Supplementary Methods
This section provides technical details on the development of the 3 distinct candidate mRNA-based predictors of response to dasatinib therapy. We examine the reproducibility and concordance of prediction results in gene expression data from identical cell lines grown in different laboratories and replicate needle biopsy specimens from the same human tumors.
General Descriptions and Data Preprocessing
The prediction values for the three dasatinib predictors will be calculated from FNA (fine-needle aspiration) biopsies of 133 breast cancers to establish a reference distribution of these prediction indices in human cancers.
The microarray data for the 133 samples were generated with Affymetrix Human Genome U133A 2.0 array from reference 1.
Quality Assessment:The quality of the arrays was assessed using package “simpleaffy” in R version 2.4.0. The results are provided in table 1 below. Please note:
- Scaling factors should be within 3-fold of each other.
- Percent.present indicates percentage of probe sets declared "present" by the algorithm. The standard is 30%~60%.
- The safe threshold recommended by Affymetrix for the 3'/5' ratios of the two housekeeing genes is 3.
- Highlighted samples are samples with insufficient QC.
Data Preprocessing:All arrays were then normalized using MAS5 with scale factor set to 1500 in the statistical software R.
Dasatinib Predictors
Dasatinib 19-genes target index
The dasatinib19-genes target index is defined as the weighted sum of theexpression of 19 distinct kinases that dasatinib binds to with highaffinity (KD50 < 1 nM) in vitro. When multiple probe sets correspond to one gene, theprobe set with the highest median intensity and greatest interquartile range was selected as the representative probe set. In order tocalculate the dasatinib target index, the normalized, log 2 transformed gene expression values for each of the 19 targets were weighed bythe corresponding KD50 values (binding constants) that are publiclyavailable form in vitro experiments (2).
We defined the individuals with the highest 25% of target index values as "potentially drug sensitive" and therefore eligible fortherapy. This 25% cut off is arbitrary but it is based on our hypothesisthat the higher the target index value the greater the sensitivity to
dasatinib is. For each new case, the 19-gene target index will be calculated bysumming the normalized gene expression values for all of the 19 probe sets, eachdivided by the corresponding log2 transformed KD50 value. If the resulting index is > 3070.56, the patient is considered to have high targetindex and will be eligible for therapy with dasatinib. The cutoff value of 3070.56 was derived from the reference population of 133 cases anddefines the lower limit of the target indices that belong to the top 25%of the overall index distribution (figure 1).
Supplemental Table 1: Probe sets used in dasatinib 19-gene dasatinib target index
Number / Probe.set / Gene.symbol / KD50.value1 / 202123_s_at / ABL1 / 0.5
2 / 206411_s_at / ABL2 / 0.2
3 / 210934_at / BLK / 0.2
4 / 203499_at / EPHA2 / 0.8
5 / 206071_s_at / EPHA3 / 0.09
6 / 215664_s_at / EPHA5 / 0.2
7 / 210753_s_at / EPHB1 / 0.4
8 / 202894_at / EPHB4 / 0.3
9 / 208438_s_at / FGR / 0.5
10 / 207178_s_at / FRK / 0.3
11 / 210105_s_at / FYN / 0.7
12 / 208018_s_at / HCK / 0.3
13 / 205051_s_at / KIT / 0.6
14 / 204891_s_at / LCK / 0.2
15 / 202626_s_at / LYN / 0.6
16 / 203131_at / PDGFRA / 0.4
17 / 202273_at / PDGFRB / 0.6
18 / 213324_at / SRC / 0.2
19 / 202933_s_at / YES1 / 0.3
Supplemental Figure 1: Overall index distribution of Dasatinib 19-gene target index
Initial SRCPathway score
The Src Pathway score was calculated based on published data (3). Investigators used human primary mammary epithelial cell (HMEC)cultures to express oncogenic pathways including Src. They thencompared the HMEC over-expressing the pathway of interest to acontrol HMEC culture overexpressing green fluorescent protein (GFP)to identify genes that are associated with overexpression of the givenoncogenic pathway. The original signature consisted of 73 probe setsfrom the Affymetrix HG-U133Plus2 array - only 46 of the 73 probesets are present in the Affymetrix HG-133A arrays therefore the gene signature for our Src index is based on these 46 probe sets.
We down-loaded the original dataset and identified the 10 HMECsoverexpressing Src and those who expressed GFP (control). Usingthe 46 genes that represent the Src pathway activity on our arrays, wecalculated two centroids that consisted of the median expressionlevels of these genes in the Src-expressing HMECs (positivesignature) and in the GFP-expressing HMECs (negative signature).For internal validation and consistency check, we examined theoriginal published data and confirmed that the Src index was high inthe Src-expressing HMECs and low in the GFP-expressing HMECs.The remaining HMEC cultures that over expressed other pathwaysshowed different levels of induction of the Src pathway, but none wasas high as the Src-expressing HMECs. This suggests the centroidscapture the varying degrees of expression of the Src pathways in theengineered cells.
Next, we compared human breast cancer samples to the Srccentroids calculated from the cell line data. As a similarity metric weused Hoeffding's D statistic. This is a non-parametric measure ofassociation like Spearman's rank but it can capture non-linearassociations unlike the Spearman's test. We defined the Src index as the difference between the two statistical metrics: "(D+) -(D-)", where D+ is Hoeffding's D relative to the positive centroid (i.eSrc overexpressing HMEC), and D- is relative to the negative (GFPoverexpressing HMEC). We calculated the Src index for 133breast cancer samples. The index values ranged from -6.56 to 8.36. We will consider a patient "positive" for Src pathway activity, andeligible for treatment, if the Src index is within the top 25 percentile ofthis reference distribution, this corresponds to Src index > 1.36.
Supplemental Table 2: Probe sets used in Src Pathway index
Number / Probe.set / Gene.symbol / Number / Probe.set / Gene.symbol1 / 31874_at / GAS2L1 / 24 / 212044_s_at / RPL27A
2 / 37028_at / PPP1R15A / 25 / 212435_at / ---
3 / 200908_s_at / RPLP2 / 26 / 212492_s_at / KIAA0876
4 / 201128_s_at / ACLY / 27 / 212560_at / SORL1
5 / 201447_at / TIA1 / 28 / 212928_at / TSPYL4
6 / 201737_s_at / TEB4 / 29 / 213056_at / GRSP1
7 / 201879_at / ARIH1 / 30 / 213069_at / KIAA1237
8 / 202245_at / LSS / 31 / 213164_at / MRPS6
9 / 202506_at / SSFA2 / 32 / 213243_at / COH1
10 / 202569_s_at / MARK3 / 33 / 213262_at / SACS
11 / 202643_s_at / TNFAIP3 / 34 / 213279_at / DHRS1
12 / 202648_at / RPS19 / 35 / 213350_at / RPS11
13 / 203301_s_at / DMTF1 / 36 / 213352_at / KIAA0779
14 / 204404_at / SLC12A2 / 37 / 213485_s_at / ABCC10
15 / 204614_at / SERPINB2 / 38 / 213865_at / ESDN
16 / 204847_at / ZNF-U69274 / 39 / 215867_x_at / AP1G1
17 / 206011_at / CASP1 / 40 / 218397_at / FANCL
18 / 206414_s_at / DDEF2 / 41 / 218796_at / C20orf42
19 / 206591_at / RAG1 / 42 / 219181_at / LIPG
20 / 206976_s_at / HSPH1 / 43 / 219571_s_at / ZNF325
21 / 209537_at / EXTL2 / 44 / 220687_at / TRRAP
22 / 209773_s_at / RRM2 / 45 / 221284_s_at / SRC
23 / 211756_at / PTHLH / 46 / 221900_at / COL8A2
BMS cell line derived sensitivity predictor
Investigators at Bristol-Myers Squibb (BMS) determined the sensitivity of 23breast cancer cell lines to dasatinib in vitro (3). They used the meanlog10IC50 values across all 23 cell lines (cut off 0.6 mM) to assigneither dasatinib sensitive (IC50 bellow the mean) or resistant (IC50above the mean) category to each cell line.
To identify differentially expressed genes between dasatinib sensitive and resistant cases, two-sample unequal-variance t tests were performed on the gene expression profiling of the 23 breast cancer cell lines and then genes were rank-ordered by p values. The top 142 probe sets that showed false discovery rates of 10% (FDR<0.1) were selected as the informative gene set. A multi-gene response predictor was developed using the top-ranked 142 probe sets and weighted votingalgorithm to assign response category to new case. Initially, [µ1(g), σ1(g)] and [µ2(g), σ2(g)] were defined as the medians and standard deviations of the log expression levels of gene g for the cell line samples in class 1 (sensitive class) and class 2 (resistant class). The weight for gene g wasdefined as
ag = [µ1(g)-µ2(g)]/[σ1(g)+σ2(g)]; the decision boundary for g was defined as bg = [µ1(g)+µ2(g)]/2. Whenever xgdenoted the normalized log expression value of gene g in a new sample, then the vote for gene g was (xg –bg) and the weighted vote of gene g would bevg=ag(xg –bg), with a positive value indicating a vote for class 1 and negative value indicating a vote for class 2. The votes of the selected genes were summed to obtain total votes. For class 1, V1 = where I1=1 if vg>0 otherwise I1=0; for class 2, V2 = where I2=1 if vg<0 otherwise I2=0. The new sample would be assigned to the class receiving the higher number of absolute value of votes.The cutoff for dasatinib sensitive is 0.
This cell line basedpredictor was tested on 12 independent breast cancer cell lines withknown dasatinib sensitivity. It correctly predicted response categoryfor 3 of the 4 sensitive cell lines (i.e sensitivity 75%) and all 8 resistant cell lines (specificity 100%).
Supplemental Table 3: Probe sets used in the Cell Line Derived Sensitivity predictor
Number / Probe.set / Gene.symbol / Number / Probe.set / Gene.symbol1 / 220914_at / --- / 72 / 214338_at / DNAJB12
2 / 216268_s_at / JAG1 / 73 / 201251_at / PKM2
3 / 209099_x_at / JAG1 / 74 / 218985_at / SLC2A8
4 / 209193_at / PIM1 / 75 / 208790_s_at / PTRF
5 / 209098_s_at / JAG1 / 76 / 219692_at / KREMEN2
6 / 220352_x_at / MGC4278 / 77 / 200889_s_at / SSR1
7 / 208966_x_at / IFI16 / 78 / 215867_x_at / CA12
8 / 206332_s_at / IFI16 / 79 / 204753_s_at / HLF
9 / 209160_at / AKR1C3 / 80 / 218214_at / FLJ11773
10 / 204343_at / ABCA3 / 81 / 211526_s_at / RTEL1
11 / 205535_s_at / PCDH7 / 82 / 209019_s_at / PINK1
12 / 217943_s_at / RPRC1 / 83 / 217655_at / FXYD5
13 / 208965_s_at / IFI16 / 84 / 200697_at / HK1
14 / 209154_at / TAX1BP3 / 85 / 204508_s_at / CA12
15 / 218084_x_at / FXYD5 / 86 / 203928_x_at / MAPT
16 / 214581_x_at / TNFRSF21 / 87 / 200771_at / LAMC1
17 / 213506_at / F2RL1 / 88 / 214446_at / ELL2
18 / 211749_s_at / VAMP3 / 89 / 214331_at / TSFM
19 / 218717_s_at / LEPREL1 / 90 / 215219_at / C21orf5
20 / 209589_s_at / EPHB2 / 91 / 206571_s_at / MAP4K4
21 / 213139_at / SNAI2 / 92 / 212097_at / CAV1
22 / 215196_at / --- / 93 / 220949_s_at / MGC5242
23 / 204279_at / PSMB9 / 94 / 203247_s_at / ZNF24
24 / 219871_at / FLJ13197 / 95 / 202388_at / RGS2
25 / 208789_at / PTRF / 96 / 201813_s_at / TBC1D5
26 / 208908_s_at / CAST / 97 / 209625_at / PIGH
27 / 203306_s_at / SLC35A1 / 98 / 208944_at / TGFBR2
28 / 211453_s_at / AKT2 / 99 / 210735_s_at / CA12
29 / 201263_at / TARS / 100 / 66053_at / HNRPUL2
30 / 221087_s_at / APOL3 / 101 / 215173_at / LRRC50
31 / 216854_at / GDF11 / 102 / 210547_x_at / ICA1
32 / 201336_at / VAMP3 / 103 / 210859_x_at / CLN3
33 / 207467_x_at / CAST / 104 / 205062_x_at / ARID4A
34 / 217373_x_at / MDM2 / 105 / 204567_s_at / ABCG1
35 / 203324_s_at / CAV2 / 106 / 211241_at / ANXA2P3
36 / 200770_s_at / LAMC1 / 107 / 209618_at / CTNND2
37 / 203499_at / EPHA2 / 108 / 207011_s_at / PTK7
38 / 210024_s_at / UBE2E3 / 109 / 209757_s_at / MYCN
39 / 209040_s_at / PSMB8 / 110 / 215505_s_at / STRN3
40 / 202062_s_at / SEL1L / 111 / 204747_at / IFIT3
41 / 218856_at / TNFRSF21 / 112 / 201155_s_at / MFN2
42 / 205729_at / OSMR / 113 / 64486_at / CORO1B
43 / 218718_at / PDGFC / 114 / 204820_s_at / BTN3A3
44 / 220468_at / ARF7 / 115 / 201391_at / TRAP1
45 / 214046_at / --- / 116 / 219772_s_at / SMPX
46 / 215989_at / CBX2 / 117 / 213238_at / ATP10D
47 / 205398_s_at / SMAD3 / 118 / 214771_x_at / M-RIP
48 / 203234_at / UPP1 / 119 / 204646_at / DPYD
49 / 200600_at / MSN / 120 / 219456_s_at / RIN3
50 / 213913_s_at / KIAA0984 / 121 / 221773_at / ELK3
51 / 208835_s_at / CROP / 122 / 202077_at / NDUFAB1
52 / 221831_at / LUZP1 / 123 / 212594_at / PDCD4
53 / 203802_x_at / NSUN5 / 124 / 38241_at / BTN3A3
54 / 208102_s_at / PSD / 125 / 206734_at / JRKL
55 / 218307_at / RSAD1 / 126 / 216215_s_at / RBM9
56 / 219751_at / FLJ21148 / 127 / 220457_at / FLJ10211
57 / 203954_x_at / CLDN3 / 128 / 203895_at / PLCB4
58 / 212294_at / GNG12 / 129 / 201590_x_at / ANXA2
59 / 221874_at / KIAA1324 / 130 / 217501_at / WDR39
60 / 203478_at / NDUFC1 / 131 / 203910_at / ARHGAP29
61 / 208816_x_at / ANXA2P2 / 132 / 205009_at / TFF1
62 / 201661_s_at / ACSL3 / 133 / 209835_x_at / CD44
63 / 215464_s_at / TAX1BP3 / 134 / 207021_at / ZPBP
64 / 204238_s_at / C6orf108 / 135 / 203623_at / PLXNA3
65 / 208782_at / FSTL1 / 136 / 201740_at / NDUFS3
66 / 214030_at / DKFZp667G2110 / 137 / 205634_x_at / ZDHHC24
67 / 219691_at / SAMD9 / 138 / 212754_s_at / KIAA1040
68 / 212253_x_at / DST / 139 / 213398_s_at / C14orf124
69 / 212534_at / ZNF24 / 140 / 209018_s_at / PINK1
70 / 220329_s_at / C6orf96 / 141 / 204876_at / ZNF646
71 / 206329_at / EXTL1 / 142 / 211888_x_at / CASP10
Overlap in prediction results
For the reference population of 133 breast cancer patients, we examined the correlation among the dasatinib 19-genes target index, the Src Pathway index from protocol version 3and the BMS cell line based predictor. Three samples (M154, M234 and M310) were predicted to be dasatinib sensitive by all of them. Table 4 and figure 2 show the agreements between predictors in more detail. The prediction results for each individual are listed in table 6.
Supplemental Table 4: Agreement between predictors
Dasatinib 19-gene target indextop25 / others
BMS cell line based predictor / Sensitive / 8 / 5
Resistance / 25 / 95
SRC Pathway index
top25 / others
BMS cell line based predictor / Sensitive / 4 / 9
Resistance / 29 / 91
SRC Pathway index
top25 / others
Dasatinib 19-gene target index / top25 / 10 / 23
others / 23 / 77
Supplemental Figure 2: Agreements between 3 predictors
Reproducibility of prediction results
We have replicate gene expression measurements from 39 individual fine needle aspiration (FNA) samples and 6 core biopsy (CBX) samples. The total RNA from these samples was split and profiled twice either in different labs or several months apart in the same laboratory. These represent technical replicates. We also have 37matched pairs of FNA and CBX (29 from a cohort clinical trail, 8 from a pilot study) These represent biopsy replicates.
The microarray data for all these samples were also generated with Affymetrix Human Genome U133A 2.0 array. The quality of the arrays was assessed using package “simpleaffy” in R version 2.4.0., and then they were normalized using MAS5 with scale factor set to 1500 in the statistical software R in the exactly same way as we did for the 133 patients. Table 6 and 7 are the QC results.
Supplemental Table5: QC metrics for the 39 FNA samples used for replicate hybridization (technical replicates)
Sample / Pair Number / Passed QC / scale_factor / percent present / AFFX-HSAC07 3'/5' / AFFX-HUMGAPDH 3'/5'M153 / 1 / Y / 1.759 / 40.219 / 1.309 / 0.258
M153R / 1 / Y / 0.957 / 23.973 / 0.571 / 0.364
M230 / 2 / Y / 2.979 / 43.603 / 0.356 / 0.214
M230R / 2 / Y / 2.088 / 41.897 / 0.780 / 0.297
M231 / 3 / N / 7.365 / 31.333 / 0.767 / 0.982
M231R / 3 / Y / 3.506 / 37.634 / 1.718 / 0.864
M233 / 4 / Y / 3.498 / 39.829 / 2.080 / 1.178
M233R / 4 / N / 5.161 / 35.516 / 3.842 / 2.262
M235 / 5 / Y / 1.127 / 49.948 / 0.415 / -0.063
M235R / 5 / Y / 0.814 / 51.739 / 0.415 / -0.084
M236 / 6 / Y / 0.964 / 52.735 / 0.904 / 0.409
M236R / 6 / Y / 1.311 / 47.960 / 0.707 / 0.555
M237 / 7 / Y / 1.175 / 49.737 / 2.481 / 1.148
M237R / 7 / Y / 1.693 / 46.865 / 2.407 / 1.374
M238 / 8 / Y / 2.863 / 45.043 / -0.177 / 0.699
M238R / 8 / Y / 1.370 / 49.320 / 0.708 / 0.597
M239 / 9 / Y / 1.977 / 44.352 / 0.170 / 0.429
M239R / 9 / Y / 1.781 / 43.607 / 0.636 / 0.304
M245 / 10 / Y / 3.488 / 42.867 / 0.081 / 0.567
M245R / 10 / Y / 1.928 / 44.653 / 0.849 / 0.211
M246 / 11 / Y / 1.125 / 49.827 / 0.767 / -0.116
M246R / 11 / Y / 1.768 / 43.437 / 0.697 / 0.025
M247 / 12 / N / 10.053 / 26.540 / 6.022 / 5.486
M247R / 12 / N / 13.061 / 21.819 / 6.045 / 5.600
M256 / 13 / Y / 1.245 / 50.411 / 1.257 / 0.736
M256R / 13 / Y / 1.222 / 50.801 / 1.369 / 0.696
M257 / 14 / N / 5.214 / 32.455 / -0.489 / 0.326
M257R / 14 / Y / 2.635 / 39.429 / 0.439 / 0.833
M258 / 15 / Y / 0.973 / 51.681 / 1.031 / 0.475
M258R / 15 / Y / 1.233 / 49.069 / 0.992 / 0.717
M264 / 16 / Y / 0.600 / 55.042 / 0.216 / -0.208
M264R / 16 / Y / 0.778 / 52.286 / 0.212 / -0.199
M266 / 17 / Y / 3.970 / 41.637 / 0.076 / 0.627
M266R / 17 / Y / 1.412 / 48.270 / 0.785 / 0.482
M270 / 18 / Y / 0.755 / 51.129 / 0.389 / -0.267
M270R / 18 / Y / 1.306 / 46.367 / 0.064 / 0.128
M280 / 19 / Y / 3.155 / 42.104 / -0.168 / 0.498
M280R / 19 / Y / 1.312 / 47.606 / 0.632 / 0.143
M286 / 20 / Y / 2.259 / 45.371 / 0.502 / -0.389
M286R / 20 / Y / 3.327 / 40.888 / 3.191 / 1.414
M302 / 21 / Y / 2.951 / 44.119 / 1.140 / 0.339
M302R / 21 / Y / 3.899 / 38.976 / 2.740 / 1.327
M316 / 22 / Y / 1.969 / 32.900 / 2.571 / 1.921
M316R / 22 / Y / 1.763 / 41.965 / 2.130 / 1.201
M322 / 23 / Y / 0.983 / 43.033 / 0.662 / 0.260
M322R / 23 / Y / 1.024 / 47.193 / 0.242 / 0.037
M323 / 24 / Y / 2.237 / 37.845 / 1.122 / 0.817
M323R / 24 / Y / 2.237 / 37.845 / 1.122 / 0.817
M330 / 25 / Y / 1.311 / 41.606 / 0.927 / 0.609
M330R / 25 / Y / 2.495 / 41.978 / 0.619 / 0.328
M331 / 26 / Y / 0.990 / 39.124 / 0.237 / 0.360
M331R / 26 / Y / 1.540 / 42.544 / -0.141 / 0.055
M332 / 27 / Y / 1.357 / 45.317 / 1.029 / 0.282
M332R / 27 / Y / 1.045 / 49.607 / 0.997 / 0.087
M333 / 28 / Y / 1.116 / 39.173 / 0.889 / 0.546
M333R / 28 / Y / 1.674 / 45.452 / 0.367 / 0.284
M337 / 29 / Y / 0.960 / 43.360 / 0.513 / 0.277
M337R / 29 / Y / 1.087 / 48.037 / 0.068 / 0.078
M371 / 30 / Y / 1.387 / 38.639 / 1.272 / 0.415
M371R / 30 / Y / 1.966 / 41.731 / 0.394 / -0.070
M375 / 31 / Y / 1.187 / 44.743 / 1.213 / 0.769
M375R / 31 / Y / 1.378 / 48.768 / 1.177 / 0.394
M386 / 32 / Y / 1.282 / 44.325 / 1.099 / 0.619
M386R / 32 / Y / 1.297 / 48.862 / 0.773 / 0.313
M387 / 33 / Y / 0.843 / 44.949 / 0.930 / 0.576
M387R / 33 / Y / 1.202 / 49.764 / 0.610 / 0.390
M402 / 34 / Y / 0.868 / 42.620 / 0.361 / 0.158
M402R / 34 / Y / 1.545 / 45.465 / -0.056 / -0.135
M406 / 35 / N / 2.156 / 37.078 / 3.353 / 2.509
M406R / 35 / Y / 3.209 / 40.870 / 2.775 / 1.814
M419 / 36 / Y / 0.868 / 42.633 / 0.047 / 0.103
M419R / 36 / Y / 1.085 / 44.195 / -0.081 / -0.076
ML20 / 37 / N / 8.897 / 19.225 / 5.661 / 6.790
ML20R / 37 / Y / 5.410 / 29.722 / 7.142 / 6.309
PERU06R / 38 / Y / 1.429 / 45.891 / 1.131 / 1.397
PERU13.15R / 38 / Y / 1.685 / 44.716 / 0.622 / 0.533
PEU06 / 39 / Y / 1.454 / 50.518 / 1.830 / 0.987
PEU13.15 / 39 / Y / 1.393 / 52.044 / 0.980 / 0.272
Supplemental Table 6: QC metrics for the 37 paired FNA/CBX samples and 6 replicate CBX samples
Sample / Pair Number / Type / Passed QC / scale factor / percent present / AFFX-HSAC07, 3'/5' / AFFX-HUMGAPDH, 3'/5'M123 / 1 / FNA / Y / 0.988 / 47.242 / 0.930 / 0.186
M123CBX / 1 / CBX / Y / 1.608 / 39.290 / 0.802 / 0.161
M135 / 2 / FNA / Y / 2.862 / 37.800 / 0.674 / -0.300
M135CBX / 2 / CBX / Y / 1.513 / 40.039 / 0.980 / 0.178
M136 / 3 / FNA / Y / 0.648 / 49.553 / 0.458 / 0.266
M136CBX / 3 / CBX / Y / 1.025 / 49.953 / 0.212 / -0.005
M145 / 4 / FNA / Y / 1.398 / 54.135 / 0.573 / 0.043
M145CBX / 4 / CBX / Y / 0.886 / 49.311 / 0.633 / 0.179
M150 / 5 / FNA / Y / 0.616 / 49.396 / 0.269 / 0.809
M150CBX / 5 / CBX / Y / 0.661 / 47.700 / 0.229 / 0.315
M151 / 6 / FNA / Y / 1.410 / 32.962 / 1.913 / 0.215
M151CBX / 6 / CBX / Y / 0.912 / 45.739 / 1.381 / 0.217
M157 / 7 / FNA / Y / 0.992 / 49.060 / 1.003 / 0.557
M157CBX / 7 / CBX / Y / 0.879 / 47.848 / 0.353 / 0.319
M177 / 8 / FNA / Y / 0.858 / 42.844 / 1.206 / 0.673
M177CBX / 8 / CBX / Y / 0.478 / 49.518 / 0.533 / 0.151
M182 / 9 / FNA / Y / 1.778 / 41.489 / 0.798 / 0.878
M182CBX / 9 / CBX / Y / 0.797 / 46.583 / 0.634 / 0.081
M189 / 10 / FNA / Y / 0.965 / 46.951 / 1.691 / 0.429
M189CBX / 10 / CBX / Y / 0.504 / 50.653 / 0.401 / -0.139
M239 / 11 / FNA / Y / 1.977 / 44.352 / 0.170 / 1.287
M239CBX / 11 / CBX / Y / 1.240 / 43.441 / 1.232 / 0.226
M300 / 12 / FNA / Y / 1.360 / 33.649 / 2.488 / 1.921
M300CBX / 12 / CBX / Y / 0.942 / 46.421 / 1.569 / 0.287
M316 / 13 / FNA / Y / 1.969 / 32.900 / 2.571 / 0.126
M316CBX / 13 / CBX / Y / 1.072 / 44.805 / 0.684 / 0.357
M325 / 14 / FNA / Y / 1.458 / 42.925 / 1.125 / 1.017
M325CBX / 14 / CBX / Y / 1.215 / 43.289 / 0.508 / 0.263
M339 / 15 / FNA / Y / 0.580 / 49.639 / 1.832 / 1.065
M339CBX / 15 / CBX / Y / 0.519 / 52.897 / 0.275 / -0.093
M352 / 16 / FNA / Y / 1.456 / 40.183 / 2.288 / 1.041
M352CBX / 16 / CBX / Y / 2.289 / 27.276 / 1.833 / 0.337
M375 / 17 / FNA / Y / 0.924 / 48.943 / 3.116 / 1.138
M375CBX / 17 / CBX / Y / 0.747 / 44.761 / 0.538 / -0.154
M376 / 18 / FNA / Y / 3.272 / 24.139 / 3.169 / 0.305
M376CBX / 18 / CBX / Y / 0.676 / 50.599 / 1.213 / 0.225
M582 / 19 / FNA / Y / 1.447 / 40.897 / 0.997 / 0.279
M582CBX / 19 / CBX / Y / 1.509 / 40.336 / 1.809 / 0.349
M641 / 20 / FNA / Y / 0.637 / 50.272 / 1.358 / 0.306
M641CBX / 20 / CBX / Y / 0.724 / 52.246 / 1.293 / 0.163
M722 / 21 / FNA / Y / 1.314 / 40.511 / 0.724 / 0.368
M722CBX / 21 / CBX / Y / 1.693 / 37.841 / 2.405 / 0.766
M771 / 22 / FNA / Y / 0.978 / 34.627 / 0.482 / 0.517
M771CBX / 22 / CBX / Y / 1.330 / 42.413 / 1.003 / 0.035
M800 / 23 / FNA / Y / 1.256 / 38.846 / 2.255 / 0.968
M800CBX / 23 / CBX / Y / 0.862 / 46.067 / 1.015 / 0.195
M801 / 24 / FNA / Y / 1.633 / 34.129 / 2.238 / 0.955
M801CBX / 24 / CBX / Y / 0.688 / 49.702 / 0.912 / 0.009
M807 / 25 / FNA / Y / 1.117 / 42.194 / 2.311 / 0.688
M807CBX / 25 / CBX / Y / 0.721 / 48.589 / 1.402 / 0.262
M813 / 26 / FNA / Y / 1.292 / 39.173 / 1.812 / 0.972
M813CBX / 26 / CBX / Y / 0.833 / 46.336 / 1.419 / 0.202
M815 / 27 / FNA / Y / 1.692 / 38.191 / 2.876 / 1.157
M815CBX / 27 / CBX / Y / 1.340 / 45.151 / 1.491 / 0.172
M834 / 28 / FNA / Y / 1.193 / 41.314 / 1.968 / 0.892
M834CBX / 28 / CBX / Y / 1.127 / 38.725 / 1.862 / 0.494
M855 / 29 / FNA / Y / 1.569 / 35.875 / 1.503 / 0.926
M855CBX / 29 / CBX / Y / 1.200 / 46.933 / 0.608 / -0.007
C.ANON.A / 30 / FNA / Y / 0.418 / 49.069 / 0.462 / 0.284
C.ANON.A.CBX / 30 / CBX / Y / 0.872 / 45.748 / 0.472 / 0.425
C.ANON.B / 31 / FNA / Y / 0.703 / 43.513 / 0.056 / 0.101
C.ANON.B.CBX / 31 / CBX / Y / 0.575 / 46.067 / 0.210 / -0.032
C.ANON.C / 32 / FNA / Y / 0.941 / 40.811 / 1.744 / 0.811
C.ANON.C.CBX / 32 / CBX / Y / 0.617 / 50.523 / 0.789 / 0.333
C.ANON.D / 33 / FNA / Y / 0.785 / 42.449 / 1.738 / 0.587
C.ANON.D.CBX / 33 / CBX / Y / 0.850 / 46.556 / 1.173 / 0.566
C.ANON.E / 34 / FNA / Y / 0.875 / 42.656 / 1.306 / 0.282
C.ANON.E.CBX / 34 / CBX / Y / 0.714 / 51.007 / 1.288 / 0.497
C.ANON.F / 35 / FNA / Y / 0.775 / 46.919 / 1.625 / 0.309
C.ANON.F.CBX / 35 / CBX / Y / 0.916 / 49.383 / 1.211 / 0.430
C.ANON.G / 36 / FNA / Y / 1.060 / 40.699 / 1.163 / 0.350
C.ANON.G.CBX / 36 / CBX / Y / 0.853 / 48.638 / 0.866 / 0.290
C.ANON.H / 37 / FNA / Y / 0.624 / 50.025 / 1.547 / 0.185
C.ANON.H.CBX / 37 / CBX / Y / 0.649 / 50.693 / 0.861 / 0.090
M146 / 38 / CBX / Y / 2.145 / 51.910 / 0.598 / 0.309
M146CBX / 38 / CBX / Y / 0.571 / 50.936 / 0.386 / -0.044
M569CBX1 / 39 / CBX / Y / 3.049 / 35.175 / 1.542 / 0.480
M569CBX2 / 39 / CBX / Y / 0.687 / 47.660 / 0.219 / 0.116
M594 / 40 / CBX / Y / 1.195 / 46.008 / -0.074 / 0.390
M594CBX / 40 / CBX / Y / 0.806 / 46.986 / 0.354 / 0.304
M601 / 41 / CBX / Y / 0.820 / 43.764 / 0.313 / 0.159
M601CBX / 41 / CBX / Y / 1.095 / 44.586 / 0.307 / 0.170
M639 / 42 / CBX / Y / 0.550 / 48.257 / 0.159 / -0.270
M639CBX / 42 / CBX / Y / 0.836 / 46.399 / 2.483 / 0.241
M677 / 43 / CBX / Y / 0.629 / 49.742 / 0.413 / 0.054
M677CBX / 43 / CBX / Y / 0.739 / 48.804 / 0.774 / -0.346
We made predictions on these technical replicates and biopsy pairs using the 19-genes target index, Src Pathway index as in protocol version 3 and BMS cell line-derived predictor. The concordance correlation coefficient of the prediction scores are shown in table 8 and figures 2-6.
Supplemental Table 7: Concordance correlation coefficientsfor prediction scores
These results indicate high technical reproducibility of the 19-gene target index values and the BMS cell line based predictor values but very poor reproducibility of the SRC Pathway activity score. Matching FNA and CBX samples yielded modes correlation coefficients
SupplementalFigure 3:Replicate FNA(N=39)
SupplementalFigure4: Matching FNA/CBX pairs from same tumor (N=37)
Supplemental Figure 5: FNA/CBX cohort pairs (N=29)
Supplemental Figure6: FNA/CBX ANON pairs (N=8)
SupplementalFigure7: CBX pairs from same tumor (N=6)
The individual predictions from the 3 predictors on each specimen are presented in tables 8-10.
Supplemental Table 8: Prediction results on 39 FNA pairs
In technical replicates, 4 of 39 cases had discrepant prediction result for the 19-gene target index and 1 discordant result for the BMS cell-line based predictor. The Discordance rate was 24/39 for the SRC Pathway score as in version 3 of the protocol.
SupplementalTable 9: Prediction results on 37 FNA/CBX pairs
Supplemental Table 10: Prediction results on 6 CBX pairs
These results indicated very poor reproducibility for the original SRC Pathway score that is based on comparing human gene expression data to cell line centroids.
We next examined the potential reasons for poor reproducibility. We plotted scatter plots and calculated the concordance correlation coefficients for the expression values of the individual probe sets that are included in the three dasatinib predictors to examine reproducibility of probe-level mRNA measurements. The results are shown in figure 7.
SupplementalFigure 8: Concordance for 39 FNA pairs using expression values of individual probes
The concordance measures for the 3 distinct probe set were similar, although the 46 SRC pathway probes were the poorest (0.91). However, the scale for the gene expression values from the three sets of probes was quite different. Probes that represented the SRC pathway tended to have mostly low normalized expression values. cell line predictor have very high intensities. If we plot the reproducibility of the measurements on the same scale by omitting the highest expressed probes from the BMS and 19-gene target index predictors, then the 43 probe sets in the SRC pathway index have lower concordance than the others. This partly explains why the SRC index has the worst reproducibility.
SupplementalFigure 9: Concordance for 39 FNA pairs using the expression values of subset probes.
Next we performed multi-dimension scaling (MDS) analysis with the cell lines used to develop the Src index and the MDACC 133 patients. The pattern of expression of the 46 probes in the human patients is completely different from those in the Src expressing cell lines (Src-pos centroid) or in the GFP expressing cell lines (Src-neg control), Figure 10. The SRC Patway index as described in version 3 of the protocol subtracts two small numbers representing the correlations of a human sample with each of the two cell line centroids, this difference is bound to be in the range of noise because the correlations themselvs are low. Therefore, this centroid-based Src index does not appear to be appropriate for application to real human samples and needs revision
SupplementalFigure 10:Multi-Dimension Scaling (MDS) analysis with the cell lines used to develop the Src index and the MDACC 133 patients
Revised SRC Pathway Activity score that was used in the clinical trial
We revised the Src pathway index by using the average expression of the 46 Src signature genes as the measure of the "Src pathway activity". We calculated this averageSrc pathway score for the 133 breast cancer samples again. The index values ranged from 1372.33 to 4370.46. The distribution is shown in figure 8. The top 25 percentile of the reference population is defined by the threshold of 2124.29. We will consider a patient "positive" for Src pathway activity, andeligible for treatment, if the Src pathway is greater than this value.
Supplemental Figure 11. Revised SRC Pathway activity score distribution in the reference population
Table 11 shows the prediction results from the dasatinib 19-genes target index, BMS cell line based predictor and the revised Src pathway score for each of the 133 patients.
SupplementalTable 11: Prediction results for the 133 breast cancer samples
dasatinib 19-genes target index / BMS cell line based predictor / Src pathway (revised)sample / predict.score / predict.group / predict.score / predict.group / predict.score / predict.group
M106 / 1560.371 / Res / -0.722 / Res / 2113.428 / Res
M107 / 1855.104 / Res / -0.583 / Res / 3140.083 / Sen
M108 / 2264.272 / Res / -0.497 / Res / 2146.947 / Sen
M111 / 3045.952 / Res / -0.064 / Res / 1605.830 / Res
M113 / 2434.783 / Res / -0.303 / Res / 1627.739 / Res
M116 / 2519.998 / Res / -0.566 / Res / 1685.664 / Res
M117 / 2941.787 / Res / -0.249 / Res / 2036.121 / Res
M120 / 2954.483 / Res / 0.222 / Sen / 1610.896 / Res
M121 / 3477.340 / Sen / -0.082 / Res / 1657.914 / Res
M123 / 2385.099 / Res / -0.541 / Res / 2061.919 / Res
M126 / 3247.001 / Sen / -0.096 / Res / 1806.678 / Res
M128 / 1928.046 / Res / -0.418 / Res / 2515.661 / Sen
M129 / 1732.394 / Res / -0.421 / Res / 1885.747 / Res
M130 / 3014.710 / Res / -0.104 / Res / 1709.459 / Res
M133 / 3510.109 / Sen / 0.130 / Sen / 1933.475 / Res
M135 / 2384.435 / Res / -0.353 / Res / 1985.607 / Res
M136 / 2769.932 / Res / 0.021 / Sen / 1598.212 / Res
M139 / 1966.203 / Res / -0.405 / Res / 2241.645 / Sen
M141 / 3472.020 / Sen / -0.270 / Res / 1763.488 / Res
M145 / 1825.865 / Res / -0.326 / Res / 1844.297 / Res
M146 / 4456.698 / Sen / 0.234 / Sen / 1515.258 / Res
M153 / 4021.856 / Sen / -0.367 / Res / 1825.248 / Res
M154 / 3797.585 / Sen / 0.016 / Sen / 1925.882 / Res
M155 / 2453.841 / Res / -0.731 / Res / 2425.683 / Sen
M156 / 2101.418 / Res / -0.353 / Res / 2013.856 / Res
M157 / 1884.402 / Res / -0.560 / Res / 2400.980 / Sen
M158 / 2209.789 / Res / -0.295 / Res / 2084.695 / Res
M159 / 2567.923 / Res / -0.420 / Res / 2162.065 / Sen
M161 / 2778.221 / Res / -0.342 / Res / 2160.393 / Sen
M165 / 2191.613 / Res / -0.622 / Res / 2044.890 / Res
M176 / 1676.093 / Res / -0.600 / Res / 2498.079 / Sen
M177 / 2050.303 / Res / -0.557 / Res / 2262.327 / Sen
M179 / 2446.987 / Res / -0.506 / Res / 1915.762 / Res
M180 / 2075.842 / Res / -0.287 / Res / 2188.257 / Sen
M181 / 1806.730 / Res / -0.588 / Res / 2279.487 / Sen
M182 / 2171.300 / Res / -0.299 / Res / 2099.083 / Res
M186 / 3872.015 / Sen / -0.122 / Res / 1967.370 / Res
M188 / 1771.148 / Res / -0.655 / Res / 1783.658 / Res
M189 / 1753.884 / Res / -0.509 / Res / 2334.356 / Sen
M196 / 1872.536 / Res / -0.616 / Res / 2001.578 / Res
M199 / 3995.931 / Sen / 0.357 / Sen / 1372.334 / Res
M201 / 1856.354 / Res / -0.657 / Res / 2235.814 / Sen
M205 / 3706.108 / Sen / -0.078 / Res / 1649.171 / Res
M206 / 3687.556 / Sen / -0.282 / Res / 1887.031 / Res
M211 / 3028.102 / Res / -0.210 / Res / 1827.522 / Res
M212 / 3273.942 / Sen / -0.292 / Res / 1980.113 / Res
M214 / 1664.057 / Res / -0.736 / Res / 2186.346 / Sen
M215 / 3270.940 / Sen / -0.383 / Res / 2130.515 / Sen
M216 / 2224.211 / Res / -0.803 / Res / 2567.846 / Sen
M217 / 2465.387 / Res / -0.453 / Res / 1948.885 / Res
M220 / 2539.829 / Res / -0.427 / Res / 1915.554 / Res
M226 / 2390.097 / Res / -0.342 / Res / 1745.659 / Res
M227 / 2100.226 / Res / -0.634 / Res / 2132.464 / Sen
M228 / 5139.968 / Sen / -0.068 / Res / 2049.546 / Res
M230 / 3012.136 / Res / -0.225 / Res / 1645.019 / Res
M231 / 3263.776 / Sen / -0.316 / Res / 2067.004 / Res
M233 / 4170.031 / Sen / -0.193 / Res / 1486.792 / Res
M234 / 3733.250 / Sen / 0.289 / Sen / 1469.901 / Res
M235 / 2412.524 / Res / -0.125 / Res / 1599.510 / Res
M236 / 2016.681 / Res / -0.722 / Res / 2007.480 / Res
M237 / 2892.461 / Res / -0.471 / Res / 2109.664 / Res
M238 / 2489.153 / Res / -0.376 / Res / 1636.772 / Res
M239 / 2755.825 / Res / -0.346 / Res / 1966.862 / Res
M245 / 2007.670 / Res / -0.574 / Res / 1550.506 / Res
M246 / 2153.908 / Res / -0.541 / Res / 2020.786 / Res
M247 / 3243.547 / Sen / -0.429 / Res / 1955.476 / Res
M251 / 2319.036 / Res / -0.508 / Res / 1742.739 / Res
M255 / 3979.232 / Sen / -0.104 / Res / 1936.792 / Res
M256 / 1900.368 / Res / -0.280 / Res / 2124.286 / Res
M257 / 2820.029 / Res / -0.575 / Res / 2195.292 / Sen
M258 / 2513.176 / Res / -0.415 / Res / 1666.096 / Res
M259 / 2679.226 / Res / -0.465 / Res / 2476.572 / Sen
M260 / 3070.558 / Res / -0.484 / Res / 1744.648 / Res
M261 / 2072.629 / Res / -0.481 / Res / 2025.583 / Res
M264 / 1603.266 / Res / -0.748 / Res / 2035.962 / Res
M266 / 2169.502 / Res / -0.542 / Res / 1776.648 / Res
M270 / 1464.457 / Res / -0.782 / Res / 1798.108 / Res
M280 / 1906.260 / Res / -0.490 / Res / 2184.024 / Sen
M282 / 3481.706 / Sen / -0.398 / Res / 1916.532 / Res
M286 / 2384.639 / Res / 0.328 / Sen / 1513.734 / Res
M287 / 3037.253 / Res / -0.036 / Res / 1592.575 / Res
M295 / 3261.304 / Sen / -0.296 / Res / 1616.611 / Res
M297 / 1287.601 / Res / -0.515 / Res / 4370.463 / Sen
M301 / 3685.530 / Sen / -0.036 / Res / 1633.941 / Res
M302 / 2075.051 / Res / -0.570 / Res / 2049.573 / Res
M304 / 1721.869 / Res / -0.622 / Res / 2139.964 / Sen
M309 / 3062.388 / Res / -0.329 / Res / 1838.930 / Res
M310 / 3489.505 / Sen / 0.112 / Sen / 1479.226 / Res
M315 / 3385.334 / Sen / -0.208 / Res / 2596.695 / Sen
M316 / 3508.653 / Sen / -0.342 / Res / 1880.156 / Res
M322 / 1267.759 / Res / -0.502 / Res / 2981.190 / Sen
M323 / 2450.432 / Res / -0.326 / Res / 1951.798 / Res
M330 / 2714.144 / Res / -0.274 / Res / 2044.079 / Res
M331 / 4146.637 / Sen / -0.002 / Res / 1529.863 / Res
M333 / 3121.135 / Sen / -0.461 / Res / 1884.424 / Res
M334 / 1487.764 / Res / -0.463 / Res / 1889.103 / Res
M341 / 1363.838 / Res / -0.373 / Res / 2829.927 / Sen
M343 / 1871.725 / Res / -0.376 / Res / 1956.473 / Res
M353 / 1712.483 / Res / -0.427 / Res / 1934.188 / Res
M356 / 4045.248 / Sen / -0.024 / Res / 1837.802 / Res
M357 / 1477.488 / Res / -0.562 / Res / 2107.317 / Res
M367 / 2095.822 / Res / -0.517 / Res / 1851.692 / Res
M371 / 2175.607 / Res / -0.406 / Res / 1750.536 / Res
M373 / 3343.779 / Sen / 0.048 / Sen / 1573.397 / Res
M375 / 2311.354 / Res / -0.339 / Res / 1703.794 / Res
M384 / 4099.146 / Sen / -0.145 / Res / 1780.961 / Res
M386 / 2310.848 / Res / -0.461 / Res / 2119.265 / Res
M387 / 2227.913 / Res / -0.366 / Res / 2069.809 / Res
M399 / 4408.322 / Sen / -0.334 / Res / 1420.276 / Res
M402 / 4460.088 / Sen / 0.043 / Sen / 1773.492 / Res
M421 / 1827.186 / Res / -0.597 / Res / 1971.836 / Res
M434 / 1378.828 / Res / -0.499 / Res / 2546.413 / Sen
M442 / 2105.213 / Res / -0.713 / Res / 2135.699 / Sen
M447 / 1891.881 / Res / -0.454 / Res / 2412.938 / Sen
M463 / 2922.847 / Res / -0.290 / Res / 2340.691 / Sen
M469 / 1732.151 / Res / -0.617 / Res / 1800.812 / Res
M482 / 1483.145 / Res / -0.385 / Res / 2022.909 / Res
M484 / 1647.984 / Res / -0.491 / Res / 1999.860 / Res
M485 / 1805.282 / Res / 0.131 / Sen / 1709.915 / Res
M497 / 2817.414 / Res / -0.366 / Res / 1592.699 / Res
M503 / 2794.905 / Res / 0.065 / Sen / 1542.869 / Res
M506 / 1486.259 / Res / -0.690 / Res / 1791.740 / Res
M523 / 1761.096 / Res / -0.284 / Res / 1790.725 / Res
M524 / 2232.254 / Res / -0.197 / Res / 1545.604 / Res
M525 / 1828.651 / Res / -0.260 / Res / 2471.155 / Sen
M534 / 1867.995 / Res / -0.505 / Res / 1936.773 / Res
ML20 / 3261.458 / Sen / -0.300 / Res / 1835.949 / Res
PERU01 / 1786.072 / Res / -0.421 / Res / 1859.541 / Res
PERU07 / 1464.207 / Res / -0.613 / Res / 1678.822 / Res
PERU09 / 1523.407 / Res / -0.572 / Res / 2196.675 / Sen
PERU11 / 2623.003 / Res / -0.016 / Res / 1390.399 / Res
PERU12 / 1065.173 / Res / -0.663 / Res / 2831.502 / Sen
PERU14 / 2201.940 / Res / -0.435 / Res / 1844.100 / Res
Using the revised Src pathway, there are no cases in the 133 patients predicted to be dasatinib sensitive by all of the three classifiers. 11 cases are double-positive in some combination of the results. This suggest largely independent predictors (Figure 12).