Gene list and DepMap data assembly
A curated list of 104 AD-associated genes, consisting of 44 AD risk genes and 60 AD network hub genes, were utilized for this study (see Supplementary Table S1). There were 22 tissue types with data for at least 25 cell lines each from the DepMap 2020 Q2 release which were selected for further analysis (Table 1). We specifically highlighted results from three tissue types most relevant to AD etiology: hematopoietic and lymphoid tissue (HL, given the now established role of immune response and phagocytotic pathway in AD), central nervous system (CNS), and autonomic ganglia (AG)[16–18].
Table 1
Overview of Tissue Types and Number of Unique Cell Lines Used for Cell Line Genetic Dependency Analysis.
Lineage
|
Dependency.
CERES
|
Dependency.
DEMETER2
|
mRNA
|
AUTONOMIC.GANGLIA
|
20
|
9
|
28
|
BILIARY.TRACT
|
28
|
1
|
39
|
BONE
|
28
|
15
|
37
|
BREAST
|
34
|
82
|
57
|
CENTRAL.NERVOUS.SYSTEM
|
60
|
55
|
84
|
ENDOMETRIUM
|
24
|
20
|
33
|
HAEMATOPOIETIC.AND.LYMPHOID.TISSUE
|
82
|
61
|
205
|
KIDNEY
|
21
|
29
|
34
|
LARGE.INTESTINE
|
37
|
44
|
67
|
LIVER
|
23
|
18
|
25
|
LUNG
|
97
|
129
|
192
|
OESOPHAGUS
|
25
|
24
|
32
|
OVARY
|
42
|
38
|
58
|
PANCREAS
|
34
|
33
|
51
|
PLEURA
|
9
|
6
|
17
|
SKIN
|
51
|
46
|
67
|
SOFT.TISSUE
|
38
|
25
|
65
|
STOMACH
|
26
|
25
|
41
|
UPPER.AERODIGESTIVE.TRACT
|
31
|
18
|
43
|
URINARY.TRACT
|
29
|
12
|
36
|
Table 1. 22 tissue types from the Cancer Dependency Map (DepMap) project with data for at least 25 cell lines were utilized for this analysis.
Pan-tissue genetic dependency
To examine the genetic dependency of AD-associated genes across tissues, we identified the proportion of cell lines within each tissue type showing significant negative dependency scores (< -0.5), as lower scores indicate that a gene is required for cell survival and proliferation. From 45 AD risk genes found through GWAS, KAT8 showed predominant negative dependency scores in the CRISPR knockout data (CERES score (< -0.5), with nearly 100% negative dependency scores across all 22 examined tissue types (Fig. 1a). FERMT2 had varying levels of negative CERES scores across tissues with ovary, pleura, and skin cell lines having the highest proportion of negative scores. MEF2C and SPI1 have negative CERES scores localized to the hematopoietic and lymphoid tissue (HL) cell lineage, with 33% and 29%, respectively. We also identified the proportion of cell lines with significant negative genetic dependency scores (DEMETER2 score < -0.5) utilizing data from large-scale RNAi screens. From the list of AD risk genes, KAT8 had a range of significant negative DEMETER2 scores from 17–75% across tissue types, with a peak of 75% in endometrium cell lines (Fig. 1b). ADAMTS4, BIN1, SORL1, and SUZ12P1 had their highest levels of negative DEMETER2 scores in autonomic ganglia (AG) cell lines. SPI1 had its highest proportion of negative DEMETER2 scores in HL cell lines.
Among the 60 AD network hub genes, BUB1, DTL, MED6, PCBP2, RPS18, RPS27, and TIMELESS had significant negative CERES scores of approximately 100% across tissue types (Fig. 2a). UBE2C, ACTG1, CREBBP had varying levels of negative CERES scores ranging from 20–60% across tissue types. GLS and GAB2 showed highest proportion of negative CERES scores in HL (33%) and AG (35%) cell lines, respectively. From AD-associated network hub genes, RPS18 had approximately 100% negative DEMETER2 scores across all tissue types (Fig. 2b). ACTG1, BUB1, DTL, MED6, PCBP2, RPS27, and TIMELESS had varying levels of negative DEMETER2 scores from 10–75% across tissue types. The largely concordant CRISPR/RNAi screen results suggest that knocking out/down of several genes, including KAT8 and FERM2, results in widespread consequences affecting most cells’ survival, whereas other candidates (e.g., MEF2C in HL, GAB in AG) more selectively affect fractions of AD-relevant cell types, thus may serve as better targets.
Expression-driven cellular dependencies of AD risk genes
We reasoned that AD-associated genes may show aberrant expression in disease-driving/affected cells. Thus, the candidate genes would likely serve as better targets if their knockout or knockdown most strongly influenced the cells showing aberrant expression of the targeted genes. We next sought to further filter for genes whose expression is significantly correlated with dependencies of cell lines within these tissue types, i.e., expression-driven dependency. We conducted a systematic Pearson correlation analysis to identify such genes of interest, and identified four genes whose expression was significantly associated with cellular dependencies (FDR < 0.05), including MEF2C in HL cell lines (R = -0.6, FDR = 8.13e-06), SPI1 in HL cell lines (R = -0.6, FDR = 9.38e-06), PSEN2 in HL cell lines (R = -0.4, FDR = 0.0131), and CNTNAP2 in HL cell lines (R = -0.3, FDR = 0.0433) (Fig. 3a). We further conducted the correlation analysis using RNAi knockdown-based DEMETER2 scores to identify similarities and differences to CERES results. We found five genes with significant expression-driven dependency (FDR < 0.05), including SPI1 in HL cell lines (R = -0.5, FDR = 0.00147), MEF2C in HL cell lines (R = -0.5, FDR = 0.00545), HESX1 in HL cell lines (R = -0.5, FDR = 0.00578), CNTNAP2 in HL cell lines (R = 0.5, FDR = 0.0424), KAT8 in CNS cell lines (R = 0.5, FDR = 0.00348) (Fig. 3b).
We next highlighted the cell lines that were most affected by knockout/knockdowns; given the challenge of functionally modeling AD in human systems[20], these lines may provide alternatives that show aberrantly high expression of the selected AD-associated genes. For SPI1 in HL cell lines, these include NOMO1, THP1, MONOMAC1, THP1, and EOL1. For MEF2C in HL cell lines, these were MM1S, KMS20, OCIMY7, KMS28BM, KASUMI2, and L363 (Fig. 3c-d).
Expression-driven cellular dependencies of AD network hub genes
We applied the same expression-driven dependency analysis for the AD network hub genes. Based on the CRISPR screen data, we identified six genes whose expression was significantly associated with cellular dependencies (FDR < 0.05), including GAB2 in HL cell lines (R = -0.5, FDR = 0.000221), ACTG1 in AG cell lines (R = -0.7, FDR = 0.0203), RFX4 in CNS cell lines (R = -0.4, FDR = 0.0203), AQP4 in CNS cell lines (R = -0.4, FDR = 0.0222), FANK1 in AG cell lines (R = 0.7, FDR = 0.0436), MED6 in CNS cell lines (R = 0.4, FDR = 0.043254) (Fig. 4a). Using the RNAi screen data, we identified one gene with significant expression-driven dependency (FDR < 0.05), ABCC11 in HL cell lines (R = -0.5, FDR = 0.00147) (Fig. 4b).
To highlight potential cell lines that can help study the implication of targeting aberrant expressions of these genes, JURLMK1 showed the lowest CERES dependency score and high expression for GAB2 in HL cell lines (Fig. 4c). For ACTG1 in AG cell lines, the cell lines of interest were LS, GIMEN, CHP212, SKNAS, and COGN278. Based on DEMETER dependency scores, for ABCC11 in HL cell lines, A3KAW, A4FUK, OPM2, KASUMI1, and PFEIFFER were highlighted (Fig. 4d). For AQP4 in AG, we noted that the cell lines KPNSI9S, SKNFI, SKNDZ, KPNYN, and SKNBE2 show expression. For GLS in CNS, the cell lines were LN235, U178, KNS60, LN215, and HS683 (see Supplementary Figure S1).