Molecular evolution in different subtypes of multifocal hepatocellular carcinoma

Multifocal hepatocellular carcinoma (MF-HCC) accounts for > 40% of HCCs, exhibiting a poor prognosis than single primary HCCs. Characterizing molecular features including dynamic changes of mutational signature along with clonal evolution, intrahepatic metastatic timing, and genetic footprint in the preneoplastic stage underlying different subtypes of MF-HCC are important for understanding their molecular evolution and developing a precision management strategy. We conducted whole-exome sequencing in 74 tumor samples from spatially distinct regions in 35 resected lesions and adjacent noncancerous tissues from 11 patients, 15 histologically confirmed preneoplastic lesions, and six samples from peripheral blood mononuclear cells. A previously published MF-HCC cohort (n = 9) was included as an independent validation dataset. We combined well-established approaches to investigate tumor heterogeneity, intrahepatic metastatic timing, and molecular footprints in different subtypes of MF-HCCs. We classified MF-HCCs patients into three subtypes, including intrahepatic metastasis, multicentric occurrence, and mixed intrahepatic metastasis and multicentric occurrence. The dynamic changes in mutational signatures between tumor subclonal expansions demonstrated varied etiologies (e.g., aristolochic acid exposure) underlying the clonal progression in different MF-HCC subtypes. Furthermore, the clonal evolution in intrahepatic metastasis exhibited an early metastatic seeding at 10–4–0.01 cm3 in primary tumor volume (below the limits of clinical detection), further validated in an independent cohort. In addition, mutational footprints in the preneoplastic lesions for multicentric occurrence patients revealed common preneoplastic arising clones, evidently being ancestors of different tumor lesions. Our study comprehensively characterized the varied tumor clonal evolutionary history underlying different subtypes of MF-HCC and provided important implications for optimizing personalized clinical management for MF-HCC.


Introduction
Approximately 41-75% of hepatocellular carcinoma (HCC) patients presented multifocal lesions (i.e., MF-HCC) at diagnosis with significant anatomical and biochemical heterogeneity [1], exhibiting an increased tumor burden with poor prognosis than single primary HCCs [2].MF-HCCs can occur synchronously or metachronously, either from intrahepatic metastasis or multicentric occurrence [3,4], exhibiting varied biological behaviors and different aggressiveness.A comprehensive understanding of the molecular features of MF-HCCs could bring new insights into its carcinogenesis and clinical management, including surgical prioritization and allocation, adjuvant therapy, and follow-up after resection [3].
Sequencing-based investigation has benefits for increasing pathological discrimination of MF-HCCs and revealed pervasive genomic and transcriptional heterogeneity across (interfocal) and within (intratumoral) lesions [3,5], which may result in different responsiveness to targeted drugs [6].Furthermore, it has increased our understanding of MF-HCCs that exhibit distinct genetic structures and evolutionary history [7][8][9], as well as heterogeneous immunogenomic footprints and escape mechanisms [10].However, molecular features during clonal evolution, including dynamic changes of mutational signature, quantitatively intrahepatic metastatic timing, and precancerous molecular changes underlying multicentric occurrence, have yet to be vigorously investigated.
Molecular features of MF-HCCs (e.g., mutational signatures) changed significantly during tumor evolution.The mutational process is associated with tumor subclone emergence [11], and changes in mutational signatures coincided with the etiology change that may shape tumor clonal evolution.Previous studies have shown that mutational processes in melanomas related to ultraviolet B exposure were remarkably stable between tumor subclonal expansions.In contrast, the activity of mutational signatures changed ubiquitously in pancreatic tumor [12].Tracking the changes in mutational signature informs timely intervention of extrinsic etiology.
Synchronously MF-HCCs with intrahepatic metastasis showed an aggressiveness of metastatic capability.The traditional forward-time model posited that metastatic seeding occurred late in a linear progression, while the recently proposed 'looking backward' model suggested that metastasis might have happened before the primary site could be clinically detected [13,14].Clocking the metastatic timing with quantitative evidence could reflect the evolution features in an enhanced time resolution and lay the foundation for clinical decision-making.However, quantitative evidence for metastatic timing of synchronous MF-HCC has yet to be established.In addition, preneoplastic lesions arising from cirrhosis may be significantly predisposed to multiple de novo tumors, as posited in a 'field effect' hypothesis [9].The process by which preneoplastic lesions hatch multicentric tumors (i.e., multicentric tumorigenic potential) and the mutational footprint before the malignant transformation remained unknown.
Here, we performed whole-exome sequencing in 74 tumor samples from spatially distinct regions in 35 resected foci and adjacent noncancerous tissues in 11 patients, 15 histologically confirmed preneoplastic lesions, and six peripheral blood mononuclear cell (PBMC) samples.We additionally included a previously published MF-HCC cohort (n = 9) as an independent validation dataset.We aimed to delineate molecular features of MF-HCCs during their evolution, including the dynamic changes in mutational signatures between subclonal expansions, quantitative intrahepatic metastatic timing, and the genetic structure in the preneoplastic lesions of multicentric occurrence, and thus provides implications for the precision oncology in the clinical management of MF-HCCs.

The study subjects and multi-region sampling
We recruited 11 patients diagnosed with synchronous MF-HCC before surgery in the First Affiliated Hospital of Chongqing Medical University (Chongqing, China).Based on the Chinese Liver Cancer Classification (CNLC) [15], these chemotherapy-naive patients were offered surgical treatment since multiple lesions were confined to the same segment or ipsilateral hemi-liver without or with vascular invasion.These patients showed adequate liver function and future liver remnant.Two experienced pathologists confirmed the diagnoses of MF-HCC in 10 patients and one (p360) as multiple preneoplastic lesions (misdiagnosed clinically) (Fig. S1).No evidence for 'clinically detected' extrahepatic metastases was noted.
Significant intratumoral heterogeneity was evident in various cancers [11], and multi-region sampling in genomic studies on solid tumors for reducing sampling bias was recommended [16].We thus performed multi-region sampling for 74 regions at different spatial locations from 35 resected foci, especially for lesions with large volumes (Table S1).The ellipsoid tumor volume was calculated as V = 4  3 × a × b × c , where a and b are the half-length of two axes of the tumor measured based on the magnetic resonance imaging (MRI) or computed tomography (CT) scan, and c is the mean of a and b (Table S2).The adjacent noncancerous tissues and PBMCs were also collected.We further sampled tissues from the pathologically confirmed preneoplastic lesions adjacent to each tumor lesion in multicentric occurrence patients (Table S3).The preneoplastic lesions are lowgrade dysplastic nodules (LGDNs) or high-grade dysplastic nodules (HGDNs) occurring in the background of liver cirrhosis [17], which was histopathologically determined by HE staining in FFPE sectioning of surgically resected liver tissues with adjacent noncancerous tissues.Under a microscope, the pathologist delineated the boundaries of preneoplastic lesions in the stained sections and scraped only the marked regions for DNA extraction and sequencing.The imaging characteristics of LGDNs include clear boundaries with surrounded fibrous tissue, and the liver cells have tiny atypia (manifested as an increased cell density).Large cell changes can occur with the liver plates 1-2 layers cell thick but not containing pseudo-adenoid arrangement.HGDNs show an increased density of local hepatocytes with a boundary, while its boundary under high magnification may not be clear.Irregular trabecular hepatocytes are usually arranged in the nodules, with an increased cell density (approximately twice larger than normal) and a possible three layers of cells of the liver plate thickness.Small cells are more common, pronounced, and distinguishable from regenerated nodules, while large cells are rare [17].

DNA extraction and whole-exome sequencing
Total DNA was extracted from the formalin-fixed paraffinembedded (FFPE) tumor, adjacent non-tumor, and preneoplastic tissues, and PBMC using the QIAamp DNA Mini Kit (QIAGEN, Hilden, Germany).The library construction and whole-exome sequencing have been described in detail in our previous study [18].The average sequencing depth was 142.6 ( ± 58.8) for tissues and 38.8 ( ± 5.2) for blood samples.

Assessment of MF-HCC origins
To assess MF-HCC origins, we calculated the proportion of shared mutations in pairwise tumor samples within and between lesions.Furuta et al. [3] used 49 multifocal tumors from 23 HCC patients, as well as 50 independent liver cancer samples, to statistically establish the cutoff of 5% for the shared mutation rate to discriminate intrahepatic metastasis and multicentric occurrence, given rare shared somatic mutations among multicentric pairs and potential sequencing artifacts.The cutoff was further validated based on somatic mutations characterized by Sanger and deep-exome sequencing [3].We thus used this well-established cutoff to distinguish MF-HCCs with different clonal origins.

Quantification for tumor heterogeneity
We used the mutant-allele tumor heterogeneity (MATH)score for measuring intratumor heterogeneity according to a vector of variant allele fractions (VAFs) from a tumor sample [23].The MATH-score was calculated as the percentage ratio of the median absolute deviation (MAD) of its VAFs to the median of the distribution of VAFs, adjusted by 1.4826 [MATH_score = MAD × 1.4826/median (VAFs)].A more heterogeneous tumor with a higher MATH score tends to have a wider distribution of VAFs among all mutation loci and centers at a lower fraction.A cutoff value of 32 MATH units was used to distinguish high-from low-heterogeneity tumors [24].

Reconstruction of tumor subclones
A Bayesian clustering method-PyClone-was used for grouping nonsilent somatic mutations into putative clusters while estimating their cellular prevalence (or cancer cell fraction, CCF) and accounting for allelic imbalances (i.e., segmental copy-number changes and normal-cell contamination) [25].Using the estimated cellular frequencies of mutations, we inferred the clone tree by exhaustively exploring all possible trees, reporting those with the highest likelihood, and assigning mutations to the tree nodes using citup [26].Finally, the tumor clonal evolution was visualized by mapscape [27].We included somatic nonsilent mutations with VAF > 0.01 and a minimum sequenc- ing depth of 20 × for a high-confidence clonal analysis.The clonal mutation was defined from the estimated CCF ( ≥ 0.8), whereas the subclonal mutation with CCF < 0.8.We finally calculated the Jaccard similarity index (JSI)-a measure for gauging the similarity and diversity of tumor samples-to infer the metastatic pattern.JSI was defined as JSI = W s /(L m + L p + W s ), where W s was the shared subclonal mutations between primary and metastasis, L p was the number of the private clonal mutations in primary, and L m was the private clonal mutations in metastasis [28].Polyclonal dissemination was considered when JSI > 0.3, and otherwise, monoclonal dissemination [14].

Changes of mutational signature between subclonal expansions
A rigorous method has been developed for deciphering mutational signatures using single base substitutions (SBS) [29].First, we extracted the de novo mutational signatures using the negative matrix factorization (NMF) algorithm, where an NMF rank was estimated to be three.Signatures were considered similar to the known COSMIC signatures [30] if the cutoff > 0.85.Then, we identified the optimal contribution of COSMIC signatures to the mutational profile of the samples.Finally, a bootstrapped refitting was performed to verify the refitting stability, given the number of bootstrap iterations of 500 and the maximum difference in original versus reconstructed cosine similarity between two iterations of 0.002.All analyses were implemented in the R 'MutationalPatterns' package (v3.4) [31].To characterize the changes of mutational signatures during subclonal expansions, we stratified mutations with different cancer cell fractions (CCF) into five stages, i.e., CCF in [1.0, 0.8), [0.8, 0.6), [0.6, 0.4), [0.4,0.2), and [0.2, 0), reflecting pseudo-time in clonal evolution.The same approach was used to extract and map mutation signatures from each CCF stage, and its dynamics, along with clonal evolution, were thereby tracked.

Identification of the primary tumor site
There remain challenges in identifying the primary site of intrahepatic metastasis MF-HCC.Without an established standard, we used the tumor size and combined evidence from clonal analyses to infer the primary site.First, the tumor site with the largest volume size was generally considered the primary site.Then, we evaluated the mean value of cellular prevalence (i.e., CCF) assigned to each cluster in different sequenced samples [25].In the context of tumor evolution, all subclones were derived from the ancestral tumor clone, and the primary site harbored the cluster with the most significant CCF.We thus defined the primary site as the sequenced sample with the largest CCF.Of note, the primary site inferred from the clonal analysis was consistent with that based on tumor volume.

Inference of intrahepatic metastatic timing
To quantify the evolutionary dynamics of metastasis, we first estimated the time from primary tumor initiation to metastasis (T pm ) and from metastasis to dissection (T md ) using the proposed framework [13].The H value, defined as H = L m /L p + 1, was used to estimate the metastatic timing, where H < 20 was considered early dissemination.Then, we estimated the primary tumor size at the time of dissemination (N d ) and the likelihood of tumor evolution mode [e.g., neutral (N) or subclonal selection (S)] in primary and metastasis.Early dissemination was defined as the upper bound of N d < 10 8 cells (approximately 1 cm 3 in volume); otherwise, later dissemination (i.e., N d ≥ 10 8 ).We implemented these procedures using SCIMET (spatial computational inference of metastatic timing) [13].

Characterization of mutations in preneoplastic lesions in multicentric occurrence patients
The underlying mechanism of the 'synchronous' occurrence of multiple independent lesions was unclear.However, independent lesions from the same field were genetically correlated, and a typical 'preneoplastic clone' may arise before malignant transformation [32].Therefore, we expected to move backward to the precancer stage to investigate the molecular background for multicentric occurrence and identify genetic ancestral for the synchronously different tumor lesions.Therefore, for multicentric occurrence MF-HCC patients, we collected pathologically confirmed preneoplastic tissues adjacent to each tumor lesion for whole-exome sequencing (Table S3).In addition, we used PBMCs to surrogate normal liver tissues to identify somatic mutations accumulated in the preneoplastic lesions and combine them with mutations identified in cancer tissues for further analyses.We implemented the same somatic mutation calling and clonal/subclonal reconstruction protocol in these collected preneoplastic lesions.

External validation
We used whole-exome sequencing data from nine MF-HCC patients (NCBI Sequence Read Archive (SRA) accession number: SRP062373) [2] for independently validating the dynamic changes of mutational signatures and early intrahepatic metastatic seeding.

Distinct somatic mutational landscapes in MF-HCCs subtypes
The clinical characteristics of the 11 patients are summarized in Table 1, and pathological reviews confirmed surgically collected tumor or preneoplastic lesions (Table S1).All patients were male from 24 to 66 years (median 59) at diagnosis, and 10 had a history of chronic hepatitis B virus (HBV) infection.These patients harbored 2-6 synchronous lesions (Fig. S1) with varied tumor sizes (Table S2), and most lesions showed moderate or poor differentiation.The Child-Pugh score was mild or moderate, and AJCC (American Joint Committee on Cancer) cancer staging showed stage III in three patients.Notable clinicopathological features included signs of vascular invasion (in six patients), a highly aggressive multinodular mass along with the portal vein tumor thrombosis (p551), a primary tumor accompanied by satellite nodules (p424), and a nodule-in-nodule phenotype (p331).Eight patients exhibited local recurrence with an average of 24 months of follow-up, possibly due to a high recurrence potential [33].Particularly, other intrahepatic metastatic deposits may have existed below the limits of clinical detection, and there is a higher risk for de novo recurrence under the background of liver cirrhosis.
Using the threshold (5%) of shared mutations in any two samples from different lesions (Fig. S2), we classified these patients into three subtypes, including intrahepatic metastasis (n = 3), multicentric occurrence (n = 6), and the mixed intrahepatic metastasis and multicentric occurrence (n = 2).The mixed class suggested that different mechanisms may have operated on multiple lesions during the tumor development.However, the subtype was not associated with clinical variables partly due to the small sample size (Table S4).

Dynamic changes of mutational signatures in the development of MF-HCCs
Three de novo mutational signatures contributed moderately but differed in three subtypes, e.g., the SBS22-like signature (aristolochic acid (AA) exposure) enriched in intrahepatic metastasis, the SBSA-like signature in multicentric occurrence, and the SBS5-like signature (tobacco smoking) in the mixed subtype (Fig. 2a).However, most multicentric occurrence MF-HCCs did not exhibit a dominant signature, indicating that its etiology is primarily unknown (Fig. 2b).Refitting the mutation type with the COSMIC signatures showed mild or moderate contributions (< 0.5), including SBS22, SBS5, SBS24 (aflatoxin exposure), SBS25 (unknown etiology), and SBS26 (defective DNA mismatch repair) (Fig. S5).In p424, SBS22 and SBS24 were exclusively dominant among all tumor sites, coinciding with their intrahepatic metastases showing similar mutational signatures in each site.In summary, one or two etiologies were dominant in each patient, although the heterogeneity of mutational signature was less pervasive than genetic aberrations between and within different subtypes.
Dynamic changes in the mutational signature underlying molecular evolution may demonstrate the shift of extrinsic or intrinsic etiology in shaping the cancer progression.Of the 74 samples with sufficient SNVs, 58 (78%) had an activity change (> 6%) in one or more signatures.We detected an average of 1.82 changes in the activity of mutational signatures per sample.Mutational signature activity was unstable between subclonal expansions, especially in intrahepatic metastasis patients (Fig. 2c).However, in the mixed subtype, we did not note that the activity of mutational signature changed significantly (middle panel in Fig. 2c), possibly due to the mixed phenotype exacerbating difficulty in tracking its etiology change.Specifically, SBS22, SBS24, SBS29 (tobacco chewing), and SBS89 (unknown) changed significantly during the tumor evolution (Fig. 2d).In intrahepatic metastasis patients, SBS29 increasingly contributed to the subclonal expansions (+ 18%), highlighting its role in driving cancer progression over time.However, the contribution of SBS22 (AA exposure) decreased dramatically (− 45%, left panel in Fig. 2d), suggesting a potential role of AA exposure in early carcinogenesis rather than in advanced stage.In multicentric occurrence, the increased SBS22 (+ 11%) supported its role in cancer initiation (right panel in Fig. 2d).The exact change was also noted in SBS24 (aflatoxin exposure), i.e., decreased in intrahepatic metastasis (− 6%) but increased in multicentric occurrence (+ 9%).Similar changes in SBS signatures (SBS22 and SBS24) highlighted the roles of AA and aflatoxin exposure in the early carcinogenesis of MF-HCC in the independent validation cohort (Fig. S6a-b).

Quantitative evidence for early intrahepatic metastatic timing
The reconstructed tumor clones/subclones and their phylogeny revealed either linear (p551) or branching evolution (p221 and p424) in MF-HCCs (Fig. 3a and Fig. S7) and highlighted that the metastases were disseminated from the spatial-specific primary site.The tumor clonal components differed remarkably among interfocal samples than intratumoral samples.A complex metastatic pattern (i.e., with both linear and branching evolution) was evident in the mixed subtype (p331 and p460) (Fig. 3b).Of note, two samples in p460 constituted 'pure' tumor clones (i.e., red in S3_A2 and yellow in S5_A1), indicating a sign of independent origins.The refined spatially clonal evolution pattern enabled identifying the specific lesion in the primary site for initiating metastasis.For example, metastatic sites (S8_B{1,2} and S8_C{1-4}) in p460 were seeded from a specific region (S8_B3) within the inferred primary lesion (S8_B), whereas in p331 with a 'nodule-in-nodule' phenotype, one metastatic satellite tumor (S6_C) harbored multiple clones shared with S6_A{1-3}, seeded from S6_A1.The Jaccard similarity index (JSI) further supported the metastatic pattern, i.e., a low JSI indicating a linear (or monophyletic) evolutionary pattern (p551) and a high JSI indicating branching (or polyphyletic) dissemination (p424 and p331) (Fig. 3c).
Metastatic seeding was initiated when the primary tumor was most likely at 10 -4 -0.01 cm 3 in volume (corresponding to 10 4 and 10 6 cells) (Fig. 4a), where a volume of 1 cm 3 (at a number of cells of 10 8 ) was usually required to be clinically detected.For example, the inferred primary site (S8_A) in p221 initiated metastatic seeding at 10 5 cells with a similar mutation rate ( = 0.6, per cell division in exonic regions).Two evolutionary scenarios were exhibited in primary/ metastasis (P/M) pairs, i.e., selection/neutral (S/N) (e.g., all P/M pairs in p424) and selection/selection (S/S) (e.g., six out of eight P/M pairs in p221) with selective subclonal evolution.A relatively lower H value (< 20) in all P/M pairs also supported early metastatic seeding (Fig. 4b).The estimated median time from primary tumor initiation to metastasis (T pm ) and from metastasis to dissection (T md ) was 253 (range 202-304) and 175 (123-203) days, respectively (Table S7).The H value was positively correlated with N d (Fig. 4b) and T pm (Fig. 4c), suggesting that a larger size of the primary tumor was related to a long time from initiation to dissemination due to the required time for tumor cell proliferation (Fig. 4d).One independent validation cohort also confirmed our findings that early metastatic seeding in most metastatic sites (in six out of nine intrahepatic metastatic patients [2]) (Fig. S8).Our findings provided direct quantitative evidence that early metastatic seeding before clinically detected was relatively common in intrahepatic metastasis MF-HCC.

Common mutations in preneoplastic lesions in multicentric occurrence MF-HCC
Using PBMC as a surrogate for normal liver, we noted that somatic mutations accumulated in preneoplastic lesions differed from tumors (Fig. S9a-b).The mutational burden was significantly greater than in cirrhotic tissues in 12 solitary HBV-associated single primary HCC patients in our previous study [18] (p = 0.01) (Fig. S9c).Of common somatic mutations identified in the preneoplastic lesions (Fig. 5a), five patients showed a proportion of shared nonsilent somatic mutations > 5% among preneoplastic lesions (Fig. 5b), suggesting that the divergence of independent lesions may have occurred after malignant transformation.Furthermore, somatic mutations in several known HCC drivers (e.g., APOB, ALB, BIRC6, AKAP9, and BRCA2) 1 3 were identified in preneoplastic stage (Fig. 5c), which were enriched significantly in canonical cancer signaling (e.g., Rap1, PI3K-Akt, and MAPK) and ECM-receptor interaction pathways (Fig. S9d).These results highlighted their carcinogenesis potential in the preneoplastic stage.The comparison of SBS signatures between preneoplastic lesions (Fig. S9e) and tumor stage (Fig. S5) showed that, with the progression from preneoplastic lesions to cancer, the dominant SBS6 (defective DNA mismatch repair) decreased, whereas SBS22 (AA exposure)-hardly detected in the preneoplastic lesions-emerged in cancer tissues (Fig. S4).Therefore, an internal defect in DNA mismatch repair and exposure to exogenous AA and tobacco chewing may underline carcinogenesis in multicentric occurrence MF-HCC.
Given commonly accumulated mutations in preneoplastic lesions (Fig. S10a-b), we hypothesized that 'preneoplastic arising clones' could be identified in the preneoplastic lesion and its corresponding tumor sample.Therefore, we reconstructed the clonal structure by combining somatic mutations by comparing tumor samples and preneoplastic lesions with blood.In p090, clone #9 (i.e., a preneoplastic arising clone), as the red arrow indicated, was typically identified in preneoplastic and tumor tissues with a relatively higher CCF, indicative of its role in the preneoplastic stage (Fig. 5d).The observation was applied to clone #5 in p360 and clone #11 in p058 (Fig. 5d).Although a preneoplastic arising clone was not apparent in p900, preneoplastic lesions (i.e., S6_A_Pre and S8_A_Pre) demonstrated shared nonsilent mutations (Fig. 5b, c).We could not detect a significant clone in p465 and p406, partly due to insufficient sampling for spatially heterogeneous tumors.HCC-related driver genes commonly mutated in preneoplastic lesions and their roles underlying hepatocarcinogenesis were summarized in Table S8.We provided initial evidence that preneoplastic lesions in multicentric occurrence patients were genetically correlated before the malignant transformation.

Discussion
Whole-exome sequencing for synchronous MF-HCC cohort allowed us to investigate the molecular evolution underlying its development.In the present study, we longitudinally characterized the distinct dynamic changes in mutational signatures among different MF-HCC subtypes, provided quantitative evidence for early intrahepatic metastatic seeding, and demonstrated that 'preneoplastic arising clones' carrying oncogenic drivers were predisposed to multicentric carcinogenesis.Validating our findings in an independent MF-HCC cohort increased the robustness of our conclusion.
The post-expansion mutations carry the signature of subclonal expansions that activate the mutation process [11].Thus, dynamic changes in mutational signatures may enhance our understanding of the varied etiologies underlying the progression of MF-HCC.In addition to confirming the AA exposure (SBS22) [37,38], we demonstrated that the contribution of AA exposure sharply declined between subclonal expansions in intrahepatic metastasis patients (Fig. 2).Nevertheless, it inclined with tumor evolution in multicentric occurrence patients, indicating its role in the initial carcinogenesis of MF-HCC.These results suggested a prevention strategy for exogenous carcinogens (e.g., aflatoxin exposure and tobacco smoking).Additionally, banning the prescribed some AA-containing herbal remedies for HBV infection is required, especially in East Asia [39].
We demonstrated that metastatic sites were seeded from spatial-specific primary sites in synchronous MF-HCCs (Fig. 3).We also provided quantitative evidence for early intrahepatic metastasis, where metastasis was disseminated when the primary lesion was in 10 -4 -0.01 cm 3 (Fig. 4).Early metastatic seeding may occur in colorectal cancer [13], lung cancer [40], and pancreatic tumors [41,42].In addition, intrahepatic metastases and tumor thrombi can occur early in HCC progression based on the distance between the intrahepatic metastatic site and primaries [2].Compared with 5-10 years from initiation to being clinically detected in other cancers [12], a much shorter time (< 1 year) from the most recent common ancestor to dissemination or from dissemination to surgery was inferred in intrahepatic metastasis patients indicating their aggressive history.Due to a higher risk of systemic metastases, tumor staging for intrahepatic metastasis MF-HCC might be modified by considering early metastasis.Patients with intrahepatic metastasis were staged as IIIa (p551 and p331) and IIb (p221, p424, and p460), more advanced than patients with multicentric origins, three of which were staged as Ib.These findings suggested that staging for intrahepatic metastasis MF-HCC might be modified by considering early intrahepatic metastasis.
Although multicentric occurrence has been suggested to arise from different preneoplastic lesions of the cirrhotic liver [43], it may not be thoroughly independent during its evolutionary history (Fig. 5).Leveraging 'somatic' mutations characterized in the preneoplastic lesions (i.e., compared with blood samples) and clonal analysis, we identified common 'preneoplastic clones' arising in the preneoplastic stage.Mutations in common preneoplastic clones enriched in the cancer-related pathways demonstrated their carcinogenesis potential before the malignant transformation.Thus, the multicentric occurrence may not be thoroughly independent across all stages during its carcinogenesis.However, different preneoplastic lesions Fig. 3 Intrahepatic metastasis occurred in the spatial-specific site.Clonal inference from multi-region samples indicated the metastasis occurred in spatial-specific sites in MF-HCCs patients with intrahepatic metastasis (a) and the mixed intrahepatic metastasis and multicentric occurrence (b).The primary lesion was marked with a dashed red circle.Distinct clones were present in different colors with numbers for each patient.The clonal composition was scaled to the fraction of each clone in each sample.'0' represented the ancestor clone.Samples were zoomed in on the inner circle around the anatomical diagram, where the outer circle was constituent clones from the overall clonal tree.c The Jaccard similarity index (JSI) for each tumor sample ◂ harboring varied adaptive mutations could be transformed into independent tumor lesions, as posited in the hypothesis of 'mutual exclusivity of oncogenes' [44].Identifying such mutations in preneoplastic lesions can enable riskstratifying these preneoplastic lesions that require intervention.In non-alcoholic fatty liver disease-associated HCC patients, somatic mutations and epigenetic changes were identified in the background liver, and thus, rigorous surveillance for the emergence of HCC in such cases was required [45].
There were several limitations.First, the present study included a relatively small patient number, collected within a two-year window in a single center.However, it was appropriate for our research with a high-resolution sampling and sequencing scheme combined with a publicly available cohort [2] as validation.A large prospective MF-HCC cohort is required to obtain more solid conclusions.Second, the results may not be generalized to MF-HCC patients without HBV infection.Including patients with other etiologies or risk factors (e.g., non-alcoholic fatty liver disease, alcohol intake, and diabetes) may further understand their potentially different evolutionary scenarios.Third, the HBVrelated MF-HCCs need further genome-to-genome analysis to characterize their co-evolution by tracking the HBV monoclonality and its expansion along with MF-HCC evolution, which may highlight the genetic determinants from both the host and virus underlying MF-HCC development [46,47].Finally, a complete delineation of genomic characteristics in MF-HCC remains challenging due to significant spatial heterogeneity.Of note, the tremendous spatial heterogeneity,

Conclusion
In summary, our study highlighted significantly varied tumor clonal evolutionary history underlying different subtypes of MF-HCC.This study presents a novel insight into the MF-HCC progression and may accelerate its personalized clinical management.

Fig. 1
Fig. 1 Tumor genomic heterogeneity in MF-HCC patients.a The nonsilent mutational landscape in sequenced tumor samples.The y-axis showed nonsilent mutations identified in all samples.b The nonsilent mutational burden in each sequenced sample.VC, the portal vein tumor thrombus.Different colors represented anatomical loca-

Fig. 2
Fig. 2 Mutation signatures exhibited dynamic changes along with MF-HCC evolution.a Principal component analysis for de novo mutational signatures.b Negative matrix factorization estimated three

Fig. 4
Fig. 4 Quantitative evidence for early intrahepatic metastatic timing.a The inferred metastatic timing.N neutral evolution, S subclonal selection.P primary tumor, M metastasis tumor, N d the primary tumor size (cell number) at metastatic seeding.b The correlation

Fig. 5
Fig. 5 Genetic footprints in the preneoplastic lesions in multicentric occurrence patients.a A nonsilent mutations landscape in the preneoplastic lesions.b The rate of shared nonsilent mutation in any two preneoplastic lesions.c Common nonsilent mutations were identified

Table 1
Clinical characteristics of collected patients with synchronous multifocal hepatocellular carcinoma See Table S1 for additional clinical information VC vein thrombus, * cirrhotic nodules, # the patient was diagnosed with multifocal preneoplastic lesions after pathological review, AJCC American Joint Committee on Cancer, MVI