POLE and POLD1 mutations in tumor ID-T9 appeared to be heterogeneous.

Clinical features of patients and morphological features of POLE and POLD1 mutated colorectal carcinomas

Information on clinical characteristics of the patients with POLE or POLD1 mutated tumors as well as morphologic and molecular features are detailed in Table 1. Notably, two patients’ tumors were microsatellite-unstable, one with a somatic POLE p.A465T variant and one with a somatic POLD1 p.L342M variant (tumors ID-T4 and ID-T8 in Table 1). Eventually, Lynch syndrome could be demonstrated in these patients because the tumors were seen to lack expression of the mismatch repair proteins MSH6 or MLH1 by immunohistochemistry, and germline mutations of these genes were found in the course of further genetic work-up of the patients (MSH6: p.K1140Q and MLH1: pL296ter, respectively). Histologically, both tumors were heterogenous in their composition, areas with mucinous patterns alternating with solid to medullary areas and numerous stromal and intraepithelial T cells, which, overall, is the classical morphology of Lynch syndrome-associated colorectal carcinomas. A minority of tumor cells (<1%) was PD-L1 positive in ID-T4 whereas ID-T8 remained negative.

The remaining tumors were microsatellite-stable. Of these, two POLE mutated tumors were histologically unusual: ID-T1 and ID-T2 were composed of large solid nests or strands of tumors cells with considerable nuclear anaplasia; the tumor borders were “pushing” and tumor budding and the stromal desmoplasia commonly observed in colorectal carcinomas were inconspicuous. In ID-T1, nuclear anaplasia was quite striking (Fig. 2). By CD8 immunohistochemistry, a dense T-cell infiltrate was seen in ID-T1. Tumor cells were PD-L1 negative by immunohistochemistry, although immunostaining of peritumoral histiocytes was seen in all cases.

Fig. 2: Microphotographs of tumor ID-T1.
figure 2

This poorly differentiated colorectal carcinoma had a somatic POLE variant p.V411L. Note the “pushing” tumor border (arrows in a) and the anaplastic tumor cell nuclei (asterisks in b). Infiltration by CD8+ T cells was prominent (c).

Full size image

ID-T9, the tumor with the heterozygous POLE p.V411L and POLD1 p.R525Q variants, was highly unusual: histologically, it was a sarcomatoid cancer (Fig. 3), which is an exceptional phenotype for a primary colorectal carcinoma, and, equally unusual, the patient was only 21 years old. Tumor infiltrating T cells were sparse in this cancer, and PD-L1 was not expressed. The rest of the tumors (ID-T3, ID-T5, ID-T6 with POLE, and ID-T7 with POLD1 variants) were moderately differentiated, non-mucinous, standard-type colorectal adenocarcinomas. T cell infiltration was prominent in all these tumors, but PD-L1 was not expressed by the tumor cells.

Fig. 3: Microphotographs of tumor ID-T9.
figure 3

This highly unusual colorectal carcinoma from a 21-year-old nonsyndromic patient harbored simultaneous, albeit hetereogeneous, a somatic POLE variant p.V411L and a somatic POLD1 p.R525Q. Note the almost diffuse mode of infiltration (a) and the sarcomatoid features of the cancer cells (b). CK20 was positive by immunohistochemistry (c), as was CDX2 (not shown).

Full size image

Association of POLE and POLD1 gene mutations with tumor mutational burden

Remarkably high mutational rates could be demonstrated by WES in five of the seven tumors with POLE mutations (Table 1). In these, mutational rates ranged from 126 mutations/Mbp of genomic DNA to 311 mutations/Mbp, which puts these tumors well into the class of “hypermutators” (ID-T1, ID-T2, ID-T4, ID-T5, and ID-T6 of Table 1). However, mutational rates were low in ID-T3, the germline POLE variant p.D287E carrier’s tumor, and in ID-T9, the tumor with the heterozygous POLE p.V411L and POLD1 variant p.R525Q. The TMB was also low in tumor ID-T7 which harbored the heterogeneous POLD1 variant p.E279K. A high mutational rate of 32.4 mutations/Mbp was recorded for the third POLD1 mutated tumor (ID-T8), one of the two Lynch syndrome-associated cancers.

Resequencing and allelotyping on tumor DNA obtained by laser-capture microdissection and promoter methylation studies

To exclude the confounding effect of nontumor DNA from the analyses, resequencing and allelotyping was made with DNA that was extracted from tumor cells isolated by laser-capture microdissection. These analyses could be completed successfully for tumors ID-T1 through to ID-T8, but ID-T9, the sarcomatoid carcinoma, due to its growth pattern was not amenable to laser-capture microdissection. First, we used the DNA to resequence the POLE or POLD1 variants, respectively. We observed that all variants were heteroallelic (Fig. 4). Second, we addressed if the POLE or POLD1 gene variants were combined with loss of the second allele. PCR-based allelotyping studies with polymorphic microsatellite markers were informative with at least some markers in all our cases (Table 1). Allelic loss, which on laser-capture microdissected material would have resulted in complete loss of one allele in the electropherograms, was not seen in any of the cases. There was seen a moderate allelic imbalance with one of six markers for the POLD1 mutated tumor ID-T7, but not for the remaining five markers (Table 1). In tumor ID-T8, as can be expected for a Lynch syndrome-associated cancer, microsatellite instability interfered with allelotyping in three of the six markers, but the remaining assays were informative and without indication of allelic loss.

Fig. 4
figure 4

Examples of Sanger sequencing on DNA obtained by laser-capture microdissections from three tumors with POLE or POLD1 mutations.

Full size image

Finally, POLE or POLD1 promoter methylation studies were done, which did not demonstrate methylation in any of our cases (Table 1).

Discussion

Colorectal carcinomas with POLE or POLD1 mutations have been recognized not very long ago as a separate molecular class in which the extremely high numbers of gene mutations (the hypermutator molecular phenotype) are a salient and defining feature [1, 13]. Colorectal carcinomas were first in this concept, endometrial carcinoma to be included not much later, and by now similar observations have been reported for other types of cancer too [14]. However, hypermutator-type colorectal carcinomas are rare. Finding them requires technology quite beyond what currently is used for standard pathological work-up of surgical pathology specimens. Therefore, published information on them is still limited. It remains a topic of interest in surgical pathology how POLE or POLD1 mutated colorectal carcinomas are set off from the rest by clinical, morphological, and/or molecular features.

In this study, we screened colorectal carcinomas obtained by surgery from a consecutive series of unselected patients for POLE or POLD1 mutations in the exonuclease domains. Screening 271 tumors yielded six POLE mutated and three POLD1 mutated tumors (2.2% and 1.1% of the series, respectively), an overall frequency similar to that observed in previous series [1, 14].

A somatic POLE variant p.P286R was found in two of our cases (ID-T6 and ID-T2). Both were microsatellite-stable and carried a very high load of mutations (TMB of 275 and 303, respectively). The POLE variant p.P286R is one of the by now well-appreciated driver mutations of the POLE gene. Indeed, features of POLE p.P286R mutated colorectal carcinomas of the hypermutator-type have recently been reported by Ahn et al. [15] who identified them by WES in four of 28 patients diagnosed at an age <40 years in a first exploratory series, adding another six cases by Sanger sequencing of POLE codon 286 of 83 microsatellite-stable colorectal carcinomas from patients aged <50 years at the time of diagnosis. Based on these findings, these authors propose that they represent a separate class of early-onset colorectal carcinomas. Histomorphological features, however, were reported as largely uncharacteristic (except for, as a casual observation, frequent debris of necrotic/apoptotic cells mingled with granulocytes in the lumens where tumors were cribriform), and T-cell infiltration was low. Contrary to the Ahn et al. observations, our ID-T2 was from a patient aged 57 at the time of diagnosis which is not really early-onset, anymore; and ID-T6 was observed to be "immunoreactive", at least by our criteria [7].

The POLE variant p.V411L, yet another established POLE driver-mutation, was found in our Sanger sequencing screen in two microsatellite-stable tumors (ID-T1 and ID-T5). As would be expected, TMBs were very high in both cases (199 and 126 variants/Mbp, respectively). Furthermore, quite in line with the idea that hypermutation due to a wealth of neoantigens can elicit an enhanced antitumor immune response, in both tumors T-cell infiltration was very high: intraepithelial T cells were recorded well above the threshold of 123 CD8+ T cells per square millimeter of tumor (our criterion for "immunoreactive" tumors; 7); T cell infiltration per square millimeter total in the tumor centre was around 400, which according to Domingo et al. amounts to a high density [16]. To be pointed out as additional unusual features, ID-T5 was from an unusually young patient (aged 41 years), and ID-T1 was notable for a striking nuclear anaplasia, reminiscent of what has been reported for POLE mutated endometrial carcinomas [17].

ID-T9 of our series puts our current understanding of the role of POLE mutations to a test. The patient was only 21 years old at the time of diagnosis which would be unusual even for a syndromic colorectal cancer. However, genetic work-up did not provide any evidence for an association with a known cancer predisposition syndrome. Furthermore, the tumor was sarcomatoid by histology, which is a rarity among colorectal carcinomas. Secondary involvement of the colon was excluded by the tumor’s macroscopic features (growth from the lumen towards the mesocolic fat), staging examinations (no other tumor found), and immunohistochemistry (CDX2 and CK20 positive, CK7 negative). As regards the DNA polymerase gene mutations, the tumor was unusual for simultaneous POLE and POLD1 mutations (p.V411L and p.R525Q, respectively), both of which, however, were not present throughout the tumor. Heterogeneity of these mutations within the tumor may explain why TMB was low because the POLE variant p.V411L, at least if present throughout, most likely would have resulted in a hypermutator-type cancer. As far as we are aware, heterogeneity of DNA polymerase gene mutations has not been addressed by any published studies. In our study, mutational heterogeneity was a coincidental finding that came to light when trying to reconcile the Sanger screen data with the WES data, but it may be more common than anticipated and add an unexpected layer of complexity. Indeed, the POLD1 variant p.E279K found in ID-T7 was a second example of mutational heterogeneity in our series, although this mutation may well be a passenger-type mutation that, even if uniformly present, may not lead to a hypermutator state.

Assessing the biological significance of DNA polymerase gene mutations is quite difficult in many instances. In vivo studies suggest that mutator effects may depend on the specific mutation incurred by a tumor, which may be weak/moderate to strong in terms of TMB (reviewed in 3). Well-appreciated, recurrent mutations with strong effects on TMB are few, the POLE variants p.P286R and p.V411L found in five of our mutated tumors among them. However, the remaining POLE and POLD1 mutations of our series are not straightforward to place, this being compounded in some cases by the fact that more than one POLE or POLD1 variants could be demonstrated once WES was completed (Table 1). Basically, there are three possible approaches to assessing the functional role of the mutations. First, in silicio algorithms that predict the mutations’ effects on gene function can be employed (as shown in Table 2). By these algorithms, as to be expected, POLE variants p.P286R and p.V411L were clearly flagged as functionally relevant. However, the POLE variant p.D287E was read out as functional by all four algorithms, but the TMB was low in the corresponding tumor and the variant was germline in a 76-year-old woman without polyposis, which makes this in silicio assessment implausible. POLE p.F959L, p.R1556W, and p.P2088S were three more missense variants that were predicted to be functionally compromising by all algorithms. Nevertheless, it remains difficult to draw a final conclusion regarding their functional roles because they all targeted POLE in regions outside the exonuclease or polymerase domains (Fig. 1) and they were found in the two tumors with a p.P286R mutation which by itself would explain the high TMBs. The POLD1 p.L342M of tumor ID-T7, scored as functionally relevant by the MutationTaster algorithm (although not by the other three), was associated with a high, but not ultrahigh TMB. However, this most likely can be ascribed to the mismatch repair deficiency recorded for this tumor. Finally, POLE p.A465T found in tumor ID-T4, the last of the functionally important missense variants by in silicio testing, remains difficult and interesting. For its assessment, as a second approach to the issue of function, we conferred our list with the list of POLE and POLD1 mutations published by Campbell et al. [14] whose “in vivo human mutagenesis screen” of a formidable 81,337 cancers of various types affords an impressive resource. Indeed, the POLE variant p.A465T in tumor ID-T4 is reported by them as associated with an ultrahigh mutational burden in two tumors (their Supplementary Table 2), although it is not included in their list of 29 POLE driver mutations (their Supplementary Table 3). The interpretation of the POLE variant p.A465T’s significance in our case is complicated by the fact that this tumor was Lynch syndrome-associated, which by itself might explain hypermutation. Conceivably, the POLE variant p.A465T might put a mutational load on top of that exacted by the mismatch repair deficiency, conceptualized as an “explosive mutation accumulation” by Hodel et al. [18]; consistent with this idea TMB of this tumor was highest of all. It may be added, parenthetically, that finding a POLE mutation in combination with mismatch repair deficiency is exceptional because usually microsatellite instability in colorectal carcinomas combines with POLD1, but not POLE mutations. The remaining mutations to be clarified are not found in the Campbell et al. files. Finally, a third and, obviously, most valid approach to assessing the significance for POLE or POLD1 mutations on TMB derives from model cell systems. However, the list of POLE and POLD1 mutations studied in this way remains short, so far, and our mutations in question are not among them [3].

A second part of our study addressed the state of the second allele in our POLE or POLD1 mutated colorectal carcinomas. This is a topic of interest because it is well appreciated that POLE and POLD1 are not tumor suppressor genes in a classical sense: on one hand, all POLE and POLD1 mutations associated with hypermutator-type tumors (driver mutations) on record are point mutations whereas truncating mutations or frameshifts are quite rare and, if present, appear to be passenger mutations; on the other hand, the point mutations of POLE and POLD1 in hypermutator-type tumors confer loss-of-function, and therefore, obviously, the genes cannot be classified as oncogenes, either. Conceivably, mutated POLE and POLD1 are “hybrids” and defy to be placed neatly in the oncogene-tumor suppressor gene dichotomy. In the experimental setting, a heterozygous state of POLE or POLD1 mutations is sufficient to induce and/or maintain high mutational burdens in cell lines derived from hypermutator tumors or yeast systems [19,20,21]. However, the situation is less clear in clinical tumor specimens which to-date have only been studied with DNA derived from tumor homogenates. Therefore, we performed laser-capture microdissection to separate tumor cells from the surrounding stroma and used the DNA thus obtained for resequencing and for allelotyping with polymorphic microsatellite markers. In addition, we tested for gene promoter methylations, another mechanism by which the second allele could be compromised in its function. We observed that all POLE and POLD1 mutations were monoallelic, and there was no evidence of a “second hit” on the genes, neither by allelic loss, nor by promoter methylation. These findings suggest that simply the 50% reduction of POLE gene dosage caused by the variants p.P286R or p.V411L, respectively, are sufficient in conferring ultrahigh TMBs in sporadic colorectal carcinomas.

Taken together, in this study we have added evidence to the rising concept that POLE variants p.P286R and p.V411L in colorectal carcinomas define a separate class, although heterogeneity may occasionally prove an issue to be kept in mind. These tumors are not identified in a straightforward fashion by histology, but very likely are “immunoreactive”. Thus, at least these POLE hotspots (and the few others not found here) should be included into sequencing panels, especially if dealing with young patients and/or patients with metastasizing disease in case immune-checkpoint therapy is deliberated. Of the less frequent mutations, POLE p.D287E was not associated with high TMB, but POLE p.A465T may be relevant, although in our series this may have been obscured by the additional mismatch repair defect. As an aspect of tumor biology addressed here, we have shown that a “second hit” on POLE is not required for a significant functional effect.