Specific lncRNA signatures discriminate childhood acute leukaemias: a pilot study

The procedures followed in the present study are in line with the Helsinki declaration and have been approved by the local ethical committees of the IRCCS-SDN (Ethical Committee IRCCS Pascale, Naples, Italy—protocol number 5/19 of the 19/06/2019) and the AORN Santobono-Pausilipon (Ethical Committee Cardarelli/Pausilion, Naples Italy—protocol number 07/20 of 03/06/2020). Both parents signed informed consent and all participants provided informed assents. All children enrolled in the study were included at moment of diagnosis, patients’ clinical features are presented in Tables 1 and 2 and Additional file 1: Dataset S1 (B-ALL patients) and in Tables 3 and 4 and Additional file 2: Dataset S2 (T-ALL patients).

Total RNA was extracted from leukemic cells derived from Bone Marrow blood of B-ALL and T-ALL patients and purified B lymphocytes and T lymphocytes from cord blood of healthy donor using Trizol (Thermo Fischer Scientific, Waltham, MA, USA) reagent protocol, according to manufacturer instructions. RNA concentration and quality were determined using Qubit (ThermoFisher Scientific, MA, USA) spectrophotometer. RNA-seq libraries were prepared with 3′-DGE approach and sequenced SEx100 on an Illumina Novaseq platform.

FASTQ files were aligned with STAR v. 2.7.1a [22] on the GRCh38 human genome. Raw counts were obtained using HTSeq v. 2.0.0 [23]. Normalisation and differential expression analysis were performed with DESeq2 v. 1.36.0 [24]. LncRNA annotations were done with Biomart v. 2.52.0 [25]. Hierarchical clustering and heatmap representations were performed as in Buono et al. [26]. Functional enrichment analysis of lncRNA clusters were analysed with gProfiler [27]). Expression correlations were calculated with “pearson” method in R. Statistical significance of gene overlapping were calculated using Fisher’s exact test through the R package GeneOverlap v. 0.99.0. GO enrichment analyses for correlated genes were performed with EnrichR [28] using as input the list of all the positively correlated genes with p-value < 0.001.

Total RNA was extracted from B-ALL, T-ALL and PBMCs derived from cord blood using the Trizol Reagent protocol. After extraction, RNA was quantified using NanoPhotometer NP80 (Implen, USA). Next, 1 µg of total RNA from each sample was reverted in cDNA using SuperScript III First-Strand Synthesis SuperMix kit (Thermo Fisher Scientific) according to the manufacturer’s protocol. The expression level of selected lncRNAs was measured by qRT-PCR using the following formula: 2-∆Ct on C1000 Touch Thermal Cycler (Bio-Rad, Hercules, CA, USA) using iQ SYBR Green Supermix (#1708882, Bio-Rad). Ribosomal Protein S18 (RPS18) level was used as an endogenous control to normalize lncRNAs expression. The following primers were used:

RPS18: fw 5′ - CGATGGGCGGCGGAAAATA-3′; rev 5′—CTGCTTTCCTCAACACCACA-3′

LINC00958: fw 5′ -TGCAGCAAGATAGCTCCAGG-3′; rev 5′- CCTGGCGTCTGTGTAGTGTT-3′

LINC00114: fw 5′- TAGAGGCCTGATGGAGTGGA-3′; rev 5′- CTGCCCAGGAAACTGTAGGT-3′

AL713998: fw 5′- AACATTTGGTGCCGAAAGCC-3′; rev 5′- GCGAGGGAAGTCTCTTGCAT-3′

AC008060: fw 5′- CGAGGCTTGGACAAATGCAG-3′; rev 5′- CAGTCCCAAAGGAAGCGGAT-3′

AL590226: fw 5′- GAATCCACAGATGGCGTGTG-3′; rev 5′- TCAGGTAGCTGCGAGTTCAA-3′

PCAT18: fw 5′- GTC CCA GCA CTT CAC TGG TT-3′; rev 5′- AGC TGG GAT ATG GTA GCA GC-3′

HHIP-AS1: fw 5′-TCA CAC CAC CAC TGA GCA AC-3′; rev 5′- AGC TCT GCT TGG TGA ATG GA-3′

AC247036: fw 5′- TGT CCT GTG GTG GGA AAA ACA-3′; rev 5′- ACC CGG GAG TCA TCT GAA CA-3′

LINC01222: fw 5′- AGCAGGGGTAACATTATGGGC-3′; rev 5′-AGC TGC TCC CCC TTT ATC TTC-3′

AC116351.1: fw 5′- TGGAAAGTCCAGCGACAGAC-3′; rev 5′: GTCTCCCTTCACAGTGGCAA - 3′

To gain insight into the transcriptomic landscape of childhood acute lymphoblastic leukaemia, we performed RNA-seq on RNA samples from the bone marrow of paediatric patients. We used purified naïve lymphocytes from cord blood as healthy control. In the case of acute T-cell lymphoblastic leukaemia (T-ALL), we sequenced 6 childhood T-ALL samples and 7 purified naïve T cells from cord blood (T-ALL patient clinical details in Additional file 2: Dataset S2). For acute B-cell lymphoblastic leukaemia (B-ALL), we used 9 samples from patients and 9 purified naïve B cells from cord blood (B-ALL patient clinical details in Additional file 1: Dataset S1). Differential expression analysis between leukaemia patients and the relative healthy controls revealed a comparable number of significant upregulated and downregulated genes (adjusted p-value < 0.05) in the two subtypes of acute lymphoblastic leukaemia. In T-ALL, we observed 2734 upregulated and 1998 downregulated genes (complete T-ALL differential expression analysis can be found in Additional file 3: Dataset S3). These differentially regulated genes include 187 upregulated and 164 downregulated lncRNAs. From the comparison between B-ALL patients and the healthy controls, we obtained 2852 upregulated genes and 1833 down-regulated ones (complete B-ALL differentially expression analysis can be found in Additional file 4: Dataset S4), including 181 and 216 lncRNAs, respectively (Fig. 1A). When comparing the landscape of upregulated lncRNAs in B-ALL and T-ALL, we noticed a very poor signature overlapping, with only 38 lncRNAs in common, whereas 143 and 149 are specific for the B-ALL and T-ALL disease, respectively (Fig. 1B). The overlapping of 38 lncRNAs resulted to be not significant, while the same analysis for the whole transcriptome or only for the protein coding genes showed a significant greater proportion of overlapping genes (Additional file 11: Fig. S1A and B). This data points to a sharper specificity of lncRNA core underlying oncogenic events in each type of childhood acute leukaemia. Based on this result, we performed a principal component analysis (PCA) of the whole lncRNA landscape in the four conditions examined. The graphical representation of the analysis shows a clear separation of the four groups in distinct clusters, indicating the expression of a unique lncRNA signature characterising each condition (Fig. 1C). PCA of the whole transcriptome and protein coding genes scored a slightly wider separation (Additional file 11: Fig. S1C and D), but this can be attributed to the larger number of genes taken in account for clustering, since lncRNAs represent a minority portion of the transcriptomic landscape compared to protein coding genes.

After a general characterization of the transcriptomic landscape of childhood T-ALL and B-ALL, we performed an unsupervised hierarchical clustering analysis of all the lncRNAs significantly differentially expressed in at least one of the following comparisons: T-ALL vs. T-healthy, B-ALL vs. B-healthy or T-ALL vs. B-ALL. This method of selection of the input lncRNAs for the hierarchical clustering analysis was chosen to eliminate all the lncRNAs that are not likely to be neither directly involved in the leukemogenesis of B and T lymphoblasts nor vary between the two types of disease and thus could have generated a high level of background, lowering the sharpness and accuracy of the analysis. This approach identified 4 main expression clusters, which are distinctive for each condition, delineating a specific molecular signature of the disease capable of distinguishing not only between healthy and ALL patients but also between B-ALL and T-ALL patients (Fig. 2A, complete list of lncRNA molecular signature in Additional file 5: Dataset S5). This analysis not only defined the lncRNA signature for each condition but also aggregated the samples in two sub-groups based on similarity, showing a higher resemblance between the two ALL signatures and the two healthy signatures than between the ALL subtype and its respective control. Functional enrichment analysis for the list of lncRNAs belonging to T-ALL cluster (cluster 2) returned no significant enrichment besides expected gene ontology terms such as “regulatory RNA binding” (GO:0061980) or “negative regulation of miRNA-mediated gene silencing” (GO:0035198) (data not shown). This may also be due to the poor characterisation and annotation in dedicated databases of the lncRNAs being investigated. Yet, interestingly the same analysis for the lncRNAs belonging to B-ALL signature (cluster 1) outputted the WikiPathway [29] term “CCL18 signalling pathway” with adjusted p-value = 6.060 × 10^–4 (data not shown). CCL18 is a chemokine produced mainly by antigen-presenting cells of the innate immune system, such as dendritic cells, monocytes, and macrophages. A study from Catusse et al. suggested that CCL18 might be an important factor interfering with pathophysiological homing and maturation processes of B-ALL cells through the chemokine receptor CXCR4 [30]. Our data corroborate this hypothesis and suggest that this process could be mediated by lncRNAs part of the identified B-ALL signature.

We selected 12 lncRNAs that have never been associated with childhood ALL previously for further validations, 6 from the group of specific lncRNAs for B-ALL (cluster 1: LINC00958, LINC00114, AL713998.1, AC008060.1, AL590226.1, AC002464.1) and 6 from the group of specific lncRNAs for T-ALL (cluster 2: AC002454.1, PCAT18, HHIP-AS1, AC116351.1, AC247036.1, LINC01222). All these lncRNAs were extremely upregulated when comparing B-ALL or T-ALL with their relative healthy control (Fig. 2B and C). From our transcriptomic data, most of these selected lncRNAs can also discriminate between B-ALL and T-ALL; AC002464.1 (B-ALL signature), LINC01222, and AC002454.1 (T-ALL signature) are not differentially expressed between the two subtypes of acute lymphoblastic leukaemia (complete differential analysis B-ALL VS T-ALL can be found in Additional file 6: Dataset S6).

To confirm our omics analyses, we decided to test selected lncRNAs expression levels in a different cohort of 10 B-ALL and 10 T-ALL patients compared to PBMCs derived from cord blood. Figure 3 (left panels) reported that four of five selected lncRNAs (LINC00958, LINC00114, AL713998, and AC008060) were significantly over-expressed in B-ALL patients compared to both healthy and T-ALL samples. No significant differences occur in the expression level of these lncRNAs comparing T-ALL patients and Healthy subjects. Only for one selected lncRNA (AL590226), the bioinformatic analyses were not confirmed since its expression levels were not significantly higher in B-ALL patients with respect to Healthy subjects and T-ALL patients. Just as for de-regulated lncRNAs in B-ALL patients, we also performed validation for deregulated lncRNAs in T-ALL patients using the same study cohort.

As reported in Fig. 3 (right panels), the lncRNA PCAT18, HHIP-AS1, AC247036.1, and LINC01222 results were significantly higher in T-ALL patients with respect to Healthy subjects and B-ALL patients. Also, in this case, for one of the selected lncRNA (AC116351.1) no significant differences were observed through the three populations analysed.

We correlated the expression levels of the analysed lncRNAs with patient clinical information. Interestingly, despite our limited sample size, we found a significant correlation of HHIP-AS1 expression levels with patient age; the higher lncRNA expression, the younger the patients. Further, AC247036.1 expression negative correlates with WBC counts at diagnosis (Additional file 11: Fig. S2).

To increase our statistical power, we tested our findings in a larger cohort of childhood leukaemia patients using RNA-seq data from Paediatric Cancer Genome Project by Saint Jude Children’s Research Hospital (https://​platform.​stjude.​cloud/​data/​cohorts/​pediatric-cancer) [31]. We compared 80 B-ALL, 25 T-ALL, and 38 acute myeloid leukaemia (AML) samples. Also, in this case, the PCA analysis of lncRNAs’ panorama shows clear segregation of the different conditions, suggesting that the lncRNA signature could help distinguish among the three types of childhood leukaemia disease and not only between B-ALL and T-ALL. Surprisingly, the lncRNA signature of T-ALL patients appears to be more similar to the signature of AML patients than that of patients affected by the other type of acute lymphoblastic leukaemia examined, B-ALL (Fig. 4A). This data was confirmed by the hierarchical clustering of all the lncRNAs in B-ALL, T-ALL, and AML, where the dendrogram of the three conditions shows T-ALL and AML separating from B-ALL (Fig. 4B). The candidate lncRNAs individuated as T-ALL biomarkers from our RNA-seq analysis and following real-time qPCR validations all belong to cluster 1, except for HHIP-AS1 belonging to cluster 3 (complete list of clusters in Additional file 7: Dataset S7). However, this data is not in contrast with our previous findings, but on the contrary, it highlights the low expression of HHIP-AS1 in AML. The upregulation of HHIP-AS1 in T-ALL compared to AML is confirmed by the differential expression analysis pointing out a significant and consistent upregulation of all the T-ALL lncRNA biomarker candidates (Fig. 4C). Further, all the lncRNAs previously selected as T-ALL biomarkers were part of cluster 5, except for AC002464.1, which belongs to cluster 2. However, this lncRNA was unable to discriminate between B-ALL and T-ALL (Fig. 2D), qualifying as a general biomarker of leukemic transformation compared to the healthy control, not specific for any of the examined paediatric leukemias. Indeed, subsequent differential expression analysis showed that AC002464.1 is even upregulated in AML. In contrast, all the other B-ALL selected lncRNAs are significantly upregulated in B-ALL, with log2(Fold Change) always greater than 1.

We focused our further bioinformatic analysis on the lncRNAs selected from T-ALL signature, due to both the particular aggressiveness of this disease and its rarity in paediatric age, which makes it even more difficult and scarcely feasible to study the molecular dynamics underlying its physiopathological mechanisms. For each T-ALL candidate lncRNAs individuated by our bioinformatics analyses, we calculated expression correlation with any other gene using all the Paediatric Cancer Genome Project samples by Saint Jude Children’s Research Hospital. This test highlighted some interesting correlations between the expression of the lncRNAs, ALL marker genes, and important players in the oncogenesis process (complete lists of correlated genes in Additional file 8: Dataset S8). For instance, the pathway enrichment analysis for all the genes positively correlated with AC247036.1 with a p-value < 0.001 returned significant enrichment for several terms regarding TCR signalling, T cell activation and β-catenin signalling pathway (Fig. 5A). Further, PCAT18, part of the T-ALL signature, is shown to be highly significantly correlated with the expression of the delta subunit of CD3 (CD3D), a well-known marker of T cell lineage [32] (Fig. 5B), confirming the specificity of this lncRNA to discriminate malignancy of T cells from acute leukemias of other blood cell types. Interestingly, we found HHIP-AS1 to be positively correlated with its sense adjacent coding transcript, HHIP, an inhibiting factor of Hedgehog signalling [33] (Fig. 5C). This correlation could suggest an implication of this lncRNA in the Sonic Hedgehog (SHH) pathway, an important proliferative factor that can underlying T-ALL oncogenic transformation [34]. This result is in agreement with a previous study showing HHIP-AS1 to be an important regulator of SHH pathway in solid tumours [35], and suggests an analogous HHIP-AS1 role also in leukemogenesis.