EURURO Vol. 72 No. 4

used. As an unknown proportion may have subsequently received NAC,

validation of the discovered single-sample classifier was performed only

in those 179 without chemotherapy prior to or after sample collection.

For validation of the discovered classifier, two datasets containing 107

(MDA, GSE48075) and 190 (Lund University, GSE32894) patients with

MIBC who did not receive NAC were downloaded from Gene Expression

Omnibus database.

2.4.

Assignment to subtypes

The group from UNC provided the classifiers to assign the samples to

claudin-low, basal, and luminal subtypes

[12,15]

. The one nearest-

neighbor prediction model from the MDA group assigned the tumors to

basal, p53-like, and luminal subtypes

[11]

. The model based on

classification to nearest centroids was used to assign the tumors to

the TCGA clusters

[10] .

The group from Lund provided the centroid

values for each subtype of their original model

[13]

2.5.

Discovery and validation of the single-sample classifier

A detailed description of the model development, model object, and

score generation code are provided in the supplementary material. In

brief, we developed a single-sample model to predict four subtypes

(claudin-low, basal, luminal-infiltrated, and luminal). A generalized

linear model with elastic net regularization (GLMNET) was trained.

2.6.

Statistical analyses

Statistical analyses are described in the supplementary information.

Results

Clinicopathological details of the discovery and validation

NAC metadatasets are provided in

Table 1

and Supplemen-

tary Table 1.

3.1.

Subtyping calls of the published models in the discovery

NAC dataset

The frequency of each subtype in all pre-NAC samples

(

= 223) was in expected ratios

[10,15] ( Fig. 1

A). Gene

expression of given biological functions confirmed the

concordance between the different subtyping methods

( Fig. 1 B

, Supplementary Fig. 1, and Supplementary

Table 2). For example, all luminal subtypes showed

higher expression of genes for urothelial differentiation

( Fig. 1

B, Supplementary Fig. 2 and Supplementary Fig. 3)

compared with all basal subtypes. T-cell and myeloid cell

signatures were highest in the expected subtypes

(claudin-low [UNC] and cluster IV [TCGA]). Genes from

the extracellular matrix were highly expressed in p53-

like (MDA; Supplementary Fig. 3E) and infiltrated

subtypes (Lund;

Fig. 1

B). Taken together, these results

demonstrate high concordance between the subtyping

methods, representing the biological spectrum of MIBC as

reported previously.

Table 1 – Patient characteristic table of the cisplatin-based NAC set (

= 269)

Variable

Bern

NKI

UHS

UW VGH UCD

EMC

Total

value

269

Overall survival time

(censored patients)

Median (IQR) 48 (26–73) 46 (31–87) 49 (40–59) 48 (26–61) 31 (13–41) 9 (4–23) 24 (13–36) 35 (

[8_TD$DIFF]

16–

[9_TD$DIFF]

54)

–

Age

Mean (SD) 64 (10)

57 (11)

67 (12)

62 (11)

61 (9)

65 (8)

59 (13)

61 (11)

–

Gender (%)

Female 12 (30.0)

17 (34.7)

7 (31.8)

10 (24.4)

14 (29.2)

8 (30.8) 12 (27.9)

80 (29.7)

0.97

Male 28 (70.0)

32 (65.3)

15 (68.2)

31 (75.6)

34 (70.8)

18 (69.2) 31 (72.1)

189 (70.3)

Clinical tumor

stage (%)

T1 1 (2.5)

0 (0.0)

1 (0.4)

0.001

T2 4 (10.0)

10 (20.4)

18 (81.8)

20 (48.8)

26 (54.2)

25 (96.2) 17 (39.5)

120 (44.6)

T3 21 (52.5)

26 (53.1)

4 (18.2)

13 (31.7)

19 (39.6)

1 (3.8)

18 (41.9)

102 (37.9)

T4 14 (35.0)

13 (26.5)

0 (0.0)

8 (19.5)

3 (6.2)

0 (0.0)

4 (9.3)

42 (15.6)

NA 0 (0.0)

0 (0.0)

4 (9.3)

4 (1.5)

Clinical lymph node

stage (%)

N0 14 (35.0)

21 (42.9)

20 (90.9)

33 (80.5)

38 (79.2)

0 (0.0)

32 (74.4)

158 (58.7)

0.001

[7_TD$DIFF]

N1–3 26 (65.0)

28 (57.1)

2 (9.1)

8 (19.5)

10 (20.8)

0 (0.0)

7 (16.3)

81 (30.1)

NA 0 (0.0)

0 (0.0)

26 (100.0) 4 (9.3)

30 (11.2)

Pathological tumor

stage (%)

pT0/is/a 13 (32.5)

24 (49.0)

9 (40.9)

18 (43.9)

19 (39.6)

9 (34.6) 13 (30.2)

105 (39.0)

0.618

pT1 3 (7.5)

2 (4.1)

0 (0.0)

5 (12.2)

2 (4.2)

2 (7.7)

6 (14.0)

20 (7.4)

pT2 8 (20.0)

7 (14.3)

5 (22.7)

6 (14.6)

13 (27.1)

3 (11.5)

5 (11.6)

47 (17.5)

pT3/4 16 (40.0)

15 (30.6)

8 (36.4)

12 (29.3)

14 (29.2)

11 (42.3) 18 (41.9)

94 (34.9)

NA 0 (0.0)

1 (2.0)

0 (0.0)

1 (3.8)

1 (2.3)

3 (1.1)

Pathological lymph

node stage (%)

pN0 24 (60.0)

28 (57.1)

18 (81.8)

33 (80.5)

41 (85.4)

16 (61.5) 35 (81.4)

195 (72.5)

0.04

pN1–3 16 (40.0)

12 (24.5)

4 (18.2)

8 (19.5)

6 (12.5)

8 (30.8)

7 (16.3)

61 (22.7)

NA 0 (0.0)

9 (18.4)

0 (0.0)

1 (2.1)

2 (7.7)

1 (2.3)

13 (4.8)

NAC type (%)

GemCis 40 (100.0) 16 (32.7)

22 (100.0) 22 (53.7)

48 (100.0) 19 (73.1) 33 (76.7)

200 (74.3)

0.001

MVAC 0 (0.0)

33 (67.3)

0 (0.0)

19 (46.3)

0 (0.0)

6 (23.1)

5 (11.6)

63 (23.4)

Other 0 (0.0)

0 (0.0)

1 (3.8)

5 (11.6)

6 (2.2)

GSC = genomic subtyping classifier; IQR = interquartile range; MVAC = methotrexate, vinblastine, adriamycin, and cisplatin; NA = not applicable;

NAC = neoadjuvant chemotherapy; SD = standard deviation.

UCD and EMC sets represent blinded and external validation sets not used in discovery of the GSC.

Excluding UCD due to lack of clinical lymph node status.

E U R O P E A N U R O L O G Y 7 2 ( 2 0 1 7 ) 5 4 4 – 5 5 4

546