63rd ASH Annual Meeting & Exposition – a Postscript

In molecular genetics, Walter et al. show how the diagnostic spectrum of next-generation sequencing (NGS) can be expanded using a target enrichment process to allow losses in heterozygosity (CN-LOH, copy number neutral loss of heterozygosity) to be detected as well. The prognostic importance of this is gaining in recognition. A commercially available panel for detecting changes in chromosome copy numbers was used for the purposes of enrichment. By combining target enrichment and NGS, CN-LOH was able to be detected with sensitivity and cost-effectively at a shallow depth of sequencing. In a patient cohort of 1,196 patients, 10% had at least one instance of CN-LOH, with 4q, 7q, 9p, and 11q being affected with particular frequency. Another investigation by Hörmann et al. made it clear that expanding the NGS panel for myeloid neoplasia to include the PIGA gene also has diagnostic and clinical relevance in cases where PNH is not suspected. In a cohort of 20,320 patients examined using this panel, a PIGA mutation was found in 67 patients. Of the 37 patients without a history of PNH, 20 patients were also examined by immunophenotyping, showing that a PNH clone was detectable in the vast majority of cases (85%). Truncating PIGA mutations were highly specific for the presence of a PNH clone.

In immunophenotyping, it was possible for findings from a prospective, multicenter study to lead to the optimization of MDS diagnostics. MDS patients exhibit aberrant immune phenotypes, which, however, are not defined by a single characteristic marker and can instead involve a large number of markers in different cell lines. Only by looking at it as a whole can it be assessed whether the findings of the immunophenotyping are consistent with a diagnosis of MDS. As Kern et al. have shown, only 17 markers contribute significantly to flow cytometry findings of MDS. If an aberrant expression profile can be determined for three of these markers, the correspondence with the cytomorphological diagnosis is already 80%. The data also indicates that the blast cell threshold – currently > 5% and also an important indicator for an immunophenotypic diagnosis of MDS – can be corrected to > 3%.

The use of AI remains an important topic at the MLL and was once again a major emphasis at the annual ASH conference this year. A total of four presentations on AI were given there, all of which were groundbreaking with regard to the application of AI in routine clinical diagnostics.

Concerning cytomorphology, two projects involving the automatization of blood and bone marrow smear differentiation were presented this year. In a prospective study, presented by Haferlach, T. et al., 10,082 blood smears were obtained both by highly qualified medical technicians as well as by using a tool fully automated from scanning to analysis that had been developed together with MetaSystems and Amazon Sagemaker. The correspondence between the two methods was a stunning 95% for pathogenic cases and will be significantly improved in the future through further training, including with rare cell types. The automated imaging of individual cells from bone marrow smears was also successfully established this year by Pohlkamp et al. – a major step toward automated single-cell image classification, remote diagnosis, and the generation of image galleries for human review and revision. The automatic classification of bone marrow smears will continue to develop rapidly through real-time training based on dynamic data sets.

With immunophenotyping, it has now become possible to analyze and evaluate raw data from flow cytometry matrices using AI, as the project presented by Bellos et al. confirmed. Various machine learning models have already been employed for this purpose: XGBoost, weighted SVC and LinearSVC, hierarchical model, and AutoGluon. This has made it possible for impressive R and P values to be achieved. In the future, the focus will not only be on improving existing models but also on identifying further sub-entities and applying transferred learning to enable the universal use of the tool for flow cytometry data.

One exciting project in the area of molecular genetics was presented by Nadarajah et al., in which WGS and WTS data is interpreted using an AI tool. The goal here is to initially predict a final diagnosis without any human input. The complexity of such data exceeds the capabilities of purely manual analysis, meaning that an AI tool is able to compensate for any loss of data. It was possible for an overall precision of 85% to be achieved, with hard-to-distinguish entities such as MGUS/MM or MDS-EB-2/AML/CMML playing a major role. In independent tests, a precision of up to 100% was able to be achieved for clearly defined entities. The tool will be made available via a web application. The results of the automated decisions are presented transparently in the form of a clear visualization, thus ensuring comprehensibility for users.

WGS and WTS combine the diagnostic disciplines of chromosome banding analysis (CBA), fluorescence in situ hybridization (FISH), and molecular genetics. Like CBA, both WGS and WTS allow genome-wide insights – albeit at a nucleotide base size resolution. As the previously presented study by Nadarajah et al. underscores, genetic characterization using genome-wide methods is so comprehensive that it has already enabled AI-assisted classification of diseases even today. Along with this, research on genome-wide data is also increasing our knowledge of pathogenetic mechanisms.

For example, a prognostically negative influence of TP53 aberrations (due to mutations, deletions, or CN-LOH) has been described and validated for various forms of neoplasia. In one large cohort of 4,646 patients, Stengel et al.systematically evaluated the frequency and composition of TP53 alterations and their clinical impact on 29 cases of neoplasia. It was detected in 13% of these patients, revealing a general association between lymphatic entities and TP53 aberrations. In addition, a clear association between TP53 alterations and a complex karyotype was detectable for the entire cohort as well. For HGBL, MZL, and T-NHL, a TP53 aberration did not affect overall survival. There was a negative prognostic influence in patients with MCL and MPAL, but this was independent of whether one or both alleles (“double hit”) were affected. For the remaining entities, the presence of a TP53 alteration shortened overall survival, an effect that was increased for cases with a double hit. The study by Wahida et al. furnishes an interesting insight into the biological impacts of TP53 mutations. In this study, telomere length and telomerase activity were studied using WGS and WTS in patients with AML, MDS, and PNH and compared with data from healthy subjects. Due to the high level of replication stress to which strongly proliferating tumor cells are exposed, shortened telomeres usually represent a characteristic feature of cancer. Contrary to this expectation, elongated telomeres were found in one subgroup of the AML cohort in which an association with TP53 mutations was discovered. Moreover, there was also a correlation between telomere length and the TP53 mutation burden as well as telomerase activity andthe TP53 mRNA level.

A diagnostic gap was able to be filled by WGS. Up until now, it was not possible to detect microdeletions either by sequencing or by chromosome analysis. Only the use of FISH probes permits a targeted detection, although not comprehensive screening. The abstract by Baer et al., for which various myeloid entities were tested for microdeletions, showed that this can be achieved by means of WGS. RUNX1 and TET2 were the genes most frequently affected. Other deletion mutations were found mainly in genes for which a loss-of-function profile is known from myeloid neoplasia. In addition to mutations, microdeletions represent yet another mechanism by which functional gene loss can occur.

Can WGS and WTS measure up to the current gold standard in order to establish themselves in clinical routine? The abstracts by Haferlach, C. et al., Truger et al., and Hörmann et al. provide a clear answer for a range of different entities. For acute leukemia and multiple myeloma, there is a high level of correspondence between findings from routine diagnostics and WGS/WTS. In the vast majority of cases, alterations not detected by genome-wide methods were attributable to small clone sizes or a low mutation burden. In addition, duplications of the entire set of chromosomes (e.g., tetraploidy) cannot be identified using WGS. Against this stand the benefits of genome-wide methods. In the case of ALL, this already starts during the processing of samples – while the low in vitro proliferation of the ALL cells is a limiting factor for cytogenetics, only purified nucleic acids are needed for WGS and WTS. In addition, the ALL subtypes of Philadelphia-like ALL and of ALL with DUX4 rearrangement can be identified easily using WTS, while they are difficult to map with the current gold standard. What is more, rare and cytogenetically cryptic rearrangements along with numerous chromosomal changes can also be detected using WGS/WTS. Last but not least, by providing information on the mutation status, WGS also enables the search for molecular genetic markers relevant to diagnosis, prognosis, and therapy. For mastocytosis, detecting the characteristic KIT D816V mutation using WGS was only possible in 21% of the patients. Here, too, the generally low mutation burden is assumed to be the cause. Even for the example of mastocytosis, however, the clear advantage of the genome-wide methods was the comprehensive genetic characterization they provide; at least one non-KIT mutation was able to be detected in 46% of the patients, while cytogenetic aberrations were detected in 21%. The detected alterations also had an impact on the prognosis; patients with non-KIT mutations and chromosomal changes exhibited the shortest overall survival. If there was only one type of aberration (mutation or chromosomal change), survival was already reduced compared to patients without any aberrations.