Input size mismatch with dlnetwork

I'm trying to do some GAN tranfer learning and am getting some errors when I try to add layers before existing layers when I use a dlnetwork. Here's a minimal example:

%% make network 1
layers1=[imageInputLayer([8 1],'Normalization','none','Name','in1')
    fullyConnectedLayer(20,'Name','FC1')
    tanhLayer('Name','T1')
    fullyConnectedLayer(1,'Name','Lout')
    ];
lgraph1 = layerGraph(layers1);
dlnet1 = dlnetwork(lgraph1);

%% optionally do some training on dlnet1
%% add layers in front of network 1
layersNew = [imageInputLayer([16 1],'Normalization','none','Name','in2');
    fullyConnectedLayer( 8,'Name','FC2')
    tanhLayer('Name','T2')
    ];

lgraph2 = layerGraph(dlnet1);
%lgraph2= lgraph1
lgraph2 = addLayers(lgraph2,layersNew);
lgraph2 = disconnectLayers(lgraph2,'in1','FC1');
lgraph2 = connectLayers(lgraph2,'T2','FC1');
lgraph2 = removeLayers(lgraph2,'in1');
dlnet2 = dlnetwork(lgraph2);

This results in the error

Error using dlnetwork (line 81)
Invalid network.

Error in minimum_error (line 24)
dlnet2 = dlnetwork(lgraph2);

Caused by:
    Layer 'FC1': Input size mismatch. Size of input to this layer is
    different from the expected input size.
    Inputs to this layer:
        from layer 'T2' (1×1×8 output)

Presumably this is because of the 1-D "image" I'm forced to use rather than a generic 1D vector input. T2 is giving a 1x1x8 output whereas FC is expecting 8x1 (or maybe 1x8) input. The weird thing is that if I uncomment the "lgraph2= lgraph1" line, there is no error. So there must be a matrix permutation happening in the layerGraph->dlnetwork->layerGraph conversion.

I can't find any way to permute or reshape the output of a layer to resolve this. Any workaround?