All samples were obtained from the Blood Diseases Hospital, Chinese Academy of Medical Sciences under following acquisition of informed consent from their legal guardians and/or patients authorizing the use of surplus specimens for research purposes. The study protocols were approved by the Institutional Review Board of the Institute of Hematology, Blood Diseases Hospital, PUMC/CAMS (Approval Number: NSFC2022035-EC-2). Patient characteristics are summarized in Supplementary Table S1 . All enrolled patients underwent standardized diagnostic evaluation in accordance with the most recent WHO guidelines. To enhance morphological characterization, samples were further classified using the French-American-British (FAB) classification system. The cohort comprised 12 patients, including 4 AML subtypes (M2, M3, M4, M5) and 2 ALL subtypes: Philadelphia chromosome-negative B-cell ALL (Ph^-) and Philadelphia chromosome-positive B-cell ALL (Ph⁺). Cellular morphology was visualized via MGG staining. We sorted HSPCs cells, B cells, T Cells, monocytes, and granulocytes as normal controls.

The cell processing protocol was implemented according to our previously established methodology (14). Briefly, BM were processed to isolate mononuclear cells by density gradient centrifugation. Before use, cells were washed, viability assessment, and fixation in 1% paraformaldehyde (w/v).

Samples were obtained from healthy donors (aged 20–40 years; no comorbidities) under ethical approval. Mononuclear cells were isolated by density gradient centrifugation. Following centrifugation, lymphocytes and monocytes were localized at the plasma-Ficoll-Paque interface, while granulocytes and erythrocytes were on the bottom. Monocytes were purified using magnetic bead-based separation (Miltenyi Biotec 130-050-201) according to the manufacturer instructions. We sorted HSPCs cells (CD34⁺, APC-CY7 anti-Human CD34, Biolegend 343513), B cells (CD19⁺, APC anti-Human CD19, Biolegend 302212) and T Cells (CD3⁺, FITC anti-Human CD3, BD Biosciences, 555916) by flow cytometry (BD FACSAria III). Granulocytes were isolated through hypotonic lysis using ammonium chloride solution.

The SRS imaging experiments were performed using a multimodal nonlinear optical microscopy system (Model: UltraView MK-II, Zhendian (Suzhou) Medical Technology Co., Ltd., China). Prior to analyzing samples, reference spectra were collected from BSA, DNA, triolein, and palmitic acid samples. All images processed and analyzed using ImageJ software. The details are reported in the Supplementary Material .

OPLS-DA was performed using SIMCA 14.1 software for multivariate statistical analysis. The data matrix consisted of SCR features analyzed using ImageJ software on the X-axis and cell populations on the Y-axis. Principal component analysis was employed to reduce dimensionality while maximizing variance, identifying distinct data clusters. Scores served as indispensable parameters providing biochemical insights, including key factors differentiating cell subtypes.

Group comparisons were conducted using t-tests, ANOVA, or non-parametric alternatives based on data distribution. P-values < 0.05 were considered statistically significant. Box-and-whisker plots were used for data visualization. A detailed breakdown of statistical tests applied is available in Supplementary Material.

This study evaluated the feasibility of rapid histopathological assessment of leukemic specimens using H-SRPI. As illustrated in Figure 1A , Supplementary Figure 1 , the H-SRPI system generated multiplex chemical maps by targeting four key biomolecular fingerprints. Supported by multiple literature sources (22, 27, 35, 36), we used BSA, DNA, triolein, and palmitic acid samples as protein, nucleic acid, unsaturated lipid and saturated lipid standards, respectively. Reference spectra ( Figure 1A ) were derived from purified standards, and hyperspectral data were processed using LASSO regression to estimate pixel-wise biomolecular abundances. The resulting coefficients were spatially mapped to generate single-cell resolution chemical images. Acute leukemia was selected as the disease model. BM were collected with six distinct subtypes: AML-M2, M3, M4, M5 and B-ALL Ph^-, Ph⁺ ( Figure 1B ), all of which are believed to originate from leukemic stem cells (LSCs). The healthy controls were containing: HSPCs, granulocytes, monocytes, B cells, and T cells ( Figure 1C ). As shown in Figure 1D , the workflow of H-SRPI is described.

Metabolic reprogramming and epigenetic remodeling are hallmarks of leukemogenesis and AML disease progression (37, 38). These alterations are often genotype-specific and accompanied by epigenetic and functional changes that promote oncogenic pathway activation (39, 40). To evaluate the utility of H-SRPI in capturing these metabolic shifts, cross-comparative analysis between leukemia blasts and their normal counterparts was to conduct delineate malignancy-associated biochemical signatures. The purity of the three populations exceeded 95% by flow cytometry, as reported in Supplementary Figure 2 . We first compared granulocyte-matched cells. Given the granulocytic dominance in AML M2, M3, and M4, these subtypes were analyzed against normal granulocytes. H-SRPI imaging ( Figure 2A ) revealed leukemic cells exhibited significantly elevated protein and lipid signals, suggesting enhanced biosynthetic activity, MGG staining ( Figure 2B ) confirmed the presence of abnormal myeloid blasts. And quantitative analysis ( Figure 2C ) showed significantly higher levels of proteins and lipids in AML-M2, M3, M4 cells. Notably, the increase in unsaturated lipids exceeded that of saturated lipids, indicating a preferential shift toward unsaturated lipid biosynthesis. In contrast, nucleic acid content was significantly decreased, consistent with reduced DNA abundance during leukemogenesis.

We next compared M5 cells with monocytes, given their shared monocytic lineage. H-SRPI imaging ( Figure 2D ) revealed consistently elevated levels of proteins, nucleic acids, and lipids in M5 cells. These differences were further supported by MGG staining of both monoblasts and monocytes ( Figure 2E ). Quantitative analysis ( Figure 2F ) demonstrated a substantial increase in all four biomolecular components in AML M5 cells, reflecting robust metabolic reprogramming characterized by enhanced protein synthesis and lipid accumulation.

To systematically compare biomolecular profiles between ALL and normal B cells, we compared ALL B cells with healthy donors (HD) B cells. H-SRPI revealed distinct intracellular distributions: both Ph⁺ and Ph^- B-ALL cells exhibited elevated proteins, nucleic acids, and unsaturated lipids compared to HD B cells, with the Ph⁺ group showing the highest enrichment of unsaturated lipids ( Figures 3A, C ). These molecular distinctions were corroborated by MGG staining, which highlighted clear morphological distinctions across groups ( Figure 3B ). However, in comparison to AML cells, ALL cells demonstrated weaker Raman signals and markedly reduced lipid content, suggesting that protein and nucleic acid metabolism dominate in ALL cell physiology.

We quantitatively characterized 174 individual cells obtained from 12 leukemia patients. 5 features of cellular area and chemical composition (protein, nucleic acid, saturated and unsaturated lipid content) were extracted by ImageJ software. Nevertheless, the differences of composition and morphology features between leukemia subtypes were not very evident, which was likely due to the existence of heterogeneous populations in each specimen. Therefore, we proposed to analyze the features by OPLS-DA. The resulting scatter plots, reporting the scores of the first two canonical variables, are shown in Figures 4A–C . As shown in Figures 4A and Supplementary Figure 3A , representing scores of “AMLs+ALLs”, a certain degree of separation was obtained between AML and ALL subtypes (AML: 90% precision and 90% sensitivity; ALL: 81% precision and 80% sensitivity).

When only AML subtypes are considered, OPLS-DA can separate AML M2 (100% precision and 77% sensitivity), M3 (85% precision and 89% sensitivity), M4 (92% precision and 63% sensitivity), and M5 (72% precision and 96% sensitivity) cells ( Figures 4B ; Supplementary Figure 3B ). The confusion matrix resulting shows very good accuracy for the classification of M5 and other subtypes. M2, M3, and M4 cluster together due to their similarity as granulocytes. When only ALL subtypes are analyzed, a very good separation can be seen between Ph⁻ and Ph⁺ (100% precision and 100% sensitivity) ( Figures 4C ; Supplementary Figure 3C ). The total accuracy is 88.21%. These results suggest that our H-SRPI imaging method is capable to detect leukemia cells with high sensitivity and specificity.

To further characterize these metabolic changes, we identified 281 significantly upregulated genes in AML cells ( Figure 5B ). Venn diagram analysis revealed overlapping genes involved in lipid metabolism, lipid biosynthesis, and lipid modification ( Figure 5C ), suggesting their involvement in AML-specific lipid reprogramming. 20 upregulated genes were identified within the intersection of lipid metabolism-related pathways in AML cells ( Figure 5D ). Among them, ACSS1, HACD4, LIPA, SLC44A1, and CD36 were particularly notable due to their central roles in lipid synthesis, elongation, degradation, and transport. ACSS1 and HACD4 mediate acetate utilization and very long-chain fatty acid elongation, respectively, reflecting enhanced anabolic lipid metabolism in AML (41, 42). CD36, a key fatty acid transporter (43), was significantly upregulated and correlated with the elevated unsaturated lipid levels observed by H-SRPI imaging. These findings underscore the role of lipid metabolic reprogramming in AML, as the 20 upregulated genes enhanced metabolic flexibility and leukemic progression, and may facilitate drug resistance and immune evasion.

Next, we analyzed the metabolic characteristics of ALL cells. It showed enrichment in pathways such as the carboxylic acid metabolic process, metapathway biotransformation Phase I and II, whereas normal B cells were enriched in energy-generating and biosynthetic pathways ( Figure 5E ). Notably, lipid metabolism–related pathways were absent from the top enriched terms in both groups. Consistently, GSEA results indicated lipid-associated pathways remained non-enriched ( Supplementary Figure 4E ), while significant upregulation of cyclic nucleotide metabolic process and CGMP metabolic process ( Supplementary Figure 4F ). These findings verified that ALL cells rely more heavily on protein and nucleic acid metabolism than on lipid metabolism.

Accumulating evidence indicates that AML cells undergo extensive lipid metabolic reprogramming to sustain malignant proliferation and survival (44). To investigate subtype-specific lipid metabolic adaptations, we performed GSVA across four AML subtypes. M5 showed the highest scores in overall lipid metabolism, while M3 displayed the strongest enrichment in lipid biosynthetic and modification pathways ( Figures 6A-C ), indicating distinct metabolic strategies. To elucidate underlying mechanisms, we analyzed the expression of genes involved in fatty acid transport, activation, synthesis, desaturation, and elongation. M3 cells showed high expression of SLC27A2,SLC27A3, fatty acid transport protein (FATP) family that mediate FA uptake for β-oxidation (45), along with elevated ELOVL3 (46) ( Figure 6D ). In contrast, M5 cells upregulated lipogenic genes ( Figure 6E ), including FADS1, SREBF1 (lipogenesis regulator driving ACLY, FASN, ACACA, and SCD (47)). SCD facilitates FA desaturation to support membrane dynamics (48). ELOVL5/6 extend PUFA and SFA chains to modulate lipid composition (46). Taking together, these results demonstrate distinct lipid metabolic programs in AML. The M3 favors lipid desaturation and unsaturation, whereas the M5 upregulates saturated lipid synthesis and elongation, forming a metabolic axis linked to leukemic progression and therapeutic response. To investigate lipid metabolism during AML progression process, we analyzed HSPCs and aberrant promyelocytes from M3-AML. M3 exhibited significantly higher levels of unsaturated lipids content and metabolism features ( Supplementary Figure 5 ).