New insights into the genetic progression of cancer through longitudinal analysis of oral lesions

Abstract

Genetic progression models of cancer continue to determine the roadmap of carcinogenesis, although the sequence of genetic events is inferred rather than empirically determined through longitudinal analyses. Here, we present a unique longitudinal study of oral leukoplakia lesions that transformed into carcinoma. Lesions were followed from initial diagnosis through to malignant transformation. For each lesion, biopsies at baseline, the carcinoma, and all available intermediate biopsies were studied for the presence of dysplasia, genomic copy number aberrations, and mutations in selected head and neck cancer genes. Using this information, the phylogenetic history of all carcinomas was reconstructed within the context of applied interventions. In total, 71 biopsies of 21 lesions were studied. The median time to malignant transformation was 60 months. Oral carcinogenesis emerged as a multi‐(sub)clonal process. Treatment interventions appeared to impact clonal selection, although lesions always remained after excision. Notably, the lesions demonstrated different routes of progression. A canonical pattern of progression was observed, characterized by longitudinal accumulation of abnormal morphology and genetic changes. However, some lesions followed an alternative, stable pattern of oncogenic progression, with no apparent increase in morphological changes or accumulation of genetic aberrations. This latter pattern appeared to be associated with the development of previously identified ‘copy number quiet’ head and neck tumors. This study provides a novel perspective on the temporal evolution of oral leukoplakia and the different routes to cancer progression and highlights the key role of multiclonal field cancerization in lesion recurrence and cancer development. © 2025 The Author(s). The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.

Introduction

Cancer progression models describe the gradual accumulation of genetic and epigenetic changes underlying cancer development. Current oral cancer progression models are limited by the fact that the order of events over time is inferred rather than empirically determined. These models are based on the assumption that mild dysplasia is followed by moderate dysplasia and severe dysplasia in the temporal progression of mucosal precancerous lesions, concurrent with accumulation of specific driver mutations [1, 2]. Validation of these deduced models by analysis of longitudinal biopsies is currently scarce or absent, and carcinogenesis is likely more dynamic than anticipated [3, 4, 5]. Canonical progression models were recently corroborated by reconstructing the phylogenic history of a large cohort of tumors [6]. However, such tumor‐based approaches again only infer the order of progression and are unlikely to capture dynamic changes during the progression of premalignant lesions. Leukoplakia lesions in the oral cavity present a unique opportunity for longitudinal analysis, as patients are typically followed through lifelong active surveillance.

Oral leukoplakia (OL) is defined by the World Health Organization as ‘A white plaque of questionable risk, having excluded (other) known diseases or disorders that carry no increased risk for cancer’ [7, 8]. OL is the most common potentially malignant disorder of the oral cavity, with a worldwide prevalence of 2%–4% and an annual malignant transformation rate of 1%–5% [7, 8, 9, 10, 11]. Current clinical management consists of anamnesis, cessation of possible etiological risk factors, clinical inspection, and a biopsy for histopathological diagnosis, both to exclude carcinoma and to assess the presence of epithelial dysplastic changes. Small lesions are directly excised, while for large and multifocal lesions, only incisional biopsies are performed. Irrespective of treatment, all patients are monitored closely, as there is currently no evidence that excision is effective for the prevention of oral cancer [10, 12]. Consequently, all patients remain in lifelong follow‐up at specialized oral medicine and cancer centers.

The presence of epithelial dysplasia is one of the most important and long‐recognized predictors of the risk of malignant transformation in OL. In general, only the cytological pattern of dysplasia (further referred to as classic epithelial dysplasia) was reported, graded as mild, moderate, or severe [7]. More recently, it was shown that besides classic epithelial dysplasia, the architectural pattern of dysplasia, previously called ‘differentiated dysplasia’, is also of major relevance to assess the risk of malignant transformation in OL [13, 14, 15, 16]. In addition to histomorphological alterations, the presence of genetic changes is a strong predictor of risk for oral squamous cell carcinoma (OSCC) [17, 18]. In the past decade, it has become possible to simultaneously identify genome‐wide copy number alterations (CNAs) and mutations using next‐generation sequencing (NGS) techniques. This resulted in the elucidation of the genomic landscape of head and neck squamous cell carcinoma (HNSCC) by The Cancer Genome Atlas (TCGA) consortium [19]. Genetic aberrations, such as CNAs in specific regions of interest, particularly loss of 3p14 and gain of 20p11, have been shown to be important for risk stratification in OL. Mutations in genes associated with OSCC, such as TP53 and CASP8, have been shown to be predictive markers for malignant transformation of OL [1, 17, 18, 20, 21]. Recently, we have demonstrated the combined power of histomorphological and genetic markers for the prediction of malignant transformation in OL [13, 16, 18]. Three risk groups were identified, capable of predicting malignant transformation up to 15 years later [18]. Notably, these predictive histomorphological and genetic alterations were typically identified in the OL biopsy at baseline, despite the fact that it may take up to 20 years before the lesions transform into OSCC [18]. This suggests that key genetic changes associated with cancer risk emerge early, while malignant transformation may occur one to two decades later.

Recently, computational methods using TCGA data have been employed to infer the order of genetic aberrations that precede malignant transformation, and these methods appear to support the canonical genetic progression model [6]. However, the genetic progression model has never been confirmed in longitudinal series and may be much more dynamic and heterogeneous than can be deduced from tumor data alone. For de novo OSCC, longitudinal analysis is not possible, but OL lesions are generally followed over time until malignant transformation occurs.

We monitored 247 OL patients in a longitudinal cohort study, with a biopsy at baseline and multiple intermediary biopsies on clinical indications during a follow‐up period of up to 25 years. The aim of the present study was to determine how lesions develop over time and ultimately undergo malignant transformation, with a focus on histopathology and genetic changes. Using genetic data acquired from longitudinal biopsies, we reconstruct the phylogenetic history of each lesion and, together with the histomorphological changes, demonstrate the existence of alternative routes to cancer development beyond the canonical genetic progression model.

Methods

Ethics approval and patient consent statement

This study followed the principles of the Helsinki Declaration and the national guidelines for the secondary use of human tissue of the Dutch Federation of Biomedical Scientific Societies and the General Data Protection Regulation of the European Commission [22, 23, 24]. This study was approved by the Institutional Research Board of Amsterdam UMC (FWA00017598) under protocol no. 2021.0036. This committee confirmed that the Dutch Medical Research Involving Human Subjects Act (WMO) did not apply to the presented retrospective study. Written informed consent was obtained from all participants.

Patients

Progressing lesions were selected from a consecutive cohort of 247 patients with OL who were referred to the Department of Oral and Maxillofacial Surgery/Oral Pathology at Amsterdam UMC, location VUmc, between January 1997 and July 2022 (supplementary material, Figure S1). Patients were included as previously described [13, 16, 18]. Records were updated until 31 December 2023. Patients were defined as progressors when an OSCC developed at the same site as the OL during follow‐up, as confirmed by histopathology. Follow‐up was calculated from the date of first biopsy until malignant transformation, rounded down to whole months. Patients were excluded when either the first OL biopsy or the tumor biopsy, or both, were not available in the archive for histopathological or DNA sequencing analysis. Smoking and alcohol consumption were recorded as ‘use’, ‘no use’, or ‘unknown’ at the time of inclusion.

Tissue collection and processing

For each patient, all available formalin‐fixed, paraffin‐embedded (FFPE) biopsy specimens were collected from the pathology archive at Amsterdam UMC, location VUmc, comprising all initial OL biopsies, available intermediate OL biopsies, and the final tumor biopsies. Tissue processing, DNA isolation and histopathological analyses were performed as previously described [13, 16, 18]. Slides were graded for classic epithelial dysplasia, architectural dysplasia, and expression of cytokeratin 13 (CK13) and cytokeratin 17 (CK17).

Calling of somatic copy number changes and mutations

NGS library preparation from DNA and subsequent calling of somatic copy number changes and mutations were performed as previously described [3, 4, 18, 25]. Sequencing pools were used for paired‐end 150‐bp sequencing on the Novaseq 6000 system (Illumina, San Diego, CA, USA) to obtain low‐coverage whole‐genome profiles for CNA analysis. Subsequently, these same libraries were used for the capture of 12 target genes that are often mutated in OSCC: TP53, NOTCH1, FAT1, CASP8, KMT2D, CDKN2A, HRAS, PIK3CA, FBXW7, PTEN, AJUBA and NSD1 [2, 8, 18].

All breakpoints from paired samples of each patient were combined and mapped onto each original sample to allow clonal analysis of somatic copy number changes. Subsequently, the average of all segments was rounded to the nearest integer and classified as aberrant when the rounded mean differed by at least one copy.

For somatic mutations, the minimum variant‐supporting read depth required to call nucleotide mutations or deletions with statistical certainty was determined for each base of each sample (see supplementary codes) [26]. p values were calculated using the binomial distribution as the probability of sequencing as many, or more, variant‐supporting reads given the expected error rate and the total depth at the specific base location. Variants were reported with p values <10⁻⁶. Insertions of nucleotides were called when variant allele frequencies (VAFs) exceeded 10%, with minimally 10 variant‐supporting reads.

The presence of a called variant in one of the paired biopsies was tested in all other samples of the same patient. As it is often not straightforward to distinguish between noise and somatic variants in FFPE tissue samples, binomial testing was applied. p values for mutations and deletions were calculated using the binomial distribution based on the depth of coverage of the specific bases in the paired biopsies. For insertions, the variant‐supporting reads and total depth were manually obtained through the Integrative Genomic Viewer (IGV) version 2.16.1 [27]. Mutations, deletions, and insertions in paired biopsies were referred to as true mutations at a VAF of >0.01; all other genes were defined as true WT. Program versions and settings are provided in Supplementary materials and methods.

TCGA dataset

To compare the frequencies of CNAs and mutations, the OSCC cohort from the TCGA PanCancer Atlas dataset [19] was included as an independent reference set. Selection and filtering of this dataset to obtain an OSCC cohort for CNA analysis (TCGA‐OSCC‐CNA, 312 samples) and for mutation analysis (TCGA‐OSCC‐MUT, 299 samples) has been previously described [18].

Reconstruction of phylogeny

Phylogenetic trees were inferred using methods adapted from the Clonal Architecture Learning by Directed Evolutionary Reconstruction (CALDER) algorithm [28]. First, CNAs were assigned an alternative and reference read depth that corresponded to the signal deflection whose sum was equal to the mean coverage, as observed in the targeted sequencing (see supplementary codes). It was confirmed that CNAs in paired samples contained the same breakpoints to ensure they were derived from the same event (supplementary material, Figure S2). Subsequently, CNAs and mutations were clustered using Pyclone‐VI, a Bayesian tool used for inferring the clonal structure of lesions from high‐throughput sequencing data, with a maximum of 20 clusters, binomial density, and 10,000 restarts [29, 30]. The per‐biopsy input for Pyclone‐VI is provided in supplementary material, Table S1. Based on the clustered data, phylogenetic trees were constructed using the CALDER algorithm with the glpk solver [28]. A distinction was made between subclones, those cells that all derived from a single clone sharing the same founder mutation (represented by the A in the phylogenetic tree figures), and separate clones that could not be fitted to the most optimal tree. These additional clones were manually added to the figures. Using the phylogenetic trees, the relative timing of each genetic alteration was calculated based on the order of occurrence, and a genetic progression model was created based on each change that was present in at least three patient lesions.

Statistical analyses

All statistical analyses were performed using R version 4.0.3 (R Core Team, Vienna, Austria, RRID:SCR_000432). The Wilcoxon signed‐rank test was used to compare overall differences in numbers of CNAs and mutations between the OL and OSCC cohorts. McNemar's test was used to compare differences in the presence of specific CNAs and mutations between the OL and OSCC cohorts. Fisher's exact test was used to compare differences in the presence of CNAs and mutations between the OSCC samples in this study and the TCGA cohort. All figures were created using R and Adobe Illustrator (Adobe Illustrator 2022, Adobe, San Jose, CA, USA, RRID:SCR_010279). In the individual patient figures, biopsies were sometimes shifted on the time axis for visual reasons, but the correct month was always depicted on the x‐axis.

Results

Clinical characteristics of OL cohort

At the time of the last update, a total of 49 out of 247 OL lesions had transformed into OSCC, of which 21 satisfied the inclusion criteria for this study. The other 28 patients were either diagnosed or treated at other hospitals, and samples were unavailable. The selection of patients from the total cohort is presented in supplementary material, Figure S1. The clinical characteristics for each patient are presented in Figure 1A and Table 1.

Figure 1.

Overview of all clinical variables and morphological and genetic changes for all biopsies of all patients. (A) Patient characteristics, including gender, smoking/alcohol use, and location of oral leukoplakia (OL). The bar at the bottom indicates the study ID and biopsy number. (B) Biopsy type, presence of dysplasia, and expression of cytokeratins 13 and 17. (C) Presence of copy number gains and losses for the entire genome. The total number of affected chromosome arms is provided. Asterisks indicate specific biopsy samples that were not sequenced for CNAs. (D) Mutations in 12 genes that are often associated with head and neck squamous cell carcinoma for all sequenced biopsies. The number in the green squares (multiple hits) indicates the total number of mutations in the gene in the biopsy. The total number of mutations per biopsy is provided above. Asterisks indicate specific samples that were not sequenced for mutations. Right: percentage of biopsies containing a mutation in specific gene. Bottom bar: study ID and biopsy number. AD, architectural dysplasia; C, carcinoma; CED, classic epithelial dysplasia; CK13, cytokeratin 13; CK17, cytokeratin 17; CNA, copy number alteration.

Table 1.

Patient characteristics.

Patient ID	Age at first biopsy (years)	Gender	Location of OL	Number of OL biopsies	Time to malignant transformation (months)	Smoking	Alcohol	Progression category
8	45	Female	Tongue	6	183	No use	Unknown	Undefined
11	31	Female	Tongue	4	182	Use	Use	Undefined
55	36	Female	Tongue	4	172	Use	Use	Canonical
66	34	Female	Tongue	4	58	No use	Unknown	Stable
77	31	Female	Tongue	1	25	Use	Unknown	Stable
78	67	Female	Tongue	2	126	No use	Use	Undefined
81	58	Female	Buccal mucosa	1	84	No use	No use	Canonical
98	62	Female	Tongue	2	20	Unknown	Unknown	Canonical
103	54	Male	Tongue	1	60	No use	Use	Canonical
105	67	Male	Floor of mouth	3	67	Use	Use	Undefined
111	62	Female	Tongue	1	24	Use	Unknown	Canonical
113	60	Male	Tongue	2	36	Use	Use	Undefined
114	60	Male	Floor of mouth	1	155	Use	Use	Canonical
118	64	Female	Tongue	1	17	Use	Use	Stable
138	51	Female	Tongue	1	28	Use	Use	Canonical
141	66	Female	Tongue	1	37	Unknown	Unknown	Canonical
144	63	Female	Hard palate	2	59	Use	Unknown	Stable
146	63	Female	Tongue	3	63	No use	Use	Undefined
169	64	Male	Buccal mucosa	2	126	Use	Use	Canonical
170	74	Female	Tongue	7	65	No use	Unknown	Stable
183	85	Female	Upper alveolus and gingiva	1	12	No use	No use	Stable

Most morphologic and genetic aberrations are present at baseline

All biopsies were reviewed for histomorphological changes and analyzed for genetic alterations. In addition, we applied immunostaining for differentiation markers CK13 and CK17 that facilitate identification of histomorphological changes. An overview of all changes in all biopsies is presented in Figure 1. While normal mucosal epithelium is characterized by a uniform, strong expression of CK13 and the absence of CK17, this pattern is generally reversed in OSCC [31, 32, 33]. To investigate whether this switch occurs in premalignancy, the expression of cytokeratins CK13 and CK17 was assessed for 19 initial OL biopsies, 15 tumor biopsies, and 20 intermediary OL biopsies (Figure 1B). CK13 expression was lost in 17 out of 19 (89%) initial OL biopsies, and CK17 was gained in 17 out of 19 (89%) of the initial OL biopsies. Only one out of 19 (5%) initial OL biopsies had normal CK13 and CK17 expression (although classic epithelial dysplasia was present). None of the tumor biopsies retained normal CK expression. Thirteen out of 15 (87%) tumor biopsies lost CK13 expression, while 14 out of 15 (93%) had gained CK17 expression. All 20 (100%) intermediary OL biopsies lost CK13 expression and gained CK17 expression. There was no effect of treatment on expression of CK13 or CK17, indicating that the mucosa never returned to normal, in line with the ineffectiveness of excision.

We analyzed the biopsies at different timepoints for the presence of histomorphological changes; both classic epithelial dysplasia and the recently recognized architectural form of dysplasia were determined. Classic epithelial dysplasia was present in seven out of 21 initial biopsies (33%), whereas architectural dysplasia was present in 13 out of 21 initial biopsies (62%). Combined, 20 out of 21 initial (95%) biopsies displayed dysplasia of any type in the baseline biopsy, in line with the fact that all these OL lesions transformed. Changes in dysplasia over time are shown in Figure 1B. For 11 patients, it was possible to perform histopathological analysis of subsequent OL biopsies. In five of these cases (case IDs 066, 144, 146, 169, and 170; supplementary material, Figure S3), only architectural dysplasia was present in biopsies until malignant transformation, whereas only one case (case ID 098; supplementary material, Figure S3) had classic epithelial dysplasia in all biopsies but no architectural dysplasia. Surprisingly, we observed three cases (case IDs 008, 011, and 113, supplementary material, Figure S3) in which the grade of classic epithelial dysplasia decreased. In all these cases, the biopsy with the lower degree of classic epithelial dysplasia coincided with the occurrence of architectural dysplasia. Finally, two cases (case IDs 055 and 105; supplementary material, Figure S3) displayed architectural dysplasia in the initial biopsies, which progressed to severe classic epithelial dysplasia 65 and 86 months before malignant transformation occurred.

Additionally, we determined the cancer driver genes involved, the CNAs and the clonal development in time. As we only had access to archival FFPE specimens, we analyzed a focused gene panel to obtain sufficient coverage for reliable mutation calling. Genetic aberrations for each biopsy are presented in Figure 1C,D. For each sample the whole‐genome sequencing coverage depth and targeted mutation information is provided in supplementary material, Tables S2 and S3. To investigate the pattern of genetic changes of these OSCCs that specifically emerge from OL lesions, we compared the frequency of genetic aberrations with the OSCC tumors included in the TCGA database. In general, the data were very comparable. However, two CNAs (2q loss; 14% versus 0% and 16p gain; 16% versus 4%) and mutations in three genes (FAT1; 48% versus 25%, KMT2D; 29% versus 10% and NOTCH1; 43% versus 16%) were more often found in the present cohort compared to the TCGA cohort (p < 0.05, Fisher's exact test, supplementary material, Figure S4). In contrast, three CNAs (4p loss; 0% versus 11%, 9p loss; 5% versus 15% and 13p loss; 0% versus 18%) were more often affected in the TCGA cohort compared to the cohort in the present study (p < 0.05, Fisher's exact test, supplementary material, Figure S4). These data suggest that genetic alterations in tumors arising in OL differ slightly compared to those in OSCC that present de novo.

Next, we analyzed the genetic changes between initial OLs and the OSCCs that originated from them. On average, first biopsies of OL harbored 6.4 CNAs (range 0–23) compared to 8.3 CNAs in carcinoma (range 0–21, p = 0.12, Wilcoxon signed‐rank test), 2.5 mutations (range 0–5) compared to 3.3 mutations in carcinoma (range 1–7, p = 0.01, Wilcoxon signed‐rank test) and a total of 8.9 genetic aberrations (range 0–24) compared to a total of 11.6 genetic aberrations in carcinoma (range 1–24, p = 0.06, Wilcoxon signed‐rank test). We failed to identify statistically significant increases in mutations in specific genes during progression (p > 0.25, McNemar's exact test, supplementary material, Figure S5). TP53 mutations showed the expected trend of being more commonly accumulated in carcinoma, although in one case a TP53 wild type (WT) tumor emerged, even though the biopsy of the OL at baseline harbored a mutation (Patient ID 103; supplementary material, Figure S3). Sequencing quality and coverage would certainly have allowed detection of the mutation in the tumor biopsy.

With respect to individual lesions, there was an accumulation of CNAs in eight cases, no change in CNAs in seven cases (including the cases for which no CNAs at all were reported throughout follow‐up), and a decrease in the number of CNAs in five cases. In one specific case, one gain that was present in the first OL biopsy was lost in the carcinoma while another gain emerged (case 146, Figure 1C, supplementary material, Figure S3). Detection of mutations followed a comparable trend: Mutations accumulated in nine cases, remained consistent in eight cases, decreased in two cases, and switched in two cases, the latter suggesting progression in genetically independent clones (Figure 1D).

Identification of multiple oral leukoplakia progression patterns

When analyzing the various biopsy series, we identified distinct patterns of malignant progression. The supplementary material, Table S4, contains the counts used for the classification of the lesions into the different progression patterns (as described in the Supplementary materials and methods). The canonical pattern of progression with increasing morphological and genetic changes in time was observed in nine of 21 cases. However, a second and stable pattern of progression with minimal increase in changes over time was noted in six of 21 cases. There was no significant difference in the time between initial diagnosis and malignant transformation for the canonical lesions (range = 20–172) and the stable lesions (range = 12–65, p = 0.181, Wilcoxon rank‐sum test). In some cases (six of 21), progression followed a relatively unpredictable pattern (Table 1). The morphologic and genetic progression for three example lesions is shown in Figures 2, 3, 4. The progression of case 055 followed a more or less canonical pattern of carcinogenesis, which was characterized by longitudinal progression of morphological and genetic changes, irrespective of interventions (Figure 2). At baseline, two CNAs were observed and architectural dysplasia was present. The lesion was excised and recurred with additional CNAs and architectural dysplasia after 43 months. The lesion was re‐excised at two more timepoints, and in both instances the lesion recurred with accumulated CNAs, first in the presence of moderate dysplasia and second in the presence of severe dysplasia, at 80 and 86 months from baseline, respectively. After in total 172 months and four surgical interventions, the lesion developed into cancer. Remarkably, the TP53 driver mutation was already present in the first OL biopsy. This TP53 mutation was observed throughout follow‐up, while no other mutations were detected, even in the carcinoma, at least in the 12 tested cancer genes. The carcinoma developing from this lesion exhibited many more CNAs, which may underpin its malignant state. Despite an imperfect concordance between the timing of genetic events and the observed histological changes, which might have been the result of the applied interventions, we concluded that this lesion followed the canonical progression route.

Figure 2.

Overview of clinical pictures, treatment, dysplasia, CNAs, cellularity, and mutations for all biopsies of patient 55. Top panel: Clinical pictures at several timepoints during progression. The first horizontal bar indicates whether an excisional biopsy, incisional biopsy, or CO₂‐laser evaporation was performed. The second horizontal bar shows the presence and grade of dysplasia. The columns show the presence of gains and losses for each of the 22 chromosomes and below cellularity estimates are shown, as determined by Allele‐specific Copy number Estimation (ACE). The mutations present in each biopsy are shown at the bottom of the top panel. The timepoints on the x‐axis correspond to biopsy, clinical evaluation, or both. In the bottom panel, the central phylogenetic tree, as inferred using the Clonal Architecture Learning by Directed Evolutionary Reconstruction (CALDER) algorithm, shows the evolutionary relationship between clones. Each letter indicates a specific clone. Numbers on branches indicate specific genetic aberrations that are acquired between clones. The corresponding mutations for each number on a branch are listed on the left, together with genetic aberrations for that specific number. A clonal prevalence plot is shown on the right, where the thickness of a wave indicates the relative prevalence of a given clone at each timepoint. Color coding in the clonal prevalence plot corresponds to the colors of the clones in the phylogenetic tree. At each timepoint, a vertical line indicates that the lesion was excised, while a dotted line indicates that an incisional biopsy was performed. L1, lesion 1; L2, lesion 2; L3, lesion 3; L4, lesion 4; C, carcinoma. AD, architectural dysplasia; CED, classic epithelial dysplasia; CNA, copy number alteration.

Figure 3.

Overview of clinical pictures, treatment, dysplasia, CNAs, cellularity, mutations, and reconstructed phylogenetic tree for all biopsies of patient 170. Diagram layout is as described in Figure 2 legend. L1, lesion 1; L2, lesion 2; L3, lesion 3; L4, lesion 4; L5, lesion 5; L6, lesion 6; L7, lesion 7; C, carcinoma. AD, architectural dysplasia; CED, classic epithelial dysplasia; CNA, copy number alteration.

Figure 4.

Overview of clinical pictures, treatment, dysplasia, CNAs, cellularity, mutations, and reconstructed phylogenetic tree for all biopsies of patient 011. Diagram layout is as described in Figure 2 legend. L1, lesion 1; L2, lesion 2; L3, lesion 3; L4, lesion 4; C, carcinoma. AD, architectural dysplasia; CED, classic epithelial dysplasia; CNA, copy number alteration.

In addition, stable routes of carcinogenesis were observed. The premalignant lesion of patient 170 entailed a highly recurrent OL that followed a stable pattern of carcinogenesis, which was characterized by no apparent increase of morphological or genetic aberrations, irrespective of applied interventions (Figure 3). All biopsies and the tumor sample lack any CNAs but harbor clonal FAT1 and HRAS mutations at one allele, indicating that these were present in the most recent common ancestor of all lesions and the tumor. In addition, in all OL biopsies, architectural dysplasia was observed. Interestingly, this lesion acquired five different CASP8 mutations during progression, and one OL and the carcinoma contained two CASP8 mutations on two alleles, albeit at VAFs corresponding to subclonal presence. Multiple OL lesions followed a comparable stable progression route, such as lesions of cases 144 and 146 (Table 1 and supplementary material, Figure S3). Most lesions of this group likely reflect the carcinogenesis pattern of so‐called ‘copy number quiet’ tumors: HPV‐negative, lack of CNAs, frequent CASP8 and HRAS mutations, and presence of architectural dysplasia [19, 34, 35].

However, different routes of carcinogenesis were observed as well. The progression of lesion 011 did not adhere to the canonical pattern of progression, but it did not follow the stable pattern either (Figure 4). At presentation the lesion was classified as mild dysplasia and contained a variety of CNAs and mutations in NOTCH1 and TP53. The lesion was excised but recurred after 99 months, but now with architectural dysplasia, only a few CNAs, and several additional mutations. The lesion persisted, and two additional incisional biopsies were obtained with the presence of moderate and severe dysplasia, at 169 and 181 months after diagnosis, respectively, followed by carcinoma at 182 months. These three latter biopsies contained most of the same CNAs and mutations that were present at 99 months, with some additional changes. Several lesions followed this mixed pattern of progression, which seemed to stem from multiclonality in the earliest premalignant changes. Progression becomes apparent after a strong shift in clonal composition, which seems to coincide with an intervention such as surgical excision.

Reconstruction of phylogenetic trees provides insight in OL heterogeneity and provides further evidence for oral field cancerization

We noted a variety of subclonal genetic changes in OL lesions, suggesting that carcinogenesis was multiclonal and heterogeneous. Using the genome‐wide CNAs and analyzed mutations we inferred the phylogenetic history of each OL. In Figures 2, 3, 4 the three phylogenetic trees are depicted corresponding to the cases discussed earlier. All other phylogenetic trees are presented in supplementary material, Figure S3. Although we obtained genome‐wide CNAs, the amount of mutational data was limited by the archival nature of the source material, which obviously may have impacted the deduced phylogenetic history. Nonetheless, the various constructed phylogenetic trees do represent the clonal heterogeneity and dynamics across OL progression. First, Pyclone‐VI was employed to cluster all genetic aberrations, resulting in a median of four subclones (range 1–8) per OL. Subsequently, CALDER was used to infer phylogenetic trees based on these subclones. The generated trees contained a median of four (range 1–8) subclones per biopsy. In seven cases one or two additional independent clones were present next to the clones that belonged to the main tree. Eleven subclones disappeared after incisional biopsy, suggesting that these were outcompeted by more actively proliferating subclones, and 23 subclones were lost when the lesion was excised. These data strongly suggest that interventions affect the lesion, but are not curative, as is known from clinical experience. In contrast, a median of two subclones (range 1–5) of a total of 44 observed subclones were present in the final OSCC biopsies (supplementary material, Figure S3).

CALDER infers the order of genetic aberrations, even from before the first biopsy. A total of 33 (sub)clones were identified before the first biopsy: in 11 cases a single subclone was inferred before the first biopsy, in four cases two subclones were inferred before the first biopsy, and in five cases three subclones were inferred before the first biopsy. Based on the information provided by CALDER, it was possible to infer the relative timing of each genetic change (Figure 5, supplementary material, Table S3). This represents the molecular time spanning from the occurrence of the first (i.e. oldest) mutation in the founder clone until the moment of tumor biopsy [36]. All alterations in genes or regions that were affected in at least three samples were included for this analysis. The most common early events were mutations in CDKN2A, TP53, and loss of chromosome arm 11p. Common middle events were 20p gain and mutations in PIK3CA. Common late events were losses in 8p, 9q, and 9p. Interestingly, several common aberrations, such as 3q gain, 8q gain, and 13q loss, occurred anywhere throughout the progression of OLs. Mutations in FAT1 and KMT2D developed either at the start of oncogenesis or near malignant transformation.

Figure 5.

Relative time of occurrence for genetic aberrations in OL. This plot shows the relative timing of occurrence for each genetic aberration that is present in at least three samples. On the y‐axis the genetic aberrations are shown. The x‐axis indicates the relative timing of occurrence, where closer to 0 means its closer to the initial mutations present in clone A as shown in Figures 2, 3, 4 and supplementary material, Figure S3. Closer to 1 means the aberration is more often acquired later in clonal evolution. Genetic aberrations are sorted based on mean relative timing. The bar graph on the right shows the number of lesions in which the aberration was present. n, number of genetic aberrations.

Discussion

In this study, we show for the first time the longitudinal morphological and genetic alterations in OL biopsies and the corresponding OSCC and how these change over time. Our analyses were based on FFPE tissue, which enables reliable histology but hampers large‐scale sequencing. We clearly recognize at least two different carcinogenesis routes: one more or less adhering to the canonical morphological and genetic progression model, and one that is stable without apparent progression until malignant transformation occurs.

Striking was the heterogeneous nature of many lesions. We cannot exclude the possibility that the interventions in some patients also affected the natural development of the lesions. To our knowledge, two other studies followed patients with oral lesions over time, of whom multiple biopsies were obtained [5, 21]. For the first study five patients were included with de novo OSCC and concurrent dysplasia, and for one of these patients additional longitudinal biopsies were studied retrospectively. This study showed that the relationship between dysplastic and malignant tissue was not always linear and that subclones might be lost due to surgery, but this was only based on one longitudinally studied case. For the second study two patients were included from whom multiple biopsies had been obtained that eventually developed into OSCC, and for both of these patients the dysplasia grade decreased during follow‐up (from low‐risk dysplasia to no dysplasia and from high‐risk dysplasia to low‐risk dysplasia, respectively). These results also suggest that progression is dynamic and that OL lesions do not always progress step by step from hyperplastic to dysplastic lesions with increased severity to OSCC. In fact, we noted lesions with architectural dysplasia that persisted over almost two decades until malignant transformation was triggered.

Some of these lesions with architectural dysplasia were devoid of any CNAs, something we also observed in a previous study on the genetic landscape of OL [18]. In the majority of HPV‐negative HNSCCs, CNAs are frequent and TP53 is often mutated, but more recently a novel class of HNSCC was identified, specifically in the oral cavity, which harbors few to no CNAs and is associated with WT TP53. This subclass of tumors was referred to by the TCGA as ‘copy number quiet’ HNSCC [37, 38]. These tumors are associated with mutations in HRAS and CASP8, present frequently in the oral cavity, and are associated with a better prognosis [2]. OL lesions that present with architectural dysplasia may be the precursor for this HNSCC ‘copy number quiet’ subclass. Even though ‘copy number quiet’ tumors only constitute roughly 10% of all OSCCs, in the cohort presented in this study, five out of 21 lesions contained only architectural dysplasia, presented without any CNAs, retained TP53 WT and often contained mutations in HRAS and CASP8 [35]. The precise mechanism behind the specific progression of OL lesions with architectural dysplasia, including possible etiological factors, remains to be further investigated.

Using the genetic information obtained through the employed NGS approaches, we inferred the phylogenetic history of each OL and corresponding OSCC and showed the order of genetic aberrations that occurred during OL progression. OLs show a remarkable heterogeneity, as evidenced by the inference of parallel clones in the evolutionary histories of seven OLs. We observed in five of 21 cases studied that multiple independent clones were present in the invasive carcinoma, suggesting a multiclonal origin of cancer in these cases. We must consider that some OSCCs are not of monoclonal origin but of multiclonal origin, as Slaughter previously suggested in 1953 [39]. This challenging hypothesis requires further attention. Analysis of recurrent or second primary tumors in these patients will further signify the relevance of this phenomenon.

Our study has several limitations to be considered. We had to rely on archival FFPE biopsies, which negatively impacts the quality of the sequencing data when compared to fresh‐frozen material. To account for this, we applied rigorous controls and focused on a targeted gene panel to enable deeper sequencing. While these genes are the most important driver genes for OSCC, it is possible that we have missed mutations in genes that contributed to carcinogenesis and that could have marked specific clones. Moreover, the variant allele frequency of all other mutated genes would have aided in establishing the phylogenetic trajectory from OL to OSCC. Furthermore, the sequenced biopsies are only part of a larger OL lesion, and for clinicians it is not always clear what part of the lesion should be biopsied, especially in more widespread lesions. Mutations that eventually emerge in the tumor may already have been present in a part of the lesion that was not biopsied.

Importantly, it remains a question as to how representative the progression of OL is for the progression of OSCC in general. As OL presents as a white plaque and the majority of OSCCs (~75%) arise de novo without a clinically visible lesion, it stands to reason that there might be a difference in genetic makeup between these lesions. Based on the comparison of genetic changes between the carcinomas investigated in this study and those in the TCGA database, the cohorts were very comparable, although some CNAs were more common in OSCCs from the TCGA cohort, whereas some driver genes were more often present in the OSCCs that originated from OL. Furthermore, the female:male ratio in the patient cohort is very different from typical OSCC cohorts such as the TCGA cohort, which almost exclusively comprises more male than female patients. The different female:male ratio may bias findings, although we are unaware of any such bias in carcinogenesis.

Finally, we cannot rule out the possibility that the interventions (incisional biopsies or attempts for excision) affected our findings. This is unavoidable, as OL lesions are routinely biopsied and excised when possible as part of standard clinical management. While these interventions clearly did not prevent recurrence of the lesions and subsequent malignant transformation in this cohort, they may have affected clonal dynamics in some patients, particularly after attempts to excise the lesions. We assume that lesions are part of a larger field of cancerization and may recur due to the residual aberrant cells that persist after treatment. Notably, all recurrent lesions appeared at the same anatomical location. Various techniques, such as autofluorescence imaging, are currently being explored to detect aberrant tissue beyond what is macroscopically visible. However, these approaches have yet to be integrated into the standard of OL management. Our assumption is supported by the observation that most clones are interconnected through shared mutations and CNAs. However, we also identified some clones that were not of the same lineage, suggesting that oral carcinogenesis is a far more dynamic process than previously thought.

In summary, we morphologically and genetically characterized a longitudinal series of biopsies of OL lesions that developed into cancer. One of the remarkable conclusions of the presented study is that many lesions are very heterogeneous, while the presence of the important morphological and genetic aberrations, features associated with malignant transformation, remain consistent, sometimes over two decades. We distinguished at least two separate carcinogenesis routes, one characterized by longitudinal progression leading to the typical high‐CNA OSCCs and one that is more stable and that may lead to the typical ‘copy number quiet’ OSCCs. It is these lesions in particular that may remain stable for decades without any apparent morphological and genetic progression, until a certain transition triggers malignant transformation. Whether this transition relates to further genetic changes in unknown genes or is induced by stochastic changes in gene expression or other factors in the underlying tissue such as immune infiltration remains a crucial question in the study of oral carcinogenesis.

Author contributions statement

EB and RHB conceptualized the project. LJW, JBP and EB performed the investigations. LJW and JBP performed formal analysis and visualization of the data. LJW, JBP and AB worked on methodology, software and data curation. JBP validated the data analysis. LP, IE, ERB, JGAMV, EHM and EB provided resources. JGAMV, EB and RHB were involved in supervision, project administration and funding acquisition. LJW, JBP and RHB wrote the original draft of the manuscript. All authors were involved in reviewing and editing of the final draft and had final approval of the submitted and published versions.

Supporting information

Supplementary materials and methods

Figure S1. Overview of inclusion of patient samples from consecutive OL cohort

Figure S2. Overview of CNA breakpoints for each biopsy of each sample

Figure S3. Overview of clinical pictures, treatment, dysplasia, CNAs, cellularity, mutations, and reconstructed phylogenetic tree for all biopsies of all included patients

Figure S4. Differences in frequencies of genetic aberrations between oral squamous cell carcinoma TCGA set and carcinomas included for this study

Figure S5. Differences in frequencies of genetic aberrations between baseline OL biopsies and carcinomas included for this study

Table S1. Per‐sample input for Pyclone

Table S2. Detailed overview of included mutations for all biopsies including relative timing based on phylogenetic reconstruction.

Table S3. Per‐sample sequencing characteristics

Table S4. Counts used for classification of lesions into different progression patterns

Data availability statement

Due to the EU general data protection regulation (GDPR), DNA sequencing data of patient samples cannot be made publicly available. On reasonable request to the corresponding author, the primary sequencing data will be made available under a material transfer agreement. The R code for data analysis of this study is available at https://codeocean.com/capsule/6949968/tree