Patients with sustained-CR present a TH1-enriched L-TME at AML diagnosis
We analyzed the expression profiles of patient samples from the different cohorts. We performed a DEG analysis in the NanoString assay on the Discovery Cohort between patients with sustained-CR (n = 6) vs non-sustained CR (n = 19) and identified a clear distinction between myeloid suppression-related genes and a T-cell infiltration signature represented by 67 DEGs (p ≤ 0.05) (Fig. 1a). By mapping these genes in the 2 CR groups, the T-cell signature, including cytotoxic T-cell infiltration (CD3D, CD8A, PRF1, GNLY), IFN signaling activation (IFIT3, IFITM1, IFI27, STAT1, GBP1, PARP12, TRAT1), antigen presentation and B-cell functions (TAP1, CD19) characterized patients with sustained-CR, while early loss of CR and refractory disease were marked by the macrophage-myeloid suppression signature (CD68, TREM1) and an intrinsic oncogene signaling linked to T-cell exclusion and immune suppression (WNT5A, WNT-β-catenin pathway, and EFGR) (Fig. 1b).
To further qualify this signature on a broader gene set, we first validated it with mRNA-Seq in a subset of 19 patients from the Discovery Cohort, then we confirmed it in the Internal Validation Cohort (Supplementary Fig. 1a). Finally, we combined all mRNA-Seq-based cohorts (Whole mRNA-Seq Cohort, see Table 1) to perform functional pathway analyses.
By applying t-distributed stochastic neighbor embedding (tSNE) analysis on the transcriptome of the Whole-mRNA-Seq Cohort, we observed a distinct separation pattern between sustained and non-sustained CR (Fig. 1c), indicating major biological differences between the compared groups. The IPA enrichment-pathway-analysis confirmed that patients with sustained-CR benefit from an immune-rich L-TME, with a prevalence of T helper 1 (TH1) activation pathways, IL-15 production (positive activation z-score), and inhibition of the PD1-PDL1 and myeloid TREM1 pathways (negative activation z-score), proving an association between a TH1-skewed/cytotoxic L-TME and long-term remission (Fig. 1d).
A TH1-enriched L-TME at AML diagnosis is represented by a 3-gene IFN signature
To restrict this gene signature to a robust and clinically applicable signature, we shortlisted the 3 genes common to all cohorts and most significantly upregulated in the Differential-Expression-Gene (DEG) analysis: guanylate-binding protein (GBP1); poly-ADP-ribose-polymerase-12 enzyme (PARP12) and T-cell receptor-associated transmembrane adapter (TRAT1) (Fig. 1e). Of note, all 3 genes are related to IFN: GBP1 (23, 24) and PARP12 (25, 26) are type II and type I- IFN-induced genes innate immunity enhancers with strong antimicrobial properties, while TRAT1 is involved in cytotoxic/IFNg-polarized immune responses (27) and correlates with immune cell infiltration and favorable prognosis in some solid tumors, as non-small cell lung carcinomas and Diffuse Large B Cell Lymphoma (27, 28).
Finally, we validated the identified IFN-signature in the TARGET database to evaluate its potential clinical relevance.
Also in the TARGET Cohort, the 3-genes-IFN signature successfully discriminated between children with sustained and non-sustained CR (Fig. 1f) and performed better than all other genes in distinguishing patients with the lowest Hazard Ratio (HR) (Fig. 2a). This proves that a cytotoxic and IFN-skewed BM microenvironment is relevant in controlling AML disease.
A high 3-gene IFN signature enrichment at AML diagnosis confers a longer OS to AML standard-risk patients
The intensity of the treatment received by AML patients is calibrated on the relapse risk categorized as low, standard (or “intermediate” in the AIEOP protocol, (29) ) and high based on cytogenetic abnormalities and early response to treatment (30, 31).
Standard-risk patients commonly lack prognostic factors that qualify the low- and high-risks categories and cannot benefit from a personalized therapeutic strategy. An allo-HSCT intensified treatment usually is offered if an HLA-matched related donor is available (29, 32).
Thus, standard-risk patients’ OS is significantly lower than low-risk patients and comparable with the high-risk group that benefits of an intensified treatment, as confirmed also in the analyzed TARGET Cohort (Fig. 2b).
We performed an OS analysis on the TARGET Cohort according to the 3-IFN genes signature and we discovered that a high and medium 3-genes enrichment scores (ES) in the L-TME of patients conferred a significant OS advantage per se (medium vs low p = 0.0008, high vs low p = 0.006) (Fig. 2c). The significance for OS was maintained after a multivariate analysis including clinical risk and age at diagnosis (p = 0.0399) (Fig. 2d).
We analyzed the distribution of the 3-IFN genes signature in the clinical risk groups. The signature was more represented in the low and standard clinical risk groups (Fig. 2e).
To investigate whether the 3-genes score could improve the stratification within the risk groups, we tested the OS in each group. A survival advantage was granted by a high 3-genes ES only in the standard-risk cohort (high vs low p = 0.0299) (Fig. 2f).
To assess whether the 3-genes signature represented the same TH1-enriched L-TME seen in the Discovery Cohort, we performed an enrichment analysis using a previously published immune subpopulations genes set (see Methods section) (33). The analysis confirmed a contextual cytotoxic/NK/Th1 rich-TME present at AML onset in the high 3-genes of standard-risk patients’ group (Fig. 3a).
Specifically, a logistic regression analysis revealed that children categorized as “standard-risk” and falling in the high 3-genes ES tertile have a 76% increased likelihood of a ≥ 6 months CR with respect to children in the low 3-genes ES tertile (Fig. 3b).
Of note, a high 3-genes ES at diagnosis correlated with higher OS, independent of FAB/WHO classifications, and surprisingly was more represented in infants (Fig. 3c-d). A negative correlation between age and 3-genes ES was further confirmed by a correlation analysis (Supplementary Fig. 1b). This observation along with recent literature challenging the dogma of “impaired neonatal immunity”, suggest that infant T cells may have innate-like functions and are prompt to danger signals response (34–36). Indeed, the ability to stratify infant AML with a favorable prognostic factor is very promising, considering that this patient group is usually faces a poor clinical outcome with respect to older pediatric AML patients (37).